cv2.line(vis, (x1-r, y1+r), (x1+r, y1-r), col, thickness)
cv2.line(vis, (x2-r, y2-r), (x2+r, y2+r), col, thickness)
cv2.line(vis, (x2-r, y2+r), (x2+r, y2-r), col, thickness)
vis0 = vis.copy()
for (x1, y1), (x2, y2), inlier in zip(p1, p2, status):
if inlier:
cv2.line(vis, (x1, y1), (x2, y2), green)
cv2.imshow(win, vis)
cv2.waitKey(0)
cv2.destroyAllWindows()
sift = cv2.xfeatures2d.SIFT_create()
#sift = cv2.SURF() SURF/ORB由于图片太小，特征点太少
#sift = cv2.ORB()
bf = cv2.BFMatcher(cv2.NORM_L2)#It is good for SIFT, SURF etc .http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_feature2d/py_matcher/py_matcher.html?highlight=bfmatcher
#bf = cv2.BFMatcher(cv2.NORM_HAMMING2)
def compare_SIFT(im1,im2):
kp1,des1 = sift.detectAndCompute(im1,None)
kp2,des2 = sift.detectAndCompute(im2,None)
#print 'kp2',kp2
if kp2 == []:
dprint('kp2 does not have the keypoint')
return
matches = bf.knnMatch(des1, des2, k=2)#KNNMatch，可设置K = 2 ，即对每个匹配返回两个最近邻描述符，仅当第一个匹配与第二个匹配之间的距离足够小时，才认为这是一个匹配。
#print matches
dprint ('kp1:',len(kp1))
dprint( 'kp1:',len(kp2))
p1,p2,kp_pairs = filter_matches(kp1,kp2,matches)
dprint( "matches",len(kp_pairs))
ratio = 0
import numpy as np
img_ = cv2.imread('img1_1.jpg')
img1 = cv2.cvtColor(img_, cv2.COLOR_BGR2GRAY)
img = cv2.imread('img2_2.jpg')
img2 = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
sift = cv2.xfeatures2d.SIFT_create()
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)
match = cv2.BFMatcher()
matches = match.knnMatch(des1, des2, k=2)
good = []
for m, n in matches:
if m.distance < 0.03 * n.distance:
good.append(m)
draw_params = dict(matchColor=(0, 255, 0), singlePointColor=None, flags=2)
img3 = cv2.drawMatches(img_, kp1, img, kp2, good, None, **draw_params)
MIN_MATCH_COUNT = 10
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([kp1[m.queryIdx].pt for m in good]).reshape(-1, 1, 2)
dst_pts = np.float32([kp2[m.trainIdx].pt for m in good]).reshape(-1, 1, 2)
img1 = cv2.imread('../images/mountain1.png')
img2 = cv2.imread('../images/mountain2.png')
sift = cv2.xfeatures2d.SIFT_create()
#extract sift keypoints and descriptors
(kp1, des1) = sift.detectAndCompute(img1, None)
(kp2, des2) = sift.detectAndCompute(img2, None)
# show keypoints
out1 = cv2.drawKeypoints(img1, kp1, None)
out2 = cv2.drawKeypoints(img2, kp2, None)
cv2.imshow("Keypoints for Image-1 ( press enter to continue )", out1)
cv2.imshow("Keypoints for Image-2 ( press enter to continue )", out2)
#print(des1,des2)
cv2.waitKey(0)
cv2.destroyAllWindows()
bf = cv2.BFMatcher()
matches12 = bf.knnMatch(des1,des2, k=2)
matches21 = bf.knnMatch(des2,des1, k=2)
#print(matches12.shape,matches21.shape)
# own implementation of matching
# filter matches by ratio test
t = 0.4
validratio12 = {}
for p in matches12:
if (p[0].distance/p[1].distance) <= t:
validratio12[p[0].queryIdx] = (p[0].trainIdx,p)
validratio21 = {}
for p in matches21:
if (p[0].distance/p[1].distance) <= t:
validratio21[p[0].queryIdx] = (p[0].trainIdx,p)
def __init__(self):
self.orb = cv2.ORB_create(100)
self.bf = cv2.BFMatcher(cv2.NORM_HAMMING)
self.last = None
p2 = (int(points2[i][0] + W1 + 30), int(points2[i][1]))
cv2.line(visImg, p1, p2, color)
return visImg
img1 = cv2.imread("../imgs/book1.jpg", 0)
img2 = cv2.imread("../imgs/book3.jpg", 0)
#compute SIFT
sift = cv2.xfeatures2d.SIFT_create()
key1, des1 = sift.detectAndCompute(img1, None)
key2, des2 = sift.detectAndCompute(img2, None)
#matching and triming
matches = cv2.BFMatcher(cv2.NORM_L2).match(des1, des2)
threshold = np.array([m.distance for m in matches]).mean() * 1.2
matches_trim = [m for m in matches if m.distance < threshold]
point1 = np.array([np.array(key1[m.queryIdx].pt) for m in matches_trim])
point2 = np.array([np.array(key2[m.trainIdx].pt) for m in matches_trim])
imgMatch = visMatching(img1, img2, point1, point2)
imgVis1 = visKeyPoints(img1, key1)
imgVis2 = visKeyPoints(img2, key2)
# wait key input
cv2.imshow("key points1", imgVis1)
cv2.imshow("key points2", imgVis2)
cv2.imshow("key match", imgMatch)
cv2.waitKey(0)
def __init__(self):
self.next_state = 'None'
self.number_of_egg = 11
self.hatched_egg = 0
self._detect_frame_count = 0
self._good = []
self._good_without_list = []
self._bf = cv2.BFMatcher()
self._breeder_house_flg = 0
self.send_command = 'None'
self._send_command_enb = 0
self._control_frame_count = 0
self._temp_y = 408
self.run_control_state = 'None'
self.a_breeder_count = 0
self.run_l_count = 0
self._sel_poke_flg = 0
self._breeder_comment = cv2.imread("data\\breeder_comment.png")
self._sora_sel_img = cv2.imread("data\\sora_sel.png")
self.cut_frame_h = 237
self.cut_frame_w = 1280
self.cross_arm_breeder1 = cv2.imread("data\\a_breeder1.png")
self.cross_arm_breeder2 = cv2.imread("data\\a_breeder2.png")
self.cross_arm_breeder3 = cv2.imread("data\\a_breeder3.png")
self.breeder1 = cv2.imread("data\\breeder1.png")
self.breeder2 = cv2.imread("data\\breeder2.png")
self.breeder3 = cv2.imread("data\\breeder3.png")
self.f_a_breeder = cv2.imread("data\\f_a_breeder_hand.png")
self.f_breeder3 = cv2.imread("data\\f_breeder_hand.png")
self.menu_sel_poke = cv2.imread("data\\menu_sel_poke.png")
self.gray_cross_arm_breeder1 = cv2.cvtColor(self.cross_arm_breeder1,
cv2.COLOR_RGB2GRAY)
self.gray_cross_arm_breeder2 = cv2.cvtColor(self.cross_arm_breeder2,
cv2.COLOR_RGB2GRAY)
self.gray_cross_arm_breeder3 = cv2.cvtColor(self.cross_arm_breeder3,
cv2.COLOR_RGB2GRAY)
self.gray_breeder1 = cv2.cvtColor(self.breeder1, cv2.COLOR_RGB2GRAY)
self.gray_breeder2 = cv2.cvtColor(self.breeder2, cv2.COLOR_RGB2GRAY)
self.gray_breeder3 = cv2.cvtColor(self.breeder3, cv2.COLOR_RGB2GRAY)
self.gray_f_a_breeder = cv2.cvtColor(self.f_a_breeder,
cv2.COLOR_RGB2GRAY)
self.gray_f_breeder = cv2.cvtColor(self.f_breeder3, cv2.COLOR_RGB2GRAY)
self.ah1, self.aw1 = self.gray_cross_arm_breeder1.shape
self.ah2, self.aw2 = self.gray_cross_arm_breeder2.shape
self.ah3, self.aw3 = self.gray_cross_arm_breeder3.shape
self.h1, self.w1 = self.gray_breeder1.shape
self.h2, self.w2 = self.gray_breeder2.shape
self.h3, self.w3 = self.gray_breeder3.shape
self.hf, self.wf = self.gray_f_breeder.shape
#self.gray_cross_arm_breeder1 = np.array(self.gray_cross_arm_breeder1, dtype="float")
#self.gray_cross_arm_breeder2 = np.array(self.gray_cross_arm_breeder2, dtype="float")
#self.gray_cross_arm_breeder3 = np.array(self.gray_cross_arm_breeder3, dtype="float")
self.mu_t1 = np.mean(self.gray_cross_arm_breeder1)
self.mu_t2 = np.mean(self.gray_cross_arm_breeder2)
self.mu_t3 = np.mean(self.gray_cross_arm_breeder3)
self.temp1 = self.gray_cross_arm_breeder1 - self.mu_t1
self.temp2 = self.gray_cross_arm_breeder2 - self.mu_t2
self.temp3 = self.gray_cross_arm_breeder3 - self.mu_t3
#self.dst = 0
if VolatileClassRun.ALGOLISM == 'AKAZE':
VolatileClassRun.MIN_MATCH_COUNT = 20 #あんまりすくないとfindHomographyがNoneになる
VolatileClassRun.ratio = 0.7
self._akaze = cv2.AKAZE_create()
self._kp1, self._des1 = self._akaze.detectAndCompute(
VolatileClassRun.breeder_house_img, None)
elif VolatileClassRun.ALGOLISM == 'SURF':
VolatileClassRun.MIN_MATCH_COUNT = 30 #あんまりすくないとfindHomographyがNoneになる
VolatileClassRun.ratio = 0.5
self._surf = cv2.xfeatures2d.SURF_create(400)
self._kp1, self._des1 = self._surf.detectAndCompute(
VolatileClassRun.breeder_house_img, None)
elif VolatileClassRun.ALGOLISM == 'USURF':
VolatileClassRun.MIN_MATCH_COUNT = 18 #あんまりすくないとfindHomographyがNoneになる
VolatileClassRun.ratio = 0.6
self._surf = cv2.xfeatures2d.SURF_create(400)
self._surf.setUpright(True)
self._kp1, self._des1 = self._surf.detectAndCompute(
VolatileClassRun.breeder_house_img, None)
elif VolatileClassRun.ALGOLISM == 'FAST':
VolatileClassRun.MIN_MATCH_COUNT = 30 #あんまりすくないとfindHomographyがNoneになる
VolatileClassRun.ratio = 0.5
self._fast = cv2.FastFeatureDetector_create()
self._brief = cv2.xfeatures2d.BriefDescriptorExtractor_create()
self._kp1 = self._fast.detect(VolatileClassRun.breeder_house_img,
None)
self._kp1, self._des1 = self._brief.compute(
VolatileClassRun.breeder_house_img, self._kp1)
self.matches = self._bf.knnMatch(self._des1, self._des1, k=2)
def pose_estimation():
global f, B
global frame_L, frame_R, frame_L_prev, frame_R_prev
global left_image, right_image, img_L, img_R
global left_featured_image, right_featured_image, matched_featured_image
global img_L, frame_L, frame_L_prev, dt_L
global flow_left_matched_featured_image, flow_left_image
global height, width, scale
frame_L_prev = frame_L
frame_R_prev = frame_R
try:
img_L = bridge.imgmsg_to_cv2(left_image, "bgr8")
except CvBridgeError as e:
print(e)
try:
img_R = bridge.imgmsg_to_cv2(right_image, "bgr8")
except CvBridgeError as e:
print(e)
img_height = len(img_L)
img_width = len(img_L[0])
# print(img_height)
# print(img_width)
#Resize image
scale = 1
width = int(img_L.shape[1] * scale)
height = int(img_L.shape[0] * scale)
dim = (width, height) #can also just specify desired dimensions
frame_L = cv2.resize(img_L, dim, interpolation=cv2.INTER_AREA)
frame_R = cv2.resize(img_R, dim, interpolation=cv2.INTER_AREA)
#Convert from BGR to gray colorspace
frame_L = cv2.cvtColor(frame_L, cv2.COLOR_BGR2GRAY)
frame_R = cv2.cvtColor(frame_R, cv2.COLOR_BGR2GRAY)
img1 = frame_L.copy()
img2 = frame_R.copy()
# Initiate ORB detector
orb = cv2.ORB_create()
# find the keypoints and descriptors with ORB
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)
# create BFMatcher object
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# Match descriptors.
matches = bf.match(des1, des2)
# Sort them in the order of their distance.
matches = sorted(matches, key=lambda x: x.distance)
top_matches = matches[:20]
plot_image = cv2.drawMatches(img1,
kp1,
img2,
kp2,
top_matches,
None,
flags=2)
# Initialize lists
list_kp1 = []
list_kp2 = []
list_disp = []
list_z = []
# For each match...
for mat in top_matches:
# Get the matching keypoints for each of the images
img1_idx = mat.queryIdx
img2_idx = mat.trainIdx
# x - columns
# y - rows
# Get the coordinates
(x1, y1) = kp1[img1_idx].pt
(x2, y2) = kp2[img2_idx].pt
if (x2 - x1 > 0.001):
z = (f * B) / (x2 - x1)
list_z.append(z)
# Append to each list
list_kp1.append((x1, y1))
list_kp2.append((x2, y2))
list_disp.append(x2 - x1)
# mean_z = 0.0
# if len(list_z)>0:
# mean_z = sum(list_z)/len(list_z)
# print(mean_z)
matched_featured_image = bridge.cv2_to_imgmsg(plot_image, "8UC3")
img3 = frame_L_prev.copy()
# find the keypoints and descriptors with ORB
kp3, des3 = orb.detectAndCompute(img3, None)
if des3 is not None:
# Match descriptors.
matches = bf.match(des1, des3)
# Sort them in the order of their distance.
matches = sorted(matches, key=lambda x: x.distance)
if (len(matches) > 100):
top_matches = matches[:50]
else:
top_matches = matches[:20]
# draw_params = dict(matchColor = (0,255,0),
# singlePointColor = (255,0,0),
# matchesMask = matchesMask,
# flags = 0)
# img3 = cv2.drawMatchesKnn(img1,kp1,img2,kp2,matches,None,**draw_params)
plot_image = cv2.drawMatches(img1,
kp1,
img3,
kp3,
top_matches,
None,
flags=2)
flow_left_matched_featured_image = bridge.cv2_to_imgmsg(
plot_image, "8UC3")
plot_image = img_L.copy()
# Initialize lists
list_kp1 = []
list_kp2 = []
list_vx_img = []
list_vy_img = []
