from color_filtering_img import apply_color_filter
from pathlib import Path
import cv2
import numpy as np
from img_utils import DisplayUtils, GeneralUtils, ShapeDetector
shape_detector = ShapeDetector()
utils = GeneralUtils()
display_utils = DisplayUtils()
def apply_ops(frame):
"""DEPRECATED, used for testing appying various operations to the frame"""
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# cv2.imshow("gray before", gray)
# blurred = cv2.bilateralFilter(gray, 25, 15, 75)
blurred = cv2.GaussianBlur(gray, (5, 5), 0)
# _, thresh = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY)
thresh = cv2.adaptiveThreshold(blurred, 255,
cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 201, 0)
# thresh = cv2.bitwise_or(thresh_norm, thresh_adapt)
# cv2.imshow("blurred", blurred)
# cv2.imshow("thresh", thresh)
median = np.median(gray)
sigma = 0.33
lower_thresh = int(max(0, (1.0 - sigma) * median))
upper_thresh = int(min(255, (1.0 + sigma) * median))
from pathlib import Path
import cv2
import numpy as np
from img_utils import DisplayUtils, GeneralUtils, ShapeDetector
shape_detector = ShapeDetector()
utils = GeneralUtