X = []
Y = []
np.array(X)
np.array(Y)
finger_center = (int(finger_box[0] + finger_box[2] / 2),
int(finger_box[1] + finger_box[3] / 2))
phone_center = (int(phone_box[0] + phone_box[2] / 2),
int(phone_box[1] + phone_box[3] / 2))
X.append(finger_center[0] - phone_center[0])
Y.append(finger_center[1] - phone_center[1])
# Set up tracker.
finger_tracker = cv2.TrackerCSRT_create()
phone_tracker = cv2.TrackerCSRT_create()
finger_ok = finger_tracker.init(frame, finger_box)
phone_ok = phone_tracker.init(frame, phone_box)
while True:
# Read a new frame
ok, frame = video.read()
if not ok:
break
# Start timer
timer = cv2.getTickCount()
# Update tracker
#else, need to explicitly call appropriate object tracker constructor
#else:
#initialise a dict that maps string to their corresponding opencv obj tracker implementation
#OPENCV_OBJECT_TRACKERS = {
#"csrt": cv2.TrackerCSRT_create,
#"kcf": cv2.TrackerKCF_create,
#"boosting": cv2.TrackerBoosting_create,
#"mil": cv2.TrackerMIL_create
#"tld": cv2.TrackerTLD_create,
#"medianflow": cv2.TrackerMedianFlow_create,
#"mosse": cv2.TrackerMOSSE_create
#}
#grab appropriate obj tracker using dict of opencv obj tracker objects
else:
tracker = cv2.TrackerCSRT_create(
) #OPENCV_OBJECT_TRACKERS[args["tracker"]]()
#init bounding box coords of obj to track
initBB = None
print("[INFO] starting video stream...")
vs = VideoStream(src=0).start()
time.sleep(1.0)
#else, grab ref to vid file
#else:
# vs = cv2.VideoCapture(args["video"])
#init fps throughput estimator
fps = None
def get_axis(path, bbox):
# Set up tracker.
sentinel = True
initial_threshold = 0
displacement_nodes = []
# Check if the initial threshold was exceeded by the projectile
threshold_crossed = False
tracker = cv2.TrackerCSRT_create()
# Read video
video = cv2.VideoCapture(path)
# Exit if video not opened.
if not video.isOpened():
print("Could not open video")
sys.exit()
# Read first frame.
ok, frame = video.read()
if not ok:
print('Cannot read video file')
sys.exit()
# Uncomment the line below to select a different bounding box
if args.select:
bbox = cv2.selectROI(frame, False)
# Initialize tracker with first frame and bounding box
ok = tracker.init(frame, bbox)
while sentinel:
# Read a new frame
ok, frame = video.read()
if not ok:
break
# Start timer
timer = cv2.getTickCount()
# Update tracker
ok, bbox = tracker.update(frame)
# Calculate Frames per second (FPS)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
# Draw bounding box
if ok:
# Tracking success
p1 = (int(bbox[0]), int(bbox[1]))
p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
displacement_nodes.append(
(int(bbox[0] + bbox[2] // 2), int(bbox[1] + bbox[3] // 2)))
cv2.rectangle(frame, p1, p2, (255, 0, 0), 2, 1)
if len(displacement_nodes) > 1:
initial_threshold = displacement_nodes[0][1] - 15
if displacement_nodes[-1][0] <= 2:
sentinel = False
if displacement_nodes[-2][1] < displacement_nodes[-1][
1] and not threshold_crossed and len(
displacement_nodes) > 30:
sentinel = False
for i in range(len(displacement_nodes) - 1):
cv2.line(frame, displacement_nodes[i],
displacement_nodes[i + 1], (0, 255, 0), 3)
if int(bbox[1] + bbox[3] // 2) < initial_threshold:
cv2.putText(frame, "Crossed Threshold Upwards", (100, 110),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
threshold_crossed = True
else:
threshold_crossed = False
else:
# Tracking failure
cv2.putText(frame, "Tracking failure detected", (100, 80),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
# Display tracker type on frame
cv2.putText(frame, "CSRT Tracker", (100, 20), cv2.FONT_HERSHEY_SIMPLEX,
0.75, (50, 170, 50), 2)
# Display FPS on frame
cv2.putText(frame, "FPS : " + str(int(fps)), (100, 50),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
if args.show:
# Display result
cv2.imshow("Tracking", frame)
# Exit if ESC pressed
k = cv2.waitKey(1) & 0xff
if k == 27:
break
df = pd.DataFrame(displacement_nodes)
return df
def __init__(self, box, frame):
bbox = (box[0], box[1], box[2] - box[0], box[3] - box[1])
self.tracker = cv2.TrackerCSRT_create()
self.tracker.init(frame, bbox)
def __init__(self):
self.tracker = cv2.TrackerCSRT_create()
multiTracker = cv2.MultiTracker_create()
cv2.putText(frame,
"please hold the marker still and close to the camera",
(5, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
if object != (0, 0, 0, 0):
if np.all(np.abs(np.subtract(posrange, object))) < 8:
counter += 1
cv2.rectangle(frame, (object[0], object[1]),
(object[0] + object[2], object[2] + object[3]),
(0, 0, 255), 2, 1)
print(counter)
else:
counter = 0
posrange = object
if counter > 100:
multiTracker.add(cv2.TrackerCSRT_create(), frame, object)
counter = 0
else:
counter = 0
success, track = multiTracker.update(frame)
if not success:
track = []
multiTracker.clear()
for i, newbox in enumerate(track):
cv2.putText(frame, "tracking", (5, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.7,
(0, 255, 0), 2)
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(frame, p1, p2, (0, 255, 0), 2, 1)
def tracking_video(root, name, point, edge = 20, save_img = True):
# Set up tracker.
# Instead of MIL, you can also use
tracker_types = ['BOOSTING', 'MIL','KCF', 'TLD', 'MEDIANFLOW', 'CSRT', 'MOSSE']
tracker_type = tracker_types[5]
if int(minor_ver) >= 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'CSRT':
tracker = cv2.TrackerCSRT_create()
if tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
# Read video
video = cv2.VideoCapture(join(root, name))
# Exit if video not opened.
if not video.isOpened():
print("Could not open video")
sys.exit()
# Read first frame.
ok, frame = video.read()
if not ok:
print('Cannot read video file')
sys.exit()
cv2.imwrite("result/%05d.jpg"%(0),frame)
# Define an initial bounding box
bbox = (point[0] - edge, point[1] - edge, edge*2 , edge*2)
# Uncomment the line below to select a different bounding box
# bbox = cv2.selectROI(frame, False)
# Initialize tracker with first frame and bounding box
ok = tracker.init(frame, bbox)
ans = []
index_ = 0
while True:
# Read a new frame
index_ += 1
ok, frame = video.read()
if not ok:
break
# Start timer
timer = cv2.getTickCount()
# Update tracker
ok, bbox = tracker.update(frame)
# Calculate Frames per second (FPS)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer);
# Draw bounding box
if ok:
# Tracking success
p1 = (int(bbox[0]), int(bbox[1]))
p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
p3 = (int(bbox[0] + bbox[2]/2), int(bbox[1] + bbox[3]/2))
ans.append(p3)
if (save_img):
cv2.rectangle(frame, p1, p2, (255,0,0), 2, 1)
else :
# Tracking failure
if (save_img):
cv2.putText(frame, "Tracking failure detected", (100,80), cv2.FONT_HERSHEY_SIMPLEX, 0.75,(0,0,255),2)
if (save_img):
# Display tracker type on frame
cv2.putText(frame, tracker_type + " Tracker", (100,20), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50,170,50),2);
# Display FPS on frame
cv2.putText(frame, "FPS : " + str(int(fps)), (100,50), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50,170,50), 2);
# Display result
cv2.imwrite("result/%05d.jpg"%(index_) , frame)
# Exit if ESC pressed
k = cv2.waitKey(1) & 0xff
if k == 27 : break
return np.array(ans)
def simple_multitracker_greedy(img_list):
def dist(point, point_dict):
min_dist = 1000000
min_dist_idx = -1
for key in point_dict.keys():
distance = np.sqrt(
np.sum((np.array(point) - np.array(point_dict[key]))**2))
if distance < min_dist:
min_dist = distance
min_dist_idx = key
return min_dist, min_dist_idx
def best_overlap(bbox,
labels): #finding the object that overlaps with the bbox
max_overlap = 0
max_overlap_idx = -1
bbox_arr = [int(i) for i in bbox]
for obj_idx in np.unique(labels)[1:]:
mask = labels == obj_idx
overlap = np.sum(
mask[bbox_arr[1]:bbox_arr[1] + bbox_arr[3],
bbox_arr[0]:bbox_arr[0] + bbox_arr[2]]) / (
bbox_arr[2] * bbox_arr[3]) #/np.sum(mask)
if overlap > max_overlap:
max_overlap = overlap
max_overlap_idx = obj_idx
return max_overlap, max_overlap_idx
n_frames = len(img_list)
cur_frame = cv2.imread(img_list[0])
cur_labels = detect.label_img(detect.segment_morph(cur_frame, False),
centroids=False)
trackers = cv2.MultiTracker()
ret_images = []
colors = []
ret_frame = cur_frame.copy()
for obj_label in np.unique(
cur_labels): #initializing trackers for each detected object
if obj_label == 0:
continue
bbox = cv2.boundingRect(
np.array(cur_labels == obj_label, dtype='uint8'))
trackers.add(cv2.TrackerCSRT_create(), cur_frame, bbox)
cv2.rectangle(ret_frame, (bbox[0], bbox[1]),
(bbox[0] + bbox[2], bbox[1] + bbox[3]), (0, 255, 0), 2)
colors.append((np.random.randint(0, 255), np.random.randint(0, 255),
np.random.randint(0, 255)))
ret_images.append(ret_frame)
min_overlap = 0.3
for frame_idx in range(1, n_frames):
cur_frame = cv2.imread(img_list[frame_idx])
cur_labels = detect.label_img(detect.segment_morph(cur_frame, False),
centroids=False)
ref_labels = cur_labels.copy()
_, cur_centroids = detect.mark_object_centroids(cur_labels)
ret_frame = cur_frame.copy()
used_box_idx = []
success, boxes = trackers.update(cur_frame)
found_objects = cur_centroids.copy()
for i, newbox in enumerate(boxes):
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
if np.sum(cur_labels[int(newbox[1]):int(newbox[1] + newbox[3]),
int(newbox[0]):int(newbox[0] + newbox[2])]
) > 0: #not showing the 'dead' tracks
overlap, best_matching_obj_idx = best_overlap(
newbox, ref_labels)
if overlap > min_overlap: #asign the object with the closest centroid
#cx, cy = cur_centroids[best_matching_obj_idx]
# p1 = (int(newbox[0]), int(newbox[1]))
# p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(ret_frame, p1, p2, colors[i], 2, 1)
found_objects.pop(best_matching_obj_idx
) #remove this object from the list
ref_labels[ref_labels == best_matching_obj_idx] = 0
used_box_idx.append(i)
#ref_labels[int(newbox[1]): int(newbox[1] + newbox[3]), int(newbox[0]):int(newbox[0] + newbox[2])] = 0
# else: #trying to find the occluding objects
# overlap, best_matching_obj_idx = best_overlap(newbox, cur_labels)
# if distance < max_dist//4:
# cx, cy = cur_centroids[best_matching_obj_idx]
# p1 = (int(newbox[0]), int(newbox[1]))
# p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
#
# cv2.rectangle(ret_frame, p1, p2, colors[i], 2, 1)
# #found_objects.pop(best_matching_obj_idx) # remove this object from the list
if len(found_objects) > 0:
for obj in found_objects.keys():
bbox = cv2.boundingRect(
np.array(cur_labels == obj, dtype='uint8'))
mask = ref_labels == obj
good_bbox = False
for i, newbox in enumerate(boxes):
overlap = np.mean(
mask[int(newbox[1]):int(newbox[1] + newbox[3]),
int(newbox[0]):int(newbox[0] + newbox[2])])
if overlap > min_overlap / 5:
if (i in used_box_idx):
good_bbox = True
break
else:
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]),
int(newbox[1] + newbox[3]))
cv2.rectangle(ret_frame, p1, p2, colors[i], 2, 1)
ref_labels[ref_labels == obj] = 0
used_box_idx.append(i)
good_bbox = True
break
if not good_bbox:
trackers.add(cv2.TrackerCSRT_create(), cur_frame, bbox)
colors.append(
(np.random.randint(0, 255), np.random.randint(0, 255),
np.random.randint(0, 255)))
cv2.rectangle(ret_frame, (bbox[0], bbox[1]),
(bbox[0] + bbox[2], bbox[1] + bbox[3]),
colors[-1], 2)
ret_images.append(ret_frame)
return ret_images
def main():
# capture video from webcam
cap = cv2.VideoCapture(0)
# set video size
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 480)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 800)
# read first frame
success, frame = cap.read()
if not success:
print('Failed to read video.')
sys.exit(1)
# create multi_tracker object
multi_tracker = cv2.multi_tracker_create()
# load mask image
mask_img = cv2.imread('mask.png', cv2.IMREAD_GRAYSCALE)
# width, height, channels = frame.shape
# mask_img = cv2.resize(mask_img, (height, width))
print('press SPACE to register cargo')
colors = []
labels = []
first_frame = None
while cap.isOpened():
success, original_frame = cap.read()
frame = original_frame.copy()
if not success:
print('Failed to read video')
sys.exit(1)
# get updated location of objects in subsequent frames
success, boxes = multi_tracker.update(frame)
# draw tracked objects
for i, newbox in enumerate(boxes):
x, y = newbox[0], newbox[1]
w, h = newbox[2], newbox[3]
p1 = (int(x), int(y))
p2 = (int(x + w), int(y + h))
cv2.rectangle(frame, p1, p2, colors[i], 2, 1)
cv2.putText(frame, labels[i], (int(x),int(y)), cv2.FONT_HERSHEY_SIMPLEX, int(w)*int(h)/100000+0.5,(0,0,255),2)
# draw geofence
frame = geofence(frame, mask_img)
# show frame
#cv2.namedWindow('Cargo Mana', cv2.WINDOW_NORMAL)
cv2.imshow('Cargo Mana', frame)
k = cv2.waitKey(1) & 0xFF
# select object to track on SPACE button
if k == ord(' '):
bbox = cv2.selectROI('multi_tracker', frame)
colors.append((randint(64, 255), randint(64, 255), randint(64, 255)))
labels.append(f'{randint(0, 1000):03}-{randint(0, 1000):02} {randint(0, 1000):04}')
# create CSRT algorithm tracker and add to multi_tracker
multi_tracker.add(cv2.TrackerCSRT_create(), frame, bbox)
success, boxes = multi_tracker.update(frame)
# send data on S button
if k == ord('s'):
payload = {label: dict(zip(('x', 'y', 'w', 'h'), box))
for label, box in zip(labels, boxes)}
ok, original_image = cv2.imencode('.jpg', original_frame)
content_settings = ContentSettings('image/jpeg')
service.create_blob_from_bytes('box', 'frame.jpg', original_image.tobytes(), content_settings=content_settings)
content_settings = ContentSettings('application/json')
service.create_blob_from_text('box', 'box.json', json.dumps(payload), content_settings=content_settings)
print(payload)
# reset geofence on R button
if k == ord('r'):
first_frame = None
# quit on ESC button
if k == 27: # Esc pressed
break
def __init__(self, frame: np.ndarray, bbox: np.ndarray, tid: int):
self.id_ = tid
self.tracker = cv2.TrackerCSRT_create()
self.tracker.init(frame, tuple(bbox))
self.age = 0
self.misses = 0
def Passing_Counter_Zone(Vehicle_x, Vehicle_y, Vehicle_w, Vehicle_h, initBB,
frame, tracker, Substracted, RED_cnt, BLUE_cnt,
vertices, inout):
# Detecting Zone
pts = detecting_zone(vertices)
global car_cnt
# 차량 검출시
for d_num in range(0, len(Vehicle_x)):
# P 좌표는 프레임에서 디텍팅 포인트 영역
p_x = Vehicle_x[d_num] + int(Vehicle_w[d_num] / 2)
p_y = Vehicle_y[d_num] + int(Vehicle_h[d_num])
crosses = 0 # 교점의 개수(짝수개이면 영역 밖에 존재, 홀수개이면 영역 안에 존재)
for p in range(0, 4): # 항상 사각형 이므로 4
next_p = (p + 1) % 4
if (pts[p][1] > p_y) != (pts[next_p][1] >
p_y): ##디텍티드 포인트의 Y좌표가 사각형의 두점사이에 존재하면
# atx가 오른쪽 반직선과의 교점이 맞으면 교점의 개수를 증가시킨다,
atX = int((pts[next_p][0] - pts[p][0]) * (p_y - pts[p][1]) /
(pts[next_p][1] - pts[p][1]) + pts[p][0])
if p_x < atX:
crosses = crosses + 1
##텍스트로 인 아웃 여부
#cv2.putText(frame, str(crosses), (atX, p_y), cv2.FONT_HERSHEY_SIMPLEX, 0.7,COLOR_GREEN, 3)
if crosses % 2 == 0: # 영역 밖에 존재하는 경우
pass
elif crosses % 2 == 1: # 영역 안에 존재하는 경우
if initBB is None:
initBB = (Vehicle_x[d_num], Vehicle_y[d_num], Vehicle_w[d_num],
Vehicle_h[d_num])
# 트래커 활성화
tracker = cv2.TrackerCSRT_create()
tracker.init(
Substracted,
initBB) # 트래커를 원본이미지가 아닌 백그라운드 Substracted 된 이미지에서 트래킹함
# 트래커 활성화시 동작
if initBB is not None:
# grab the new bounding box coordinates of the object
(success, box) = tracker.update(Substracted)
# check to see if the tracking was a success
# 트래킹 성공시
if success:
(x, y, w, h) = [int(v) for v in box]
cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 255),
2) #yellow bounding box
cv2.rectangle(Substracted, (x, y), (x + w, y + h), (0, 255, 255),
2)
cv2.putText(frame, str(car_cnt), (x + int(w / 2) - 10, y + h - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 0, 255), 2)
Tracking_Xp = x + int(w / 2)
Tracking_Yp = y + h
# Tracking Point 와 Detected Point의 거리가 150픽셀 이하인 경우 매칭 및 트래커 박스 재조정
Matched = False
Matched_Xp = 0
Matched_Yp = 0
for i in range(0, len(Vehicle_x), 1):
Vehicle_Xp = Vehicle_x[i] + int(Vehicle_w[i] / 2)
Vehicle_Yp = Vehicle_y[i] + Vehicle_h[i]
#트래커 포인트와 디텍팅 포인트와의 거리가 150이하인 경우
if int(
math.sqrt(
pow(abs(Tracking_Xp - Vehicle_Xp), 2) +
pow(abs(Tracking_Yp - Vehicle_Yp), 2))) < 200:
cv2.line(frame, (Tracking_Xp, Tracking_Yp),
(Vehicle_Xp, Vehicle_Yp), (125, 255, 125), 2)
Matched = True
Matched_Xp = Vehicle_Xp
Matched_Yp = Vehicle_Yp
#트래커의 박스를 재조정하기위한 Bounding box
tempBB = (Vehicle_x[i], Vehicle_y[i], Vehicle_w[i],
Vehicle_h[i])
#트래커삭제 및 갱신
tracker = cv2.TrackerCSRT_create()
tracker.init(
Substracted, tempBB
) # 트래커를 원본이미지가 아닌 백그라운드 Substracted 된 이미지에서 트래킹함
break
# tracker가 영역 밖에 존재하는 경우 - 삭제 (트래커기준이 아니라, 매칭된 디텍티드 포인트를 통해서 카운팅)
#if (Tracking_Xp < DetectingZone[0]) or Tracking_Xp > (DetectingZone[2]) or Tracking_Yp < (DetectingZone[1]-10) or Tracking_Yp > (DetectingZone[3]+10)or \
#(Matched is True and (Matched_Xp < DetectingZone[0]) or Matched_Xp > (DetectingZone[2]) or Matched_Yp < (DetectingZone[1]-10) or Matched_Yp > (DetectingZone[3]+10)):
#매칭이 트루이고, 매칭된 디텍티드 포인트가 영역밖에 존재하는 경우 - 삭제
if (Matched == True):
initBB_xy = (initBB[0] + int(initBB[2] / 2),
initBB[1] + initBB[3])
Matched_xy = (Matched_Xp, Matched_Yp)
RED_line_start_xy = (vertices[0][0][0], vertices[0][0][1])
RED_line_end_xy = (vertices[0][3][0], vertices[0][3][1])
BLUE_line_start_xy = (vertices[0][1][0], vertices[0][1][1])
BLUE_line_end_xy = (vertices[0][2][0], vertices[0][2][1])
#tracker에서 매칭된 디텍티드 포인트로 변경하여 주석처리
global count
if intersect(initBB_xy, Matched_xy, RED_line_start_xy,
RED_line_end_xy):
RED_cnt = RED_cnt + 1
if inout == 1: # inout을 구분하여 전역변수 변경
count = count + 1
else:
count = count - 1
car_cnt += 1
# initBB,lastBB, tracker 초기화
cv2.line(frame, (initBB[0] + int(initBB[2] / 2),
initBB[1] + initBB[3]),
(Matched_Xp, Matched_Yp), COLOR_RED, 2)
initBB = None
tracker = cv2.TrackerCSRT_create()
if intersect(initBB_xy, Matched_xy, BLUE_line_start_xy,
BLUE_line_end_xy):
BLUE_cnt = BLUE_cnt + 1
car_cnt += 1
cv2.line(frame, (initBB[0] + int(initBB[2] / 2),
initBB[1] + initBB[3]),
(Matched_Xp, Matched_Yp), COLOR_RED, 2)
# initBB,lastBB, tracker 초기화
initBB = None
tracker = cv2.TrackerCSRT_create()
return tracker, initBB, RED_cnt, BLUE_cnt
cent2 = get_centroid(bbox2)
draw_circle(frame, cent)
draw_circle(frame, cent2) # circle, outside this movements will happend
BB = bbox
BB2 = bbox2 # saving it for later
cv2.destroyAllWindows()
# fvs = FileVideoStream(path_vid).start() #vs Video Stream
bbox = BB
bbox2 = BB2
# Creating the CSRT Tracker
tracker = cv2.TrackerCSRT_create()
tracker2 = cv2.TrackerCSRT_create() # left hand for steering
# Initialize tracker with first frame and bounding box
tracker.init(FRAME, bbox)
tracker2.init(FRAME, bbox2)
cv2.putText(frame.copy(), 'Put both your hands in Postion', (100,70), \
cv2.FONT_HERSHEY_SIMPLEX, 0.75, COLOR_BLACK, 2)
TIMER_SETUP = 8
t = time.time()
while True:
frame = get_frame()
curr = (time.time() - t)
import os
import cv2
from vidstab import VidStab, layer_overlay, download_ostrich_video
# Download test video to stabilize
if not os.path.isfile("ostrich.mp4"):
download_ostrich_video("ostrich.mp4")
# Initialize object tracker, stabilizer, and video reader
object_tracker = cv2.TrackerCSRT_create()
stabilizer = VidStab()
vidcap = cv2.VideoCapture("ostrich.mp4")
# Initialize bounding box for drawing rectangle around tracked object
object_bounding_box = None
cap = cv2.VideoCapture(0)
while True:
# grabbed_frame, frame = vidcap.read()
grabbed_frame, frame = cap.read()
# Pass frame to stabilizer even if frame is None
stabilized_frame = stabilizer.stabilize_frame(input_frame=frame,