# For each match...
for mat in top_matches:
# Get the matching keypoints for each of the images
img1_idx = mat.queryIdx
img2_idx = mat.trainIdx
# x - columns
# y - rows
# Get the coordinates
(x1, y1) = kp1[img1_idx].pt
(x2, y2) = kp3[img2_idx].pt
vx_img = (x2 - x1) / dt_L
vy_img = (y2 - y1) / dt_L
# Append to each list
list_kp1.append((x1, y1))
list_kp2.append((x2, y2))
list_vx_img.append(vx_img)
list_vy_img.append(vy_img)
cv2.arrowedLine(plot_image, (int(x1), int(y1)), (int(x2), int(y2)),
(0, 0, 255),
thickness=2,
line_type=8,
shift=0,
tipLength=0.5)
# print(dt_L)
# print(sum(list_vx_img)/len(list_vx_img))
flow_left_image = bridge.cv2_to_imgmsg(plot_image, "8UC3")
def matched_positions(img1, img2, k=3, n=2, unique=False, order=0):
'''
tips: works better for scale within 0.5-2
Args:
img1: template image
img2: target image
k: k matches for each feature point
n: filter for filtering out points with distance > n* min_distance
unique: whether the object is unique
return:
positions: first n best matches positions
side effects:
create a result_with_bb.png showing the matched result with bounding box
'''
# Initiate SIFT detector
if unique:
k = 1
sift = cv.xfeatures2d.SIFT_create()
# find the keypoints and descriptors with SIFT
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)
# BFMatcher with default params
print("# of keypoints: ", len(kp1))
bf = cv.BFMatcher()
matches = bf.knnMatch(des1, des2, k)
matches_flat = []
# flat the match list
for match in matches:
for match_point in match:
matches_flat.append(match_point)
matches_flat = sorted(matches_flat, key=lambda x: x.distance)
max_distance = matches_flat[len(matches_flat) - 1].distance
min_distance = matches_flat[0].distance
good = []
for match in matches_flat:
if match.distance <= max(0.02, n * min_distance):
good.append(match)
#extract position in the target image
if len(good) < 5:
good = matches_flat[:5]
if not unique:
positions_map = map(
lambda x: (x.queryIdx, x.trainIdx, kp2[x.trainIdx].pt), good)
positions = map(lambda x: x[2], positions_map)
bandwidth = max(50,
estimate_bandwidth(positions, quantile=0.3, n_jobs=-1))
if (len(good) <= 20):
bandwidth = min(50, bandwidth)
img1_width_up = 2 * max(img1.shape)
print("upper_width", img1_width_up)
bandwidth = min(img1_width_up, bandwidth)
img1_width_low = max(img1.shape)
bandwidth = max(img1_width_low, bandwidth)
print("lower_width", img1_width_low)
print('bandwidth: ', bandwidth)
ms = MeanShift(bandwidth=bandwidth, bin_seeding=True)
ms.fit(positions)
labels = ms.labels_
labels_unique = np.unique(labels)
print("# of clusters: ", len(labels_unique))
print("labels: ", labels)
n_clusters = len(labels_unique)
matched_group = []
for i in range(0, n_clusters):
matched_group.append({'query': [], 'train': []})
for i in range(0, len(positions_map)):
for j in range(0, n_clusters):
if labels[i] == j:
matched_group[j]['query'].append(
kp1[positions_map[i][0]].pt)
matched_group[j]['train'].append(
kp2[positions_map[i][1]].pt)
h, w = img1.shape
corners = [[0, 0], [0, h - 1], [w - 1, h - 1], [w - 1, 0]]
img3 = cv.drawMatchesKnn(img1, kp1, img2, kp2, [good], None, flags=2)
for i in range(0, n_clusters):
if len(matched_group[i]['query']) >= 5:
try:
box = produce_bounding_box(corners,
matched_group[i]['query'],
matched_group[i]['train'])
# translate the box to the correct positions
box = map(lambda x: [[x[0][0] + w, x[0][1]]], box)
img3 = cv.polylines(img3, [np.int32(box)], True, 255, 3,
cv.LINE_AA)
except:
print("Not enough information to produce the bounding box")
print("result_with_bb{}.png".format(order))
cv.imwrite("result_with_bb{}.png".format(order), img3)
return positions
else:
# make every matched point has a unique query point,use the matched point with best distance
output = []
present_train_idx = []
present_query_idx = []
for element in good:
if (element.trainIdx
not in present_train_idx) and (element.queryIdx
not in present_query_idx):
present_train_idx.append(element.trainIdx)
present_query_idx.append(element.queryIdx)
output.append(element)
good = output
positions_map = map(
lambda x: (x.queryIdx, x.trainIdx, kp2[x.trainIdx].pt), good)
positions = map(lambda x: x[2], positions_map)
query_positions = map(lambda x: kp1[x[0]].pt, positions_map)
img3 = cv.drawMatchesKnn(img1, kp1, img2, kp2, [good], None, flags=2)
h, w = img1.shape
print(h, w)
corners = [[0, 0], [0, h - 1], [w - 1, h - 1], [w - 1, 0]]
box = produce_bounding_box(corners, query_positions, positions)
box = map(lambda x: [[x[0][0] + w, x[0][1]]], box)
img3 = cv.polylines(img3, [np.int32(box)], True, 255, 3, cv.LINE_AA)
print("result_with_bb{}.png".format(order))
cv.imwrite("result_with_bb{}.png".format(order), img3)
return positions
def sift_det(image,source):
# TODO:
# Sets up match count between images for sensitivity of detection - choose your value!
MIN_MATCH_COUNT = 10
# If VideoCapture(feed) doesn't work, manually try -1, 0, 1, 2, 3 (if none of those work,
# the webcam's not supported!)
#cam = cv2.VideoCapture(args["source"])
# Reads in the image
img1 = cv2.imread(image, 0)
""" # Labels the image as the name passed in
if args["label"] is not None:
label = args["label"]
else:
# Takes the name of the image as the name
if image[:2] == "./":
label = label = (image.split("/"))[2]
else:
label = image[2:-4] """
################################################### Set up Feature Detection
# Create a the SIFT Detector Object
try:
#delete! sift = cv2.xfeatures2d.SIFT_create()
orb = cv2.ORB_create()
except AttributeError:
print("Install 'opencv-contrib-python' for access to the xfeatures2d module")
# Compute keypoints
kp, des = orb.detectAndCompute(img1, None)
FLANN_INDEX_KDTREE = 0
# Option of changing 'trees' and 'checks' values for different levels of accuracy
index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
search_params = dict(checks = 50)
# Fast Library for Approximate Nearest Neighbor Algorithm
# Creates FLANN object for use below
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck = True)
#delete! flann = cv2.FlannBasedMatcher(index_params, search_params)
#ret_val, frame = cam.read()
frame = source
""" if frame is None: # Did we get an image at all?
continue """
################################################### Shape Computation
# TODO:
# What color space does OpenCV read images in, and what color space do we want process?
# Check out cvtColor! <https://docs.opencv.org/2.4/modules/imgproc/doc/miscellaneous_transformations.html>
# Read in the image from the camera
img = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
# TODO:
# Set up your HSV threshold bounds - [Hue, Saturation, Value]
lower = np.array([0, 0, 100], dtype = "uint8")
upper = np.array([100, 10, 90], dtype = "uint8")
# TODO:
# Check inRange() <https://docs.opencv.org/3.0-beta/modules/core/doc/operations_on_arrays.html?highlight=inrange#invert>
# Create mask for image with overlapping values
mask = cv2.inRange(img, lower, upper)
# TODO:
# What parameters work best for thresholding? <https://docs.opencv.org/2.4/modules/imgproc/doc/miscellaneous_transformations.html?highlight=adaptivethreshold>
imgThresh = cv2.adaptiveThreshold(mask, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 3, 1)
# TODO:
# This is OpenCV's call to find all of the contours
# Experiment with different algorithms (cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
# in the parameters of cv2.findContours!
# <https://docs.opencv.org/2.4/modules/imgproc/doc/structural_analysis_and_shape_descriptors.html?highlight=findContours>
# The underscores represent Python's method of unpacking return values, but not using them
_, contours, _ = cv2.findContours(imgThresh, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
# Optional TODO:
# Optional processing of contours - do we want to remove all non-rectangle shapes from contours?
# Read the documentation on approxPolyDP <https://docs.opencv.org/2.4/modules/imgproc/doc/structural_analysis_and_shape_descriptors.html>
# TODO:
# Orders contours - but by what metric? Check the "key" options <https://docs.opencv.org/3.1.0/dd/d49/tutorial_py_contour_features.html>
# Ex. key = cv2.contourVectorLen() (Would order the contours by vector length - not an actual function, but this is how you would set the "key")
# Python's "sorted" function applies a "key" set lambda function to each element within an array, this is not a traditional dictionary "key"
contours = sorted(contours, key = cv2.contourArea, reverse = True)[1:] # Removes contouring of display window
if len(contours) != 0:
# TODO:
# Draws the max of the contours arrays with respect to the "key" chosen above
contours_max = max(contours, key = cv2.contourArea)
# Find bounding box coordinates
rect = cv2.boundingRect(contours_max)
x, y, w, h = rect
# TODO:
# Calculates area of detection - what detection area should be the lower bound?
if w*h > 28000:
#cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 4)
greenRect = ((x, y), (x+w, y+h))
################################################### Feature Detection
# NOTE: NO CODE TO FILL IN HERE - BUT FEEL FREE TO CHANGE THE PARAMETERS
kp_s, des_s = orb.detectAndCompute(frame, None)
if(len(kp) >= 2 and len(kp_s) >= 2):
# Uses the FLANN algorithm to search for nearest neighbors between elements of two images
# Faster than the BFMatcher for larger datasets
matches = bf.knnMatch(des,des_s,k=2)
#delete! matches = flann.knnMatch(des, des_s, k = 2)
#except cv2.error:
#pass
if des_s is None and len(matches) == 0:
pass
# Store all the good matches (based off Lowe's ratio test)
good = []
for k, pair in enumerate(matches):
try:
(m, n) = pair
if m.distance < 0.75 * n.distance:
good.append(m)
except ValueError:
pass
# When there are enough matches, we convert the keypoints to floats in order to draw them later
if len(good) >= MIN_MATCH_COUNT:
try:
src_pts = np.float32([ kp[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
dst_pts = np.float32([ kp_s[m.trainIdx].pt for m in good ]).reshape(-1,1,2)
except IndexError:
pass
# Homography adds a degree of rotation/translation invariance by mapping the transformation
# of points between two images
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
matchesMask = mask.ravel().tolist()
h, w = img1.shape
pts = np.float32([ [0,0],[0,h-2],[w-2,h-2],[w-2,0] ]).reshape(-1,1,2)
if M is not None:
dst = cv2.perspectiveTransform(pts, M)
intDst = np.int32(dst)
# Draws a bounding box around the area of most matched points
#cv2.rectangle(frame, (intDst[0][0][0], intDst[0][0][1]), (intDst[2][0][0], intDst[2][0][1]), (0, 0, 255), 4, cv2.LINE_AA, 0)
redRect = ((intDst[0][0][0], intDst[0][0][1]), (intDst[2][0][0], intDst[2][0][1]))
cv2.putText(frame, label, (dst[0][0][0], dst[0][0][1]) , cv2.FONT_HERSHEY_TRIPLEX, 1.0, (0, 0, 255), lineType = cv2.LINE_AA )
else:
matchesMask = None
else:
matchesMask = None
draw_params = dict(matchColor = (0, 255, 0), # Draw matches in green color
singlePointColor = None,
matchesMask = matchesMask, # Draw only inliers
flags = 2)
try:
# Option of slicing the 'good' list to display a certain number of matches (ex. good[:6])
# Take out draw_params if we do not want to draw matches
frame = cv2.drawMatches(img1, kp, frame, kp_s, good, None, **draw_params)
except cv2.error:
pass
# Shows the current frame
return(redRect, greenRect)
def superm2(image):
mimage = np.fliplr(image)
kp1, des1 = sift.detectAndCompute(image, None)
kp2, des2 = sift.detectAndCompute(mimage, None)
for p, mp in zip(kp1, kp2):
p.angle = np.deg2rad(p.angle)
mp.angle = np.deg2rad(mp.angle)
bf = cv2.BFMatcher()
matches = bf.knnMatch(des1, des2, k=2)
houghr = np.zeros(len(matches))
houghth = np.zeros(len(matches))
weights = np.zeros(len(matches))
i = 0
good = []
for match, match2 in matches:
point = kp1[match.queryIdx]
mirpoint = kp2[match.trainIdx]
mirpoint2 = kp2[match2.trainIdx]
mirpoint2.angle = np.pi - mirpoint2.angle
mirpoint.angle = np.pi - mirpoint.angle
if mirpoint.angle < 0.0:
mirpoint.angle += 2 * np.pi
if mirpoint2.angle < 0.0:
mirpoint2.angle += 2 * np.pi
mirpoint.pt = (mimage.shape[1] - mirpoint.pt[0], mirpoint.pt[1])
if very_close(point.pt, mirpoint.pt):
mirpoint = mirpoint2
good.append(match2)
else:
good.append(match)
theta = angle_with_x_axis(point.pt, mirpoint.pt)
xc, yc = midpoint(point.pt, mirpoint.pt)
r = xc * np.cos(theta) + yc * np.sin(theta)
Mij = reisfeld(point.angle, mirpoint.angle, theta) * S(
point.size, mirpoint.size
)
houghr[i] = r
houghth[i] = theta
weights[i] = Mij
i += 1
# matches = sorted(matches, key = lambda x:x.distance)
good = sorted(good, key=lambda x: x.distance)
def draw(r, theta):
if np.pi / 4 < theta < 3 * (np.pi / 4):
for x in range(len(image.T)):
y = int((r - x * np.cos(theta)) / np.sin(theta))
if 0 <= y < len(image.T[x]):
image[y][x] = 255
else:
for y in range(len(image)):
x = int((r - y * np.sin(theta)) / np.cos(theta))
if 0 <= x < len(image[y]):
image[y][x] = 255
print(houghr)
print(houghth)
img3 = cv2.drawMatches(image, kp1, mimage, kp2, good[:15], None, flags=2)
def hex():
polys = plt.hexbin(houghr, houghth, bins=200, gridsize=image.shape[1])
# plt.colorbar()
# plt.show()
hvals = polys.get_array()
hcoords = polys.get_offsets()
return hcoords[hvals.argmax()]
best_coords = hex()
print(best_coords)
# draw(2.8, 2.4)
draw(*best_coords)
cv2.imshow('a', image); cv2.waitKey(0);
def __init__(self, detector):
self._detector = detector
self._matcher = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
self._prevImg = None
def wait_function(single_image_target, new_images_path, images_subject):
try:
percentage = 0.50
geo_portail = []
sift = cv2.xfeatures2d.SIFT_create()
images_target_chunk = [single_image_target]
for id_img, base_image in enumerate(images_subject):
# print(f"image subject {id_img} - {images_subject[id_img]}")
# cv2.imread(base_image)
base_image_original = os.path.basename(base_image)
# print(f"{str(threading.current_thread().name)}::IMAGE SUBJECT:: {base_image_original}")
# print(f" ")
base_image = cv2.imread(base_image, 0)
# print(f"__Supouse to check {single_image_target} ")
for idxInt, fileD in enumerate([single_image_target]):
# print(f"____Checking... Image Target {fileD} ")
# print(f"Image to processe {single_image_target} ")
file_path = str(fileD)
if os.path.exists(file_path):
# print(f"____file exist {True} ")
file_name = str(os.path.basename(fileD))
target_image_color = cv2.imread(file_path)
target_image = cv2.imread(file_path, 0)
kp1, des1 = sift.detectAndCompute(base_image, None)
kp2, des2 = sift.detectAndCompute(target_image, None)
bf = cv2.BFMatcher()
matches = bf.knnMatch(des1, des2, k=2)
good = []
percentage = 0.65 if base_image_original == 'logo.png' else percentage
percentage = 0.65 if base_image_original == 'logo_cadastre.png' else percentage
cont = 0
for match1, match2 in matches:
if match1.distance < percentage * match2.distance:
good.append([match1])
if cont == 0:
# print(f"____found match {True} " )
cont = 1
if base_image_original == "doble.png":
if file_path not in geo_portail and len(
matches) >= 1400 and len(good) >= 70:
# geo_portail.append(file_path)
cv2.imwrite(
os.path.join(new_images_path,
file_name),
target_image_color)
matched = True
if idx > 0 and len(
images_target_chunk) > 0:
print(
f"Saved: {base_image_original} => {file_name} current size chunk {len(images_target_chunk)} - {id_img}"
)
del images_target_chunk[idxInt]
print(
f"{base_image_original}: after delete: {len(images_target_chunk)} - {id_img}"
)
break
if base_image_original == "green.png":
if file_path not in geo_portail and len(
matches) >= 2500 and len(good) >= 120:
# geo_portail.append(file_path)
cv2.imwrite(
os.path.join(new_images_path,
file_name),
target_image_color)
matched = True
if idx > 0 and len(
images_target_chunk) > 0:
print(
f"Saved: {base_image_original} => {file_name} current size chunk {len(images_target_chunk)} - {id_img}"
)
del images_target_chunk[idxInt]
print(
f"{base_image_original}: after delete: {len(images_target_chunk)} - {id_img}"
)
break
if base_image_original == "icons.png":
if file_path not in geo_portail and len(
matches) >= 120 and len(good) >= 45:
# geo_portail.append(file_path)
cv2.imwrite(
os.path.join(new_images_path,
file_name),
target_image_color)
matched = True
if idx > 0 and len(
images_target_chunk) > 0:
print(
f"Saved: {base_image_original} => {file_name} current size chunk {len(images_target_chunk)} - {id_img}"
)
del images_target_chunk[idxInt]
print(
f"{base_image_original}: after delete: {len(images_target_chunk)} - {id_img}"
)
break
if base_image_original == "logo_cadastre.png":
if file_path not in geo_portail and len(
matches) >= 150 and len(good) >= 27:
# geo_portail.append(file_path)
cv2.imwrite(
os.path.join(new_images_path,
file_name),
target_image_color)
matched = True
if idx > 0 and len(
images_target_chunk) > 0:
print(
f"Saved: {base_image_original} => {file_name} current size chunk {len(images_target_chunk)} - {id_img}"
)