border_size=50)
# If stabilized_frame is None then there are no frames left to process
if stabilized_frame is None:
break
# Draw rectangle around tracked object if tracking has started
if object_bounding_box is not None:
def object_tracker(tracker_algo="MDF", video_path="road.mp4"):
if tracker_algo == "boosting":
tracker = cv2.TrackerBoosting_create()
if tracker_algo == "CSRT":
tracker = cv2.TrackerCSRT_create()
if tracker_algo == "TLD":
tracker = cv2.TrackerTLD_create()
if tracker_algo == "MIL":
tracker = cv2.TrackerMIL_create()
if tracker_algo == "KCF":
tracker = cv2.TrackerKCF_create()
if tracker_algo == "MDF":
tracker = cv2.TrackerMedianFlow_create()
time.sleep(1)
cap = cv2.VideoCapture(video_path)
ret, frame = cap.read()
width = cap.get(3) # float
height = cap.get(4) # float
bbox = cv2.selectROI("Tracking", frame, False)
tracker.init(frame, bbox)
def drawBox(img, bbox):
x, y, w, h = int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])
cv2.rectangle(img, (x, y), ((x + w), (y + h)), (0, 255, 0), 3, 3)
cv2.line(img, (0, int(height) - 5), (int(width), int(height) - 5),
(255, 0, 0), 3)
print(((x + x + w) / 2), ((y + y + h) / 2))
cv2.putText(img, "Tracking Started", (100, 75),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)
while True:
timer = cv2.getTickCount()
success, img = cap.read()
success, bbox = tracker.update(img)
if success:
drawBox(img, bbox)
else:
cv2.putText(img, "Lost", (100, 75), cv2.FONT_HERSHEY_SIMPLEX, 0.7,
(0, 0, 255), 2)
cv2.rectangle(img, (15, 15), (200, 90), (255, 0, 255), 2)
cv2.putText(img, "Fps:", (20, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.7,
(255, 0, 255), 2)
cv2.putText(img, "Status:", (20, 75), cv2.FONT_HERSHEY_SIMPLEX, 0.7,
(255, 0, 255), 2)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
if fps > 60: myColor = (20, 230, 20)
elif fps > 20: myColor = (230, 20, 20)
else: myColor = (20, 20, 230)
cv2.putText(img, str(int(fps)), (75, 40), cv2.FONT_HERSHEY_SIMPLEX,
0.7, myColor, 2)
cv2.imshow("Tracking", img)
if cv2.waitKey(1) & 0xff == ord('q'):
break
cv2.destroyAllWindows()
cap.release()
height = endY - startY
initBB = (startX, startY, width, height)
#print(initBB)
bboxes.append(initBB)
found = 1
totalFrames += 1
#tracker.init(frame, initBB)
#(x, y, w, h) = [int(v) for v in box]
#cv2.rectangle(frame, (x, y), (x + w, y + h),(0, 255, 0), 2)
#fps = FPS().start()
#totalFrames+=1
#found = 1
for bbox in bboxes:
multiTracker.add(cv2.TrackerCSRT_create(), frame, bbox)
retval = multiTracker.getObjects()
#print(type(bbox))
#print(retval)
#(x, y, w, h) = [int(v) for v in bbox]
#cv2.rectangle(frame, (x, y), (x + w, y + h),(0, 255, 0), 2)
cv2.imshow("MultiTracker", frame)
fps.update()
anobox = []
dellist = []
while True:
frame = vs.read()
frame = frame[1] if args.get("input", False) else frame
print('finding ROI: ', not tracking)
cv2.namedWindow('image')
cv2.setMouseCallback('image', onMouse)
cv2.imshow('image', img)
# detector = cv2.SimpleBlobDetector_create()
#tracker = cv2.TrackerBoosting_create()
#tracker = cv2.TrackerMIL_create()
#tracker = cv2.TrackerKCF_create()
#tracker = cv2.TrackerTLD_create()
#tracker = cv2.TrackerMedianFlow_create()
tracker = cv2.TrackerCSRT_create(
) # ensures enlarging and localization of the selected region and improved tracking of the non-rectangular regions or objects. It uses only 2 standard features (HoGs and Colornames)
#tracker = cv2.TrackerMOSSE_create() # good accuracy; loses with quick movements
tracker.save('tracker_params.json')
#fs = cv2.FileStorage("tracker_params.json", cv2.FileStorage_READ)
#tracker.read(fs.getFirstTopLevelNode())
display_boxes = []
display_scores = []
counter = 0
display_box = None
while True:
success, img = cap.read()
#print('{} tracking: {}'.format(counter, tracking))
def __createTracker(self):
tracker = None
if self.methodName == '' or self.methodName == 'CASCADE' or self.methodName == None:
pass
elif self.methodName == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
''' PyImgSearch
Based on the same algorithm used to power the machine learning behind Haar cascades (AdaBoost),
but like Haar cascades, is over a decade old. This tracker is slow and doesn’t work very well.
Interesting only for legacy reasons and comparing other algorithms. (minimum OpenCV 3.0.0)
'''
''' Paulo
Extremamente lento
'''
elif self.methodName == 'MIL':
tracker = cv2.TrackerMIL_create()
''' PyImgSearch
Better accuracy than BOOSTING tracker but does a poor job of reporting failure. (minimum OpenCV 3.0.0)
'''
elif self.methodName == 'KCF':
tracker = cv2.TrackerKCF_create()
''' PyImgSearch
Kernelized Correlation Filters. Faster than BOOSTING and MIL.
Similar to MIL and KCF, does not handle full occlusion well. (minimum OpenCV 3.1.0)
'''
elif self.methodName == 'TLD':
tracker = cv2.TrackerTLD_create()
''' PyImgSearch
I’m not sure if there is a problem with the OpenCV implementation of the TLD tracker or the actual algorithm itself,
but the TLD tracker was incredibly prone to false-positives. I do not recommend using this OpenCV object tracker. (minimum OpenCV 3.0.0)
'''
elif self.methodName == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
''' PyImgSearch
Does a nice job reporting failures; however, if there is too large of a jump in motion,
such as fast moving objects, or objects that change quickly in their appearance, the model will fail. (minimum OpenCV 3.0.0)
'''
''' Paulo
É bem rápido, responde a mudanças de escala, reporta oclusão.
A principio é o melhor algoritmo que temos.
'''
elif self.methodName == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
''' PyImgSearch
The only deep learning-based object detector included in OpenCV.
It requires additional model files to run (will not be covered in this post).
My initial experiments showed it was a bit of a pain to use even though it reportedly handles viewing changes well
(my initial experiments didn’t confirm this though).
I’ll try to cover it in a future post, but in the meantime, take a look at Satya’s writeup. (minimum OpenCV 3.2.0)
'''
''' Paulo
Testado com o reconhecimento de face.
Não apresentou um funcionamento muito bom, devido a constante reinicialização do rastreador.
Também não reportou corretamente oclusão do objeto.
Performance bem fraca.
Seria interessante testar com o cone.
Caso desejem fazer isso baixem os arquivos nesse link: https://www.dropbox.com/sh/77frbrkmf9ojfm6/AACgY7-wSfj-LIyYcOgUSZ0Ua?dl=0
e extraiam eles no diretório do main.py
'''
elif self.methodName == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
''' PyImgSearch
Very, very fast. Not as accurate as CSRT or KCF but a good choice if you need pure speed. (minimum OpenCV 3.4.1)
'''
''' Paulo
Realmente muito rápido e extremamente estável na detecção,
mas não responde a mudança no tamanho dos objetos que está rastreando.
Ou seja, não seriamos capazes de verificar a aproximação do objeto.