del images_target_chunk[idxInt]
print(
f"{base_image_original}: after delete: {len(images_target_chunk)} - {id_img}"
)
break
if base_image_original == "logo.png":
if file_path not in geo_portail and len(
matches) > 200 and len(good) >= 75:
# geo_portail.append(file_path)
# print(f"Save here: {os.path.join(new_images_path, file_name)}")
cv2.imwrite(
os.path.join(new_images_path,
file_name),
target_image_color)
matched = True
if idx > 0 and len(
images_target_chunk) > 0:
print(
f"Saved: {base_image_original} => {file_name} current size chunk {len(images_target_chunk)} - {id_img}"
)
del images_target_chunk[idxInt]
print(
f"{base_image_original}: after delete: {len(images_target_chunk)} - {id_img}"
)
break
if len(good) >= 125:
print('::::::::break Good:::::::')
break
# print(f"{base_image_original}: acabo chunk : {len(images_target_chunk)} - {id_img}")
#
# print(f"{base_image_original}: ACABO SUBJECT : {len(images_target_chunk)} - {images_subject[id_img]}")
# print(f"GEO PORTAILS {len(geo_portail)}")
print('Done this Image')
return True
except ValueError as e:
print('main: saw error "{}" when accessing result'.format(e))
return False
img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
img2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)
figure, ax = plt.subplots(1, 3, figsize=(16, 8))
ax[0].imshow(img1, cmap='gray')
ax[2].imshow(img2, cmap='gray')
#sift
sift = cv2.xfeatures2d.SIFT_create()
keypoints_1, descriptors_1 = sift.detectAndCompute(img1,None)
keypoints_2, descriptors_2 = sift.detectAndCompute(img2,None)
len(keypoints_1), len(keypoints_2)
#feature matching
bf = cv2.BFMatcher(cv2.NORM_L1, crossCheck=True)
matches = bf.match(descriptors_1,descriptors_2)
matches = sorted(matches, key = lambda x:x.distance)
img3 = cv2.drawMatches(img1, keypoints_1, img2, keypoints_2, matches[:50], img2, flags=2)
ax[1].imshow(img3)
plt.savefig(f"{path}result2.jpg")
plt.show()
# %%
plt.show()
# %%
def make_diff(fpath, diff_path):
image_list = os.listdir(fpath)
if '.DS_Store' in image_list:
image_list.remove('.DS_Store')
neighbour_list = os.listdir(fpath)
if '.DS_Store' in neighbour_list:
neighbour_list.remove('.DS_Store')
image_path = osp.join(fpath, 'NormalShot.jpg')
neighbour_path = osp.join(fpath, 'WideShot.jpg')
# Affine Transform
img_ = cv2.imread(image_path)
img_ = cv2.resize(img_, (512, 512))
img1 = cv2.cvtColor(img_, cv2.COLOR_BGR2GRAY)
img = cv2.imread(neighbour_path)
img = cv2.resize(img, (512, 512))
img2 = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
mean = np.average(img1)
mmax = np.amax(img1)
mmin = np.amin(img1)
variance = mmax - mmin
# lb = mean - variance / 2
# ub = mean + variance / 2
is_align = False
try:
# sift = cv2.xfeatures2d.SIFT_create(contrastThreshold=0.03,edgeThreshold=10000)
sift = cv2.xfeatures2d.SIFT_create()
# find the keypoints and descriptors with SIFT
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)
# img3 = cv2.drawKeypoints(img1, kp1, img1)
# cv2.imwrite(diff_path+'/sift_keypoints1.png', img3)
# cv2.imwrite(diff_path+'/original1.png', img1)
# img3 = cv2.drawKeypoints(img2, kp2, img2)
# cv2.imwrite(diff_path+'/sift_keypoints2.png', img3)
# cv2.imwrite(diff_path+'/original2.png', img2)
#
# stereo = cv2.StereoBM_create()
# disparity = stereo.compute(img1, img2)
# cv2.imwrite(diff_path+'/disparity.png', disparity)
bf = cv2.BFMatcher()
matches = bf.knnMatch(des1, des2, k=2)
# Apply ratio test
good = []
for m in matches:
if m[0].distance < 0.5 * m[1].distance:
good.append(m)
matches = np.asarray(good)
if len(matches[:, 0]) < 4:
raise AssertionError("Can’t find enough keypoints")
src = np.float32([kp1[m.queryIdx].pt
for m in matches[:, 0]]).reshape(-1, 1, 2)
dst = np.float32([kp2[m.trainIdx].pt
for m in matches[:, 0]]).reshape(-1, 1, 2)
H, masked = cv2.findHomography(src, dst, cv2.RANSAC, 5)
# save_path = "/Users/jason/Documents/GitHub/DAIN_py/tmp"
# cv2.imwrite(save_path + '/view1.png', img_)
# cv2.imwrite(save_path + '/view2.png', img)
dst = cv2.warpPerspective(img1, H, (img_.shape[1], img_.shape[0]))
# plt.imshow(img_)
# plt.show()
# plt.figure()
# plt.imshow(img)
# plt.show()
# plt.figure()
# plt.imshow(dst)
# plt.show()
# plt.figure()
idx = (dst == 0)
# diff =cv2.cvtColor(img-dst, cv2.COLOR_BGR2GRAY)
diff = np.absolute(img2 - dst)
diff[idx] = mean
# diff = cv2.normalize(diff, None, alpha=0, beta=1, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_32F)
# print(diff)
# diff = cv2.equalizeHist(diff)
# diff = cv2.medianBlur(diff,3)
# diff = cv2.blur(diff, (3,3))
# diff = cv2.GaussianBlur(diff, (3,3),0)
diff = cv2.bilateralFilter(diff, 9, 75, 75)
# plt.imshow(diff)
# plt.show()
# plt.figure()
# cv2.imwrite('/Users/jason/Documents/GitHub/DAIN_py/tmp/crop1.png', dst)
# cv2.imwrite('/Users/jason/Documents/GitHub/DAIN_py/tmp/crop2.png', img)
# cv2.imwrite(save_path+'/diff.png',diff)
# diff = cv2.bilateralFilter(diff, 9, 75, 75)
is_align = True
except:
# diff = cv2.cvtColor(img - img_, cv2.COLOR_BGR2GRAY)
diff = np.absolute(img2 - img1)
# diff = cv2.cvtColor(diff, cv2.COLOR_BGR2GRAY)
diff = cv2.normalize(diff, None, mmin, mmax, norm_type=cv2.NORM_MINMAX)
diff = cv2.resize(diff, (240, 240))
diff = np.stack((diff, ) * 3, axis=-1)
cv2.imwrite(diff_path + '/diff.png', diff)
return is_align
def match(detector):
#Reading the image pair
img1 = cv2.imread('../Image_Pairs/torb_small1.png')
print("Dimension of image 1:", img1.shape[0], "rows x", img1.shape[1],
"columns")
print("Type of image 1:", img1.dtype)
img2 = cv2.imread('../Image_Pairs/torb_small2.png')
print("Dimension of image 2:", img2.shape[0], "lignes x", img2.shape[1],
"columns")
print("Type of image 2:", img2.dtype)
#Beginning the calculus
t1 = cv2.getTickCount()
#Creation of objects "keypoints"
if detector == 1:
kp1 = cv2.ORB_create(
nfeatures=500, #By default : 500
scaleFactor=1.2, #By default : 1.2
nlevels=8) #By default : 8
kp2 = cv2.ORB_create(nfeatures=500, scaleFactor=1.2, nlevels=8)
print("Detector: ORB")
else:
kp1 = cv2.KAZE_create(
upright=False, #By default : false
threshold=0.001, #By default : 0.001
nOctaves=4, #By default : 4
nOctaveLayers=4, #By default : 4
diffusivity=2) #By default : 2
kp2 = cv2.KAZE_create(upright=False,
threshold=0.001,
nOctaves=4,
nOctaveLayers=4,
diffusivity=2)
print("Detector: KAZE")
#Conversion to gray scale
gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)
#Detection and description of keypoints
pts1, desc1 = kp1.detectAndCompute(gray1, None)
pts2, desc2 = kp2.detectAndCompute(gray2, None)
t2 = cv2.getTickCount()
time = (t2 - t1) / cv2.getTickFrequency()
print("Detection of points and computation of descriptors:", time, "s")
# Beginning of matching
t1 = cv2.getTickCount()
if detector == 1:
#Hamming distance for descriptor BRIEF (ORB)
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=False)
else:
#L2 distance for descriptor M-SURF (KAZE)
bf = cv2.BFMatcher(cv2.NORM_L2, crossCheck=False)
# 2-nearest-neighbours list extraction
matches = bf.knnMatch(desc1, desc2, k=2)
# Application of the ratio test
good = []
for m, n in matches:
if m.distance < 0.7 * n.distance:
good.append([m])
t2 = cv2.getTickCount()
time = (t2 - t1) / cv2.getTickFrequency()
print("Matching Computation:", time, "s")
# Displaying the matches that respect the ratio test
draw_params = dict(matchColor=(0, 255, 0),
singlePointColor=(255, 0, 0),
flags=0)
img3 = cv2.drawMatchesKnn(gray1, pts1, gray2, pts2, good, None,
**draw_params)
Nb_ok = len(good)
return img3, Nb_ok
def sift_bfma(im1, im2, lowe_ratio, min_match, display, ransac_th):
import numpy as np
import cv2 as cv
from matplotlib import pyplot as plt
from PIL import Image
#Reading images
img1 = cv.imread(im1, 1)
img2 = cv.imread(im2, 1)
#Making sift object - it takes default but also can take
#(nfeatures,nOctaveLayers,contrastThreshold,edgeThreshold,sigma)
sift = cv.xfeatures2d.SIFT_create()
#Exporting to grayscale
gray1 = cv.cvtColor(img1, cv.COLOR_BGR2GRAY)
gray2 = cv.cvtColor(img2, cv.COLOR_BGR2GRAY)
#Detecting keypoints and computing descriptors for every keypoint
kp1, des1 = sift.detectAndCompute(gray1, None)
kp2, des2 = sift.detectAndCompute(gray2, None)
bf = cv.BFMatcher()
matches = bf.knnMatch(des1, des2, k=2)
# store all the good matches as per Lowe's ratio test.
if lowe_ratio == None:
lowe_ratio = 0.75
good = []
for m, n in matches:
if m.distance < lowe_ratio * n.distance:
good.append(m)
if min_match == None:
MIN_MATCH_COUNT = 10
else:
MIN_MATCH_COUNT = min_match
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([kp1[m.queryIdx].pt
for m in good]).reshape(-1, 1, 2)
dst_pts = np.float32([kp2[m.trainIdx].pt
for m in good]).reshape(-1, 1, 2)
if ransac_th == None:
ransac_th = 5.0
M, mask = cv.findHomography(src_pts, dst_pts, cv.RANSAC, ransac_th)
matchesMask = mask.ravel().tolist()
h, w, d = img1.shape
pts = np.float32([[0, 0], [0, h - 1], [w - 1, h - 1],
[w - 1, 0]]).reshape(-1, 1, 2)
dst = cv.perspectiveTransform(pts, M)
img2 = cv.polylines(img2, [np.int32(dst)], True, 255, 3, cv.LINE_AA)
else:
print("Not enough matches are found - {}/{}".format(
len(good), MIN_MATCH_COUNT))
matchesMask = None
draw_params = dict(
matchColor=(0, 255, 0), # draw matches in green color
singlePointColor=None,
matchesMask=matchesMask, # draw only inliers
flags=2)
img3 = cv.drawMatches(img1, kp1, img2, kp2, good, None, **draw_params)
if display != None:
plt.imshow(cv.cvtColor(img3,
cv.COLOR_BGR2RGB)), plt.axis("off"), plt.show()
return (img1, img2, img3, good)
0, general_settings["camera_focal_length_y"],
general_settings["camera_center_x"]
], [0, 0, 1]],
dtype="double")
# Coefficients are [k1, k2, p1, p2, k3] as per OpenCV documentation
distortion_coefficients = np.array([
general_settings["camera_k1"], general_settings["camera_k2"],
general_settings["camera_p1"], general_settings["camera_p2"],
general_settings["camera_k3"]
])
#################################
# Detect features on the markers, compute pixel dimensions for coordinate calculation.
feature_detector = cv2.ORB_create() #Initialise the ORB feature detector.
matcher = cv2.BFMatcher(
cv2.NORM_HAMMING2, crossCheck=True
) #Initialise the Brute Force matcher. Change the algorithm if you think it's reasonable to do so.
for i in range(0, general_settings["no_of_markers"]):
#In this loop, for each marker as defined in the config file, we:
# 1., Load the image
markers[i]["image_data"] = cv2.imread(markers[i]["file_name"],
cv2.IMREAD_COLOR)
# 2., Calculate pixel width and height based on specified marker dimensions
markers[i]["height_in_px"], markers[i]["width_in_px"], _ = markers[i][
"image_data"].shape
markers[i]["pixel_width"] = markers[i]["width"] / markers[i][
"width_in_px"] #Width of one pixel
markers[i]["pixel_height"] = markers[i]["height"] / markers[i][
"height_in_px"] #Height of one pixel
# 3., Identify features (key_points and descriptors)
def main(opt):
#region Initialization
ARTools.log.setLevel(getattr(logging, opt.log.upper(), None))
# descriptor
if opt.detector == 'SIFT':
detector = cv.SIFT_create()
log.warning('Pipeline use SIFT detector (realtime is compromised)')
elif opt.detector == 'ORB':
detector = cv.ORB_create()
log.info('Pipeline use ORB detector')
else:
detector = cv.AKAZE_create()
log.info('Pipeline use AKAZE detector')
# Matcher
if opt.matcher == 'L2':
matcher = cv.BFMatcher(normType=cv.NORM_L2)
log.warning('Pipeline use BRUTEFORCE matcher')
elif opt.matcher == 'FLANN':
matcher = cv.FlannBasedMatcher()
log.info('Pipeline use FLANNBASED matcher')
else:
matcher = cv.BFMatcher(normType=cv.NORM_HAMMING)
log.info('Pipeline use BRUTEFORCE_HAMMING matcher')
pipeline = ARPipeline(capture=opt.source,
width=opt.x,
height=opt.y,
calibration=opt.calibration,
detector=detector,
matcher=matcher)
pipeline.loadMarker(opt.marker)
model = OBJ(opt.object, swapyz=True)
#endregion
#region AR Pipeline
while not Camera.checkKey():
frame = pipeline.getFrame()
if frame is not None:
# Compute pose estimation
matches, frame_kp = pipeline.computeMatches(frame)
homography, _ = pipeline.computeHomography(
matches, frame_kp, minMatches=opt.minMatches)
matches_refined = pipeline.refineMatches(matches,
frame_kp,
minMatches=opt.minMatches)
homography_refined, _ = pipeline.computeHomography(
matches_refined, frame_kp, minMatches=opt.minMatches)
warped, homography_warped = pipeline.warpMarker(
frame, homography_refined, minMatches=opt.minMatches)
rvecs, tvecs = pipeline.computePose(frame,
homography_warped) #@TODO pnp
# Rendering
ar = frame.copy()
#ar=pipeline.renderer.draw2DRectangle(ar,homography,color=(255,0,0))
#ar=pipeline.renderer.draw2DRectangle(ar,homography_refined,color=(0,255,0))
#ar=pipeline.renderer.draw2DRectangle(ar,homography_warped,color=(0,0,255))
#ar=pipeline.renderer.draw3DCube(ar,rvecs,tvecs)
ar = pipeline.renderer.drawObj(ar,
homography_warped,
model,
eye=0.4)
cv.imshow('AR Camera (press "esc" to quit)', ar)
if opt.all:
cv.imshow('Keypoints',
pipeline.renderer.drawKeypoints(frame, frame_kp))
img_matches = pipeline.renderer.drawMatches(
frame, frame_kp, matches, maxMatches=opt.maxMatches)
img_matches = cv.resize(img_matches,
(frame.shape[1], frame.shape[0]))
cv.imshow('Matches', img_matches)
img_matches_refined = pipeline.renderer.drawMatches(
frame,
frame_kp,
matches_refined,
maxMatches=opt.maxMatches)
img_matches_refined = cv.resize(
img_matches_refined, (frame.shape[1], frame.shape[0]))
cv.imshow('Matches refined', img_matches_refined)
cv.imshow('Warp', warped)
# Initialize position of window once
if not opt.unmove:
opt.unmove = True
ypadding = 60
cv.moveWindow('AR Camera (press "esc" to quit)', 0, 0)
cv.moveWindow('Keypoints', frame.shape[1], 0)
cv.moveWindow('Matches', 2 * frame.shape[1], 0)
cv.moveWindow('Matches refined', 2 * frame.shape[1],
frame.shape[0] + ypadding)
cv.moveWindow('Warp', frame.shape[1],
frame.shape[0] + ypadding)
else:
# No frame available
break
def __init__(self, nfeatures=500, draw=False):
self.SIFT = cv2.xfeatures2d_SIFT.create(nfeatures=nfeatures)
self.bf = cv2.BFMatcher()
self.draw = draw
kpts2, desc2 = sift.detectAndCompute(img2, None)
print len(kpts1)
print len(kpts2)
# cv2.imshow('img1', img1)
# cv2.imshow('img2', img2)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
## second neighbour pre-filtering
nnThreshold = 0.8
time1 = clock()
tree = spatial.KDTree(desc1, leafsize=10)
time2 = clock()
print 'time: ', time2 - time1
query_desc = desc2
matcher = cv2.BFMatcher(cv2.NORM_L2)
time3 = clock()
raw_matches = matcher.knnMatch(desc2, trainDescriptors=desc1, k=2)
time4 = clock()
print 'query time: ', time4 - time3
print len(raw_matches)
print raw_matches[1][0].queryIdx
print raw_matches[1][0].trainIdx
print raw_matches[0][0].imgIdx
print raw_matches[1][0].imgIdx