'''
elif self.methodName == 'CSRT':
tracker = cv2.TrackerCSRT_create()
''' PyImgSearch
Discriminative Correlation Filter (with Channel and Spatial Reliability).
Tends to be more accurate than KCF but slightly slower. (minimum OpenCV 3.4.2)
'''
''' Paulo:
Responde muito bem a escola, é relativamente estável, tem uma precisão muito boa.
Performance ruim para o Pi, e não reporta oclusão.
'''
else:
print("Algortimo '" + self.methodName + "' não reconhecido")
self.methodName = ""
return tracker
def tracking_face(img_path="./data_try/0__0", t_type=0):
# Set up tracker.
# Instead of MIL, you can also use
(major_ver, minor_ver, subminor_ver) = (cv2.__version__).split('.')
tracker_types = [
'BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'MOSSE',
'CSRT'
]
tracker_type = tracker_types[t_type]
global tracker
# if int(minor_ver) < 3:
if int(major_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
if tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
if tracker_type == "CSRT":
tracker = cv2.TrackerCSRT_create()
images_list = [img for img in os.listdir(img_path) if img.endswith(".png")]
images_list = sort_img(images_list)
img_txt_path = img_path.replace(img_path.split("/")[0], "./txt")
# txt_path = img_txt_path+"_txt"
txt_path = img_txt_path
if not os.path.isdir(txt_path):
os.makedirs(txt_path)
frame = cv2.imread(os.path.join(img_path, images_list[0]))
# Define an initial bounding box
height, width, layers = frame.shape
for i in range(len(images_list)):
frame = cv2.imread(os.path.join(img_path, images_list[i]))
bbox, score = bbox_score(frame)
bbox = np.squeeze(bbox)[0]
score = np.squeeze(score)[0]
# write_txt(txt_path,images_list[i])
if score <= 0.7:
bbox = bbox_transfer(bbox, height, width)
write_txt(txt_path, images_list[i], bbox, "bad")
else:
bbox = bbox_transfer(bbox, height, width) #(269,47,62,80)
# bbox = (269,47,62,80)
tracker.init(frame, bbox)
while i < len(images_list):
frame = cv2.imread(os.path.join(img_path, images_list[i]))
timer = cv2.getTickCount()
# Update tracker
ok, bbox = tracker.update(frame)
bbox_w = bbox_int(bbox)
write_txt(txt_path, images_list[i], bbox_w)
# Calculate Frames per second (FPS)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
# Draw bounding box
if ok:
# Tracking success
p1 = (int(bbox[0]), int(bbox[1]))
p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
cv2.rectangle(frame, p1, p2, (255, 0, 0), 2, 1)
i = i + 1
else:
# Tracking failure
cv2.putText(frame, "Tracking failure, re-initialize ...",
(100, 80), cv2.FONT_HERSHEY_SIMPLEX, 0.75,
(0, 0, 255), 2)
# Display tracker type on frame
cv2.putText(frame, tracker_type + " Tracker", (100, 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
# Display FPS on frame
cv2.putText(frame, "FPS : " + str(int(fps)), (100, 50),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
# Display result
cv2.imshow("Tracking", frame)
# Exit if ESC pressed
k = cv2.waitKey(1) & 0xff
if k == 27: break
break
def tracking_video_rectangle_tovideo(root, name, ad_name, point, result = 'cool_project.avi', edge = 20, save_img = False, save_result = True, method_num = 5, save_img2 = True, middle_halt = -1):
# Set up tracker.
# Instead of MIL, you can also use
tracker_types = ['BOOSTING', 'MIL','KCF', 'TLD', 'MEDIANFLOW', 'CSRT', 'MOSSE']
tracker_type = tracker_types[method_num]
ad = cv2.imread(join(root,ad_name))
if int(minor_ver) >= 3:
tracker = [cv2.Tracker_create(tracker_type) for _ in range(4)]
else:
if tracker_type == 'BOOSTING':
tracker = [cv2.TrackerBoosting_create() for _ in range(4)]
if tracker_type == 'MIL':
tracker = [cv2.TrackerMIL_create() for _ in range(4)]
if tracker_type == 'KCF':
tracker = [cv2.TrackerKCF_create() for _ in range(4)]
if tracker_type == 'TLD':
tracker = [cv2.TrackerTLD_create() for _ in range(4)]
if tracker_type == 'MEDIANFLOW':
tracker = [cv2.TrackerMedianFlow_create() for _ in range(4)]
if tracker_type == 'CSRT':
tracker = [cv2.TrackerCSRT_create() for _ in range(4)]
if tracker_type == 'MOSSE':
tracker = [cv2.TrackerMOSSE_create() for _ in range(4)]
# Read video
video = cv2.VideoCapture(join(root, name))
# Exit if video not opened.
if not video.isOpened():
print("Could not open video")
sys.exit()
# Read first frame.
ok, frame = video.read()
if not ok:
print('Cannot read video file')
sys.exit()
cv2.imwrite("result/%05d.jpg"%(0),frame)
out = cv2.VideoWriter(result , cv2.VideoWriter_fourcc(*'DIVX'), 10, (frame.shape[1],frame.shape[0]))
# Define an initial bounding box
bbox = [(point[_][0] - edge, point[_][1] - edge, edge*2 , edge*2) for _ in range(4)]
# Uncomment the line below to select a different bounding box
# bbox = cv2.selectROI(frame, False)
# Initialize tracker with first frame and bounding box
for _ in range(4):
ok = tracker[_].init(frame, bbox[_])
index_ = 0
while True:
# Read a new frame
index_ += 1
ok, frame = video.read()
if not ok:
break
# Start timer
timer = cv2.getTickCount()
ans = []
for _ in range(4):
# Update tracker
ok, bbox = tracker[_].update(frame)
# Calculate Frames per second (FPS)
# fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer);
# Draw bounding box
if ok:
# Tracking success
p1 = (int(bbox[0]), int(bbox[1]))
p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
p3 = (int(bbox[0] + bbox[2]/2), int(bbox[1] + bbox[3]/2))
ans.append(p3)
if (save_img or save_img2):
cv2.rectangle(frame, p1, p2, (255,0,0), 2, 1)
else :
# Tracking failure
if (save_img or save_img2):
cv2.putText(frame, "Tracking failure detected", (100,80), cv2.FONT_HERSHEY_SIMPLEX, 0.75,(0,0,255),2)
if (save_img):
# Display tracker type on frame
cv2.putText(frame, tracker_type + " Tracker", (100,20), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50,170,50),2);
# Display FPS on frame
# cv2.putText(frame, "FPS : " + str(int(fps)), (100,50), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50,170,50), 2);
# Display result
cv2.imwrite("result/%05d_%d.jpg"%(index_, _) , frame)
if (save_img2):
cv2.imwrite("result/%05d_.jpg"%(index_), frame)
if (save_result):
frame = Trans_forward(frame, np.array(ans , dtype = np.float32), ad)
cv2.imwrite("result2/%05d_.jpg"%(index_), frame)
out.write(frame)
# Exit if ESC pressed
k = cv2.waitKey(1) & 0xff
if k == 27 : break
if index_ == middle_halt : break
out.release()
return np.array(ans)
def detect_video(yolo: YOLO, video_path: str, output_path: str = ""):
vid = cv2.VideoCapture(video_path)
if not vid.isOpened():
raise IOError("Couldn't open webcam or video")
video_FourCC = int(vid.get(cv2.CAP_PROP_FOURCC))
video_fps = vid.get(cv2.CAP_PROP_FPS)
video_size = (int(vid.get(cv2.CAP_PROP_FRAME_WIDTH)),
int(vid.get(cv2.CAP_PROP_FRAME_HEIGHT)))
isOutput = True if output_path != "" else False
if isOutput:
print("!!! TYPE:", type(output_path), type(video_FourCC),
type(video_fps), type(video_size))
out = cv2.VideoWriter(output_path, video_FourCC, video_fps, video_size)
accum_time = 0
curr_fps = 0
fps = "FPS: ??"
prev_time = timer()
detected = False
trackers = []
font = ImageFont.truetype(font='font/FiraMono-Medium.otf', size=30)
thickness = 1
frame_count = 0
while True:
return_value, frame = vid.read()
image = Image.fromarray(frame)
image_data = np.array(image) / 255.
draw = ImageDraw.Draw(image)
if detected:
for tracker, predicted_class in trackers:
success, box = tracker.update(frame)
left, top, width, height = box
right = left + width
bottom = top + height
label = '{}'.format(predicted_class)
label_size = draw.textsize(label, font)
if top - label_size[1] >= 0:
text_origin = np.array([left, top - label_size[1]])
else:
text_origin = np.array([left, top + 1])
# My kingdom for a good redistributable image drawing library.
for i in range(thickness):
draw.rectangle([left + i, top + i, right - i, bottom - i],
outline=yolo.colors[c])
draw.rectangle(
[tuple(text_origin),
tuple(text_origin + label_size)],
fill=yolo.colors[c])
draw.text(text_origin, label, fill=(0, 0, 0), font=font)
frame_count += 1
if frame_count == 100:
for tracker in trackers:
del tracker
trackers = []
frame_count = 0
detected = False
else:
if tf.executing_eagerly():
boxes, scores, classes = yolo.detect_image(image_data, False)
else:
boxes, scores, classes = yolo.detect_image(image, False)
for i, c in enumerate(classes):
predicted_class = yolo.class_names[c]
top, left, bottom, right = boxes[i]
height = abs(bottom - top)
width = abs(right - left)
tracker = cv2.TrackerCSRT_create()
#tracker = cv2.TrackerKCF_create()
#tracker = cv2.TrackerMOSSE_create()
tracker.init(frame, (left, top, width, height))
trackers.append([tracker, predicted_class])
label = '{}'.format(predicted_class)
label_size = draw.textsize(label, font)
if top - label_size[1] >= 0:
text_origin = np.array([left, top - label_size[1]])
else:
text_origin = np.array([left, top + 1])
# My kingdom for a good redistributable image drawing library.
for i in range(thickness):
draw.rectangle([left + i, top + i, right - i, bottom - i],
outline=yolo.colors[c])
draw.rectangle(
[tuple(text_origin),
tuple(text_origin + label_size)],
fill=yolo.colors[c])
draw.text(text_origin, label, fill=(0, 0, 0), font=font)
detected = True
del draw
result = np.asarray(image)
curr_time = timer()
exec_time = curr_time - prev_time
prev_time = curr_time
accum_time = accum_time + exec_time
curr_fps = curr_fps + 1
if accum_time > 1:
accum_time = accum_time - 1
fps = "FPS: " + str(curr_fps)
curr_fps = 0
cv2.putText(result,
text=fps,
org=(3, 15),
fontFace=cv2.FONT_HERSHEY_SIMPLEX,
fontScale=0.50,
color=(255, 0, 0),
thickness=2)
cv2.namedWindow("result", cv2.WINDOW_NORMAL)
cv2.imshow("result", result)
if isOutput:
out.write(result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
yolo.close_session()
import cv2
# cap = cv2.VideoCapture(0) # for cam
cap = cv2.VideoCapture('vtest.avi') # for video
while True:
success, img = cap.read()
cv2.putText(img, "When You Ready Press 's' ", (100, 100),
cv2.FONT_HERSHEY_COMPLEX_SMALL, 0.7, (255, 23, 100), 2)
cv2.imshow('Tracking', img)
if cv2.waitKey(80) & 0xFF == ord('s'):
break
# install opencv-contrib-python
# Trackers
# tracker = cv2.TrackerMOSSE_create() # high speed & low accuracy
tracker = cv2.TrackerCSRT_create() # low speed & high accuracy
# tracker = cv2.TrackerGOTURN_create() # some error
# tracker = cv2.TrackerMedianFlow_create() # working accurate
# tracker = cv2.TrackerTLD_create() #working
# tracker = cv2.TrackerKCF_create() #working
# tracker = cv2.TrackerMIL_create()
# tracker = cv2.TrackerBoosting_create()
success, img = cap.read()
bbox = cv2.selectROI('Tracking', img, True)
tracker.init(img, bbox)
def drawBox(img, bbox):
x, y, w, h = int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])
cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 255), 3, 1)
cv2.putText(img, 'Tracking', (75, 75), cv2.FONT_HERSHEY_SIMPLEX, 0.7,
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
'--classification_model',
help='Path of classification model.',
required=False,
default='all_models/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite'
)
parser.add_argument(
'--detection_model',
help='Path of detection model.',
required=False,
default=
'all_models/ssd_mobilenet_v2_coco_quant_postprocess_edgetpu.tflite')
parser.add_argument('--image', help='Path of the image.', required=False)
parser.add_argument('--classification_labels',
required=False,
default='all_models/inat_bird_labels.txt')
parser.add_argument('--detection_labels',
required=False,
default='all_models/coco_labels.txt')
args = parser.parse_args()
# initialize the video stream and allow the camera sensor to warmup
print("[INFO] starting video stream...")
vs = VideoStream(src=0, resolution=(2048, 1536)).start()
#vs = VideoStream(usePiCamera=False).start()
time.sleep(2.0)
detection_model = DetectionEngine(args.detection_model)
classification_model = ClassificationEngine(args.classification_model)
detection_labels = load_labels(args.detection_labels)
print("detection_labels : {}".format(len(detection_labels)))
classification_labels = load_labels(args.classification_labels)
multiTracker = cv2.MultiTracker_create()
tracking_mode = False
tracking_expire = None
# loop over the frames from the video stream
while True:
# grab the frame from the threaded video stream and resize it
# to have a maximum width of 500 pixels
frame = vs.read()
#resized_frame = imutils.resize(frame, width=500)
resized_frame = frame
orig = resized_frame.copy()
# prepare the frame for classification by converting (1) it from
# BGR to RGB channel ordering and then (2) from a NumPy array to
# PIL image format
resized_frame = cv2.cvtColor(resized_frame, cv2.COLOR_BGR2RGB)
resized_frame = Image.fromarray(resized_frame)
# make predictions on the input frame
start = time.time()
success, boxes = multiTracker.update(orig)
if tracking_expire and time.time() > tracking_expire:
tracking_mode = False
for tracker in multiTracker.getObjects():
tracker.clear()
multiTracker = cv2.MultiTracker_create()
print('success {}'.format(success))
print('boxes {}'.format(boxes))
if success:
for box in boxes:
(x, y, w, h) = [int(v) for v in box]
cv2.rectangle(orig, (x, y), (x + w, y + h), (0, 0, 255), 2)
text = "{}: {:.2f}% ({:.4f} sec)".format(
"bird", score * 100, end - start)
cv2.putText(orig, text, (x, y), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(0, 0, 255), 2)
objs = detection_model.detect_with_image(resized_frame, top_k=1)
end = time.time()
for obj in objs:
# draw the predicted class label, probability, and inference
# time on the output frame
score = obj.score
box = obj.bounding_box
height, width, channels = orig.shape
label = detection_labels[obj.label_id]
if label == "bird":
p0, p1 = list(box)
x0, y0 = list(p0)
x1, y1 = list(p1)
x0, y0, x1, y1 = int(x0 * width), int(y0 * height), int(
x1 * width), int(y1 * height)
cv2.rectangle(orig, (x0, y0), (x1, y1), (0, 255, 0), 2)
text = "{}: {:.2f}% ({:.4f} sec)".format(
"bird", score * 100, end - start)
cv2.putText(orig, text, (x0, y0), cv2.FONT_HERSHEY_SIMPLEX,
0.5, (0, 255, 0), 2)
if score > 0.2:
#im = Image.new('RGB', (x1-x0, y1-y0))
#im.putdata(frame[y0:y1,x0:x1])
#print("raw {}".format(frame[y0:y1,x0:x1]))
#classification_thread = threading.Thread(target=classification_job,args=(classification_model, frame[y0:y1,x0:x1], 1))
#classification_thread.start()
#classification_thread.join()
is_intersection = False
for box in boxes:
(x, y, w, h) = [int(v) for v in box]
if bb_intersection_over_union(
[x0, y0, x1, y1], [x, y, x + w, y + h]) > 0:
is_intersection = True
print("intersect.. already tracking")
if not is_intersection:
tracking_expire = time.time() + 60
tracker = cv2.TrackerCSRT_create()
print("add tracker {} {} {} {}".format(
x0, y0, width, height))
multiTracker.add(tracker, orig,
(x0, y0, width / 2, height / 2))
# show the output frame and wait for a key press
cv2.namedWindow("Frame", cv2.WINDOW_NORMAL)
cv2.resizeWindow("Frame", 800, 600)
cv2.imshow("Frame", orig)
key = cv2.waitKey(1) & 0xFF
# if the `q` key was pressed, break from the loop
if key == ord("q"):
break
# do a bit of cleanup
cv2.destroyAllWindows()
vs.stop()
def get_points(filename, tracker):