print raw_matches[1][0].distance
print type(raw_matches)
good_match = []
for i in range(len(raw_matches)):
if (raw_matches[i][0].distance / raw_matches[i][1].distance) <= float(
def detect():
if request.method == 'POST':
f = request.files['file']
f.save(os.path.join(app.config['UPLOAD_FOLDER'], "test.jpg"))
notes = [
"10.jpg", "10_new.jpg", "20.jpg", "20_new.jpg", "50.jpg", "50_new.jpg",
"100.jpg", "100_new.jpg", "200.jpg", "500.jpg", "2000.jpg", "test.jpg"
]
res = []
for i in range(len(notes) - 1):
img1 = cv2.imread("images//" + notes[i], 0)
img2 = cv2.imread("images//" + notes[11], 0)
img1 = resize_img(img1, 0.5)
img2 = resize_img(img2, 0.2)
# ORB Detector
orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)
#print(des1)
# Brute Force Matching
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1, des2)
#print(matches)
print(len(matches))
res.append(len(matches))
matches = sorted(matches, key=lambda x: x.distance)
matching_result = cv2.drawMatches(img1,
kp1,
img2,
kp2,
matches[:50],
None,
flags=2)
#cv2.imshow("Test", img2)
start_point = (0, 0)
end_point = (0, 0)
color = (0, 0, 0)
thickness = 0
image = cv2.rectangle(img2, start_point, end_point, color, thickness)
#cv2.imshow("result", image)
#cv2.imshow("Matching result", matching_result)
cv2.waitKey(0)
cv2.destroyAllWindows()
val = res.index(max(res))
print(res.index(max(res)))
font = cv2.FONT_HERSHEY_SIMPLEX
deno = 0
if (val == 0 or val == 1):
deno = 10
elif (val == 2 or val == 3):
deno = 20
elif (val == 4 or val == 5):
deno = 50
elif (val == 6 or val == 7):
deno = 100
elif (val == 8):
deno = 200
elif (val == 9):
deno = 500
else:
deno = 2000
font = cv2.FONT_HERSHEY_SIMPLEX
org = (25, 25)
fontScale = 1
color = (255, 0, 0)
thickness = 2
image = cv2.putText(image,
'Currency Amount Detected is ' + str(deno) + ' INR',
org, font, fontScale, color, thickness, cv2.LINE_AA)
flash('Currency Amount Detected is ' + str(deno) + ' INR')
with open('config.json') as f:
data = json.load(f)
cnt = data['params']['count']
data['params']['count'] = data['params']['count'] + 1
with open('config.json', 'w') as f:
json.dump(data, f)
name = "res" + str(cnt) + ".jpg"
cv2.imwrite("static//result//" + name, image)
address = "static/result/res" + str(cnt) + ".jpg"
print(address)
return render_template('index.html', addr=address)
) # ORB is a fast working algorithm and it's free unlike swift/surf
kp1, des1 = orb.detectAndCompute(
img1, None) # kp1 will be the features, None is meant for the mask
kp2, des2 = orb.detectAndCompute(img2, None)
# descriptors details
# print(des1)
# print(des1.shape) # (498, 32)
# print(des2.shape) # (500, 32)
# ORB detector uses 500 features by default. So it will try to find 500 features in the images.
# For each feature, ORB will describe it in 32 values.
# Now that we have the descriptors, we can use a "matcher" to match these descriptors together.
bf = cv2.BFMatcher() # brute force matcher
matches = bf.knnMatch(
des1, des2, k=2) # k=2 as we want two values that we can compare later.
# To decide whether it's a good match or not
good_matches = []
for m, n in matches: # m, n are values from k=2
if m.distance < 0.75 * n.distance:
good_matches.append([m])
print(len(good_matches))
# Plotting the good matches
img3 = cv2.drawMatchesKnn(img1, kp1, img2, kp2, good_matches, None, flags=2)
# To view the keypoints
def __init__(self):
self.sift = cv2.xfeatures2d.SIFT_create()
self.brute = cv2.BFMatcher()
def match(detector):
#Reading the image pair
img1 = cv2.imread('../Image_Pairs/torb_small1.png')
print("Dimension of image 1:", img1.shape[0], "rows x", img1.shape[1],
"columns")
print("Type of image 1:", img1.dtype)
img2 = cv2.imread('../Image_Pairs/torb_small2.png')
print("Dimension of image 2:", img2.shape[0], "lignes x", img2.shape[1],
"columns")
print("Type of image 2:", img2.dtype)
#Beginning the calculus
t1 = cv2.getTickCount()
#Creation of objects "keypoints"
if detector == 1:
kp1 = cv2.ORB_create(
nfeatures=500, #By default : 500
scaleFactor=1.2, #By default : 1.2
nlevels=8) #By default : 8
kp2 = cv2.ORB_create(nfeatures=500, scaleFactor=1.2, nlevels=8)
print("Detector: ORB")
else:
kp1 = cv2.KAZE_create(
upright=False, #By default : false
threshold=0.001, #By default : 0.001
nOctaves=4, #By default : 4
nOctaveLayers=4, #By default : 4
diffusivity=2) #By default : 2
kp2 = cv2.KAZE_create(upright=False,
threshold=0.001,
nOctaves=4,
nOctaveLayers=4,
diffusivity=2)
print("Detector: KAZE")
#Conversion to gray scale
gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)
#Detection and description of keypoints
pts1, desc1 = kp1.detectAndCompute(gray1, None)
pts2, desc2 = kp2.detectAndCompute(gray2, None)
#Un-matched points will appear in grey
img1 = cv2.drawKeypoints(gray1, pts1, None, color=(127, 127, 127), flags=0)
img2 = cv2.drawKeypoints(gray2, pts2, None, color=(127, 127, 127), flags=0)
t2 = cv2.getTickCount()
time = (t2 - t1) / cv2.getTickFrequency()
print("Detection points and descriptors computation:", time, "s")
# Beginning of Matching
t1 = cv2.getTickCount()
if detector == 1:
#Hamming distance for descriptor BRIEF (ORB)
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
else:
#L2 distance for descriptor M-SURF (KAZE)
bf = cv2.BFMatcher(cv2.NORM_L2, crossCheck=True)
matches = bf.match(desc1, desc2)
# Sorting the matches
matches = sorted(matches, key=lambda x: x.distance)
t2 = cv2.getTickCount()
time = (t2 - t1) / cv2.getTickFrequency()
print("Matching Computation:", time, "s")
# Display the N best matches
Nbest = 200
img3 = cv2.drawMatches(img1,
pts1,
img2,
pts2,
matches[:Nbest],
None,
flags=2)
return img3, Nbest
mask = cv.normalize(mask, mask, 0, 1, cv.NORM_MINMAX)
#PATH = 'C:/Users/zszentim/Documents/ENGG6100Assign3/rutgers_apc_dataset/all_data/cheezit_big_original-pose-C-1-2-0.yml'
img2 = img2 * mask
img1 = cv.normalize(img1, img1, 0, 255, cv.NORM_MINMAX)
img2 = cv.normalize(img2, img2, 0, 255, cv.NORM_MINMAX)
# Initiate SIFT detector
sift = cv.xfeatures2d.SIFT_create()
# find the keypoints and descriptors with SIFT
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)
# BFMatcher with default params
bf = cv.BFMatcher()
matches = bf.knnMatch(des1, des2, k=2)
newm = []
# Apply ratio test
good = []
good_without_list = []
for m, n in matches:
if m.distance < 0.75 * n.distance: #0.6 for cheezit but I should try a higher value, might get better results
good.append([m])
good_without_list.append(m)
#cv.drawMatchesKnn expects list of lists as matches.
img3 = cv.drawMatchesKnn(
img1,
kp1,
img2,
for i in corners_2:
x, y = i.ravel()
cv2.circle(img_zero_2, (x, y), 1, (255, 255, 255), 2)
fe_2 = fe_2 + 1
cv2.imshow("src", img_zero_1)
cv2.imshow("dst", img_zero_2)
print(fe_1)
print(fe_2)
detector = cv2.AKAZE_create()
kpts1, desc1 = detector.detectAndCompute(img_zero_1, None)
kpts2, desc2 = detector.detectAndCompute(img_zero_2, None)
matcher = cv2.BFMatcher(cv2.NORM_HAMMING)
matches = matcher.match(desc1, desc2)
h1 = img_zero_1.shape[0]
w1 = img_zero_1.shape[1]
h2 = img_zero_2.shape[0]
w2 = img_zero_2.shape[1]
img_dst = np.zeros((max(h1, h2), w1 + w2, 3), np.uint8)
cv2.drawMatches(img_src_1, kpts1, img_src_2, kpts2, matches, img_dst)
cv2.imwrite("output.jpg", img_dst)
cv2.imshow('out', img_dst)
cv2.waitKey(0)
cv2.destroyAllWindows()
# -*- coding: utf-8 -*-
"""
Created on Mon Mar 19 13:26:59 2018
@author: dpr
"""
"""
will match an object contained in 2 different images without caring about orientation
"""
import numpy as np
import cv2
import matplotlib.pyplot as plt
img1 = cv2.imread('F:/temp/opencv-feature-matching-template.jpg', 0)
img2 = cv2.imread('F:/temp/opencv-feature-matching-image.jpg', 0)
orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1, des2)
matches = sorted(matches, key=lambda x: x.distance)
img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[:10], None, flags=2)
plt.imshow(img3)
plt.show()
def init_feature():
detector = cv2.BRISK_create()
norm = cv2.NORM_HAMMING
matcher = cv2.BFMatcher(norm)
return detector, matcher
import numpy as np
import cv2
dataset = cv2.imread('dataset/imagemPare_11.jpg')
target = cv2.imread('target/pare.jpg')
datasetGray = cv2.cvtColor(dataset, cv2.COLOR_BGR2GRAY)
targetGray = cv2.cvtColor(target, cv2.COLOR_BGR2GRAY)
sift = cv2.xfeatures2d.SIFT_create()
kpDataset, desDataset = sift.detectAndCompute(datasetGray, None)
kpTarget, desTarget = sift.detectAndCompute(targetGray, None)
bf = cv2.BFMatcher(cv2.NORM_L1, crossCheck=False)
matches = bf.match(desDataset, desTarget)
matches = sorted(matches, key=lambda x: x.distance)
img3 = cv2.drawMatches(dataset,
kpDataset,
target,
kpTarget,
matches[:10],
None,
flags=2)
cv2.imshow('teste1', img3)
img3 = cv2.drawMatches(dataset,
class board:
mask = None
board_field = None
showflag = True
exact = None
moneys = []
money_files = {"1b.png": "1 yuan back", "1f.png": "1 yuan front",
"10b.png": "10 yuan back", "10f.png": "10 yuan front",
"20b.png": "20 yuan back", "20f.png": "20 yuan front",
"50b.png": "50 yuan back", "50f.png": "50 yuan front",
"100b.png": "100 yuan back", "100f.png": "100 yuan front",
"5b.png": "5 yuan back", "5f.png": "5 yuan front",
"1jb.png":"1 jiao back","1jf.png": "1 jiao front",
"5jb.png": "5 jiao back", "5jf.png": "5 jiao front"
}
detector = cv2.ORB(800)
norm = cv2.NORM_HAMMING
matcher = cv2.BFMatcher(norm)
def __init__(self):
self.cam = cv2.VideoCapture(0)
self.cam.set(3, 1280)
self.cam.set(4, 720)
w = self.cam.get(3)
h = self.cam.get(4)
print w, h
cv2.namedWindow("cam")
cv2.setMouseCallback("cam", self.on_mouse)
# img1 = cv2.bilateralFilter(img1, 11, 17, 17) # 双边滤波
# img1 = cv2.Canny(img1, 30, 200) # 边缘canny检测
for d in self.money_files:
print d
img = cv2.imread(d)
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
kp, desc = self.detector.detectAndCompute(img, None)
self.moneys.append((desc, self.money_files[d]))
def on_mouse(self, event, x, y, flags, param):
if event == cv2.EVENT_LBUTTONDOWN:
print (x, y)
if x < 100 and y < 100: # 点击窗体左上角完成保存文件
im = None
if self.showflag:
im = self.img
else:
im = self.hsv
if im != None: # 保存图片至png文件
cv2.imwrite(datetime.now().strftime(
"%m%d%H%M%S") + ".png", im) # 时间
print "save png file to:\n", datetime.now().strftime("%m%d%H%M%S") + ".png"
def rotate_about_center(self, src, angle, scale=1.):
w = src.shape[1]
h = src.shape[0]
rangle = np.deg2rad(angle) # angle in radians
# now calculate new image width and height
nw = (abs(np.sin(rangle) * h) + abs(np.cos(rangle) * w)) * scale
nh = (abs(np.cos(rangle) * h) + abs(np.sin(rangle) * w)) * scale
# ask OpenCV for the rotation matrix
rot_mat = cv2.getRotationMatrix2D((nw * 0.5, nh * 0.5), angle, scale)
# calculate the move from the old center to the new center combined
# with the rotation
rot_move = np.dot(rot_mat, np.array(
[(nw - w) * 0.5, (nh - h) * 0.5, 0]))
# the move only affects the translation, so update the translation
# part of the transform
rot_mat[0, 2] += rot_move[0]
rot_mat[1, 2] += rot_move[1]
return cv2.warpAffine(src, rot_mat, (int(math.ceil(nw)), int(math.ceil(nh))), flags=cv2.INTER_LANCZOS4)
def find_contours(self, img):
self.gray = img.copy()
# img=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
# img=cv2.bitwise_not(img) #图像取反
# img = cv2.GaussianBlur(img, (5, 5), 0) # Gauss滤波
img = cv2.bilateralFilter(img, 11, 17, 17) # 双边滤波
# t,img=cv2.threshold(img, 0, 255, cv2.THRESH_OTSU ) #二值化
img = cv2.Canny(img, 30, 200) # 边缘canny检测
# img=cv2.dilate(img,cv2.getStructuringElement(cv2.MORPH_CROSS,(3,3))) #膨胀算法
# img=cv2.medianBlur(img,3) #中值滤波
# self.hsv=img.copy() #保存以便显示
cnts, _ = cv2.findContours(
img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
cnts = sorted(cnts, key=cv2.contourArea, reverse=True)[:20]
squares = []
for cnt in cnts:
cnt_len = cv2.arcLength(cnt, True)
cnt = cv2.approxPolyDP(cnt, 0.02 * cnt_len, True) # 最小包裹
if len(cnt) == 4 and cv2.isContourConvex(cnt): # 4边凸包
squares.append(cnt)
break
if len(squares) > 0:
screenCnt = squares[0]
pts = screenCnt.reshape(4, 2)
rect = np.zeros((4, 2), dtype="float32")
# the top-left point has the smallest sum whereas the
# bottom-right has the largest sum
s = pts.sum(axis=1)
rect[0] = pts[np.argmin(s)]
rect[2] = pts[np.argmax(s)]
# compute the difference between the points -- the top-right
# will have the minumum difference and the bottom-left will
# have the maximum difference
diff = np.diff(pts, axis=1)
rect[1] = pts[np.argmin(diff)]
rect[3] = pts[np.argmax(diff)]
(tl, tr, br, bl) = rect
widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))
widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))
# ...and now for the height of our new image
heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))
heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))
# take the maximum of the width and height values to reach
# our final dimensions
maxWidth = max(int(widthA), int(widthB))
maxHeight = max(int(heightA), int(heightB))
ratio = max(maxHeight, maxWidth) / 600.0
maxWidth, maxHeight = int(maxWidth / ratio), int(maxHeight / ratio)
dst = np.array([
[0, 0],
[maxWidth - 1, 0],
[maxWidth - 1, maxHeight - 1],
[0, maxHeight - 1]], dtype="float32")
M = cv2.getPerspectiveTransform(rect, dst)
self.exact = cv2.warpPerspective(
self.img, M, (maxWidth, maxHeight))
if maxWidth < maxHeight:
self.exact = self.rotate_about_center(self.exact, 90)
#self.exact = cv2.GaussianBlur(self.exact, (7, 7), 0)
#self.exact = cv2.Canny(self.exact, 5, 100)
self.recognition(self.exact)
cv2.drawContours(self.img, squares, -1, (255, 128, 0), 3)
def recognition(self, img):
kp1, desc1 = self.detector.detectAndCompute(img, None)
for md, mn in self.moneys:
matches = self.matcher.knnMatch(desc1, md, k=2)
good = []
for m, n in matches:
# print m.distance, n.distance, m.distance / n.distance
# filter those pts similar to the next good ones
if m.distance < 0.7 * n.distance:
good.append(m)
if len(good) > 5:
print mn,":",len(good)
def run(self): # 主函数
while True:
ret, self.img = self.cam.read() # 读取摄像头图片
self.gray = cv2.cvtColor(self.img, cv2.COLOR_BGR2GRAY)
self.recognition(self.gray.copy())
if self.showflag: # 切换显示窗体
cv2.imshow("cam", self.img)
cv2.setMouseCallback("cam", self.on_mouse) # 重新注册鼠标事件
if self.exact != None:
cv2.imshow("exact", self.exact)
key = cv2.waitKey(20)
if(key == 27):
break # esc退出程序
self.cam.release()
cv2.destroyAllWindows()