# Set up tracker.
# Instead of MIL, you can also use
tracker_types = ['BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'MOSSE', 'CSRT']
tracker_type = tracker_types[tracker]
if int(minor_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
if tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
if tracker_type == "CSRT":
tracker = cv2.TrackerCSRT_create()
# Read video
path = '/home/ilkka/Python/trajectory/videos/' + filename
video = cv2.VideoCapture(path)
# Exit if video not opened.
if not video.isOpened():
print('Could not open video')
sys.exit()
# Read first frame.
ok, frame = video.read()
if not ok:
print('Cannot read video file')
sys.exit()
# Define an initial bounding box
bbox = (287, 23, 86, 320)
# Uncomment the line below to select a different bounding box
bbox = cv2.selectROI(frame, False)
# Initialize tracker with first frame and bounding box
ok = tracker.init(frame, bbox)
# The list of points
points = np.array([[0, 0]])
while True:
# Read a new frame
ok, frame = video.read()
if not ok:
break
# Start timer
timer = cv2.getTickCount()
# Update tracker
ok, bbox = tracker.update(frame)
# Calculate Frames per second (FPS)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer);
# Draw bounding box
if ok:
# Tracking success
p1 = (int(bbox[0]), int(bbox[1]))
p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
cv2.rectangle(frame, p1, p2, (0,0,255), 2, 1)
# Center of bbox
x = (2*bbox[0] + bbox[2])/2
y = (2*bbox[1] + bbox[3])/2
# Add (x, y) to the list of points
points = np.append(points, [[x, y]], axis=0)
# Add a mark in the center of bbox
q1 = (int(x)-1, int(y)-1)
q2 = (int(x)+1, int(y)+1)
cv2.rectangle(frame, q1, q2, (0,0,255), 2, 1)
else:
# Tracking failure
cv2.putText(frame, "Tracking failure detected", (100,80),
cv2.FONT_HERSHEY_SIMPLEX, 0.75,(0,0,255),2)
# Display tracker type on frame
cv2.putText(frame, tracker_type + " Tracker", (100,20),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50,170,50),2);
# Display FPS on frame
cv2.putText(frame, "FPS : " + str(int(fps)), (100,50),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50,170,50), 2);
# Display result
cv2.imshow("Tracking", frame)
# Exit if ESC pressed
k = cv2.waitKey(1) & 0xff
if k == 27 : break
# Remove the initial zero row from points
points = points[1:len(points)]
return points
def __init__(self):
self.position_arr = []
self.position_dq = deque(maxlen=16)
self.roi_tracker = cv2.TrackerCSRT_create()
self.init_bb = None
def main(args):
input_video_paths = glob(
path.join(args.input_video_dir, '*.' + args.video_extension))
for input_video_path in input_video_paths:
print(input_video_path)
cam_name = '_'.join(
path.basename(input_video_path).split('.')[0].split('_')[:2])
with open(
path.join(args.input_bbox_dir,
path.basename(input_video_path) + '.pkl'),
'rb') as f:
bboxes = pkl.load(f)
with open(path.join(args.input_roi_dir, cam_name + '.txt')) as f:
roi_coords = [[int(coord) for coord in line.split(',')]
for line in f.read().split('\n')[:-1]]
roi = np.load(path.join(args.input_roi_dir, cam_name + '.npy'))
input_video = cv2.VideoCapture(input_video_path)
print(input_video.get(cv2.CAP_PROP_FPS))
continue
width = int(input_video.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(input_video.get(cv2.CAP_PROP_FRAME_HEIGHT))
output_video = cv2.VideoWriter(filename=path.join(
args.output_video_dir, path.basename(input_video_path)),
fourcc=cv2.VideoWriter_fourcc(*'mp4v'),
fps=30.,
frameSize=(width, height),
isColor=True)
num_frames = int(input_video.get(cv2.CAP_PROP_FRAME_COUNT))
trackers = []
tracked_bboxes = []
bbox_ids = []
current_bbox_id = 0
for frame_idx in tqdm(range(num_frames)):
success, frame = input_video.read()
# Keep cars and trucks
frame_bboxes = np.concatenate(
[bboxes[frame_idx][1], bboxes[frame_idx][2]], axis=0)
# Keep bboxes with confidence score more than threshold
frame_bboxes = [
bbox[:4].astype(np.int32) for bbox in frame_bboxes
if bbox[4] > args.confidence_thresh
]
# Remove bboxes that cannot be tracked or exists over a threshold
untracked_ids = []
for i, bbox_id in enumerate(bbox_ids):
success, bbox = trackers[i].update(frame)
bbox = [bbox[0], bbox[1], bbox[2] + bbox[0], bbox[3] + bbox[1]]
if success and verify_bbox(roi, bbox, tracked_bboxes[i],
args.dist_thresh, args.time_thresh,
bbox_id):
tracked_bboxes[i].append(np.array(bbox))
else:
untracked_ids.append(i)
if len(untracked_ids) > 0:
for index in untracked_ids[::-1]:
del tracked_bboxes[index]
del trackers[index]
del bbox_ids[index]
if len(frame_bboxes) > 0 and len(tracked_bboxes) > 0:
latest_bboxes = [
tracked_car[-1] for tracked_car in tracked_bboxes
]
ious = mask_util.iou(
np.array(frame_bboxes), np.array(latest_bboxes),
np.zeros((len(latest_bboxes), ), dtype=np.bool))
elif len(frame_bboxes) > 0:
ious = np.zeros((len(frame_bboxes), 1))
max_iou_per_new = np.asarray(ious).max(axis=1).tolist()
arg_max_iou_per_new = np.asarray(ious).argmax(axis=1).tolist()
for iou, arg, xyxy in zip(max_iou_per_new, arg_max_iou_per_new,
frame_bboxes):
if iou <= args.iou_lower_thresh:
if verify_bbox(roi, xyxy):
tracked_bboxes.append([xyxy])
bbox_ids.append(current_bbox_id)
trackers.append(cv2.TrackerCSRT_create())
xywh = (xyxy[0], xyxy[1], xyxy[2] - xyxy[0],
xyxy[3] - xyxy[1])
trackers[-1].init(frame, xywh)
current_bbox_id += 1
elif iou >= args.iou_upper_thresh:
tracked_bboxes[arg][-1] = xyxy
trackers[arg] = cv2.TrackerCSRT_create()
xywh = (xyxy[0], xyxy[1], xyxy[2] - xyxy[0],
xyxy[3] - xyxy[1])
trackers[arg].init(frame, xywh)
for tracked_seq, bbox_id in zip(tracked_bboxes, bbox_ids):
frame = draw_bbox(frame, tracked_seq[-1], bbox_id)
frame = draw_roi(frame, roi_coords)
output_video.write(frame)
def StartTracking(self,frame, object_box):
self.__tracker = cv2.TrackerCSRT_create()
self.__tracker.init(frame,object_box)
def __init__(self):
self.mtracker = cv2.TrackerCSRT_create() # CSRT, KCF,Boosting,MIL,TLD,MedianFlow,MOSSE
self.trackingBoundingBox = (0,0,0,0)
def tracking_face(vi_path="try.avi", t_type=7):
# Set up tracker.
# Instead of MIL, you can also use
(major_ver, minor_ver, subminor_ver) = (cv2.__version__).split('.')
tracker_types = [
'BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'MOSSE',
'CSRT'
]
tracker_type = tracker_types[t_type]
# if int(minor_ver) < 3:
if int(major_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
if tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
if tracker_type == "CSRT":
tracker = cv2.TrackerCSRT_create()
# Read video
video = cv2.VideoCapture(vi_path)
# Exit if video not opened.
if not video.isOpened():
print("Could not open video")
sys.exit()
# Read first frame.
ok, frame = video.read()
if not ok:
print('Cannot read video file')
sys.exit()
# Define an initial bounding box
height, width, layers = frame.shape
# box
# bbox_transfer()
bbox, score = bbox_score(frame)
bbox = np.squeeze(bbox)[0]
score = np.squeeze(score)[0]
while True:
if score > 0.7:
break
ok, frame = video.read()
bbox, score = bbox_score(frame)
bbox = np.squeeze(bbox)[0]
score = np.squeeze(score)[0]
bbox = bbox_transfer(bbox, height, width) #(269,47,62,80)
# bbox = (bbox[1]*width,bbox[0]*height,bbox[2]-bbox[0],bbox[3]-bbox[1])
# bbox = tuple(bbox)
# bbox = (265, 48, 70, 82)# x,y, w, h
# bbox = (48,151,128,186) # *H ymin, xmin, ymax, xmax
# bbox = (85,270,228,332) # *W
# bbox = (85,151,228,186) A - 151 85 (228-85) 186-151
# bbox = (48,270,128,332) B
# print(bbox.shape,score.shape)
# Uncomment the line below to select a different bounding box
# bbox = cv2.selectROI(frame, False)
# Initialize tracker with first frame and bounding box
ok = tracker.init(frame, bbox)
# bbox = bbox_transfer(bbox)
while True:
# Read a new frame
ok, frame = video.read(
) # I think this line should be put after tracker.update
if not ok:
break
# Start timer
timer = cv2.getTickCount()
# Update tracker
ok, bbox = tracker.update(frame) # bbox --> [x1,y1,width,height]
# Calculate Frames per second (FPS)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
# Draw bounding box
if ok:
# Tracking success
p1 = (int(bbox[0]), int(bbox[1])) #(x1,y1)
p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3])) #(x2,y2)
cv2.rectangle(frame, p1, p2, (255, 0, 0), 2, 1)
else:
# Tracking failure
cv2.putText(frame, "Tracking failure detected", (100, 80),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
# Display tracker type on frame
cv2.putText(frame, tracker_type + " Tracker", (100, 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
# Display FPS on frame
cv2.putText(frame, "FPS : " + str(int(fps)), (100, 50),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
# Display result
cv2.imshow("Tracking", frame)
# Exit if ESC pressed
k = cv2.waitKey(1) & 0xff
if k == 27: break
tracker_types = ['MIL', 'KCF', 'GOTURN', 'CSRT']
print(f"Select tracker type by index: {tracker_types}")
i = int(input(">> "))
tracker_type = tracker_types[i]
if int(minor_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
if tracker_type == "CSRT":
tracker = cv2.TrackerCSRT_create()
# Read video
video = cv2.VideoCapture("vid.avi")
# Exit if video not opened.
if not video.isOpened():
print("Could not open video")
sys.exit()
# Read first frame.
ok, frame = video.read()
if not ok:
print('Cannot read video file')
sys.exit()
def main():
tracker_types = [
'BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'MOSSE',
'CSRT'
]
tracker_type = tracker_types[2]
if int(minor_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
if tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
if tracker_type == "CSRT":
tracker = cv2.TrackerCSRT_create()
x0 = 200
y0 = 200
w0 = 224
h0 = 224
track_window = (x0, y0, w0, h0)
# Reference Distance
L0 = 100
S0 = 50176 #224x224 #take#here.
# Base Distance
LB = 100
# Define an initial bounding box
bbox = (x0, y0, w0, h0) #(287, 23, 86, 320)
#CX=int(bbox[0]+0.5*bbox[2]+3) #adding
#CY=int(bbox[1]+0.5*bbox[3]+3) #adding
drone = tellopy.Tello()
try:
drone.connect()
drone.wait_for_connection(60.0)
retry = 3
container = None
while container is None and 0 < retry:
retry -= 1
try:
container = av.open(drone.get_video_stream())
except av.AVError as ave:
print(ave)
print('retry...')
drone.takeoff()
# skip first 300 frames
frame_skip = 300
while True:
#------------------------------------------for start
for frame in container.decode(video=0):
if 0 < frame_skip:
frame_skip = frame_skip - 1
continue
start_time = time.time()
image = cv2.cvtColor(numpy.array(frame.to_image()),
cv2.COLOR_RGB2BGR)
# Start timer
timer = cv2.getTickCount()
#cv2.imshow('Canny', cv2.Canny(image, 100, 200))
#cv2.waitKey(1)
# Update tracker
ok, bbox = tracker.update(image)
# Calculate Frames per second (FPS)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
# Draw bounding box
if ok:
#print('Tracking ok')
(x, y, w, h) = (int(bbox[0]), int(bbox[1]), int(bbox[2]),
int(bbox[3]))
CX = int(bbox[0] + 0.5 * bbox[2]) #Center of X
CY = int(bbox[1] + 0.5 * bbox[3])
S0 = bbox[2] * bbox[3]
print("CX,CY,S0,x,y=", CX, CY, S0, x, y)
# Tracking success
p1 = (x, y)
p2 = (x + w, y + h)
cv2.rectangle(image, p1, p2, (255, 0, 0), 2, 1)
p10 = (x0, y0)
p20 = (x0 + w0, y0 + h0)
cv2.rectangle(image, p10, p20, (0, 255, 0), 2, 1)
d = round(L0 * m.sqrt(S0 / (w * h)))
dx = x + w / 2 - CX0 #no change dx
dy = y + h / 2 - CY0
print(d, dx, dy)
tracking(drone, d, dx, dy, LB)
else:
# Tracking failure
#print('Tracking failure')
cv2.putText(image, "Tracking failure detected", (100, 80),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
cv2.imshow('Original', image)
key = cv2.waitKey(1) & 0xff
if key == ord('q'):
print('Q!')
break
if key == ord('r'):
roi_time = time.time()
bbox = cv2.selectROI(image, False)
print(bbox)
(x0, y0, w0, h0) = (int(bbox[0]), int(bbox[1]),
int(bbox[2]), int(bbox[3]))
CX0 = int(x0 + 0.5 * w0) #Center of X
CY0 = int(y0 + 0.5 * h0)
# Initialize tracker with first frame and bounding box
ok = tracker.init(image, bbox)
'''
if frame.time_base < 1.0/60:
time_base = 1.0/60
else:
time_base = frame.time_base
frame_skip2 = int((time.time() - roi_time)/time_base)
if 0 < frame_skip2:
frame_skip2 = frame_skip2 - 1
continue
'''
if frame.time_base < 1.0 / 60:
time_base = 1.0 / 60
else:
time_base = frame.time_base
frame_skip = int((time.time() - start_time) / time_base)
#-------------------------------------------------for end
break
print('stop fly')
except Exception as ex:
exc_type, exc_value, exc_traceback = sys.exc_info()
traceback.print_exception(exc_type, exc_value, exc_traceback)
print(ex)
finally:
drone.quit()
drone.land()
cv2.destroyAllWindows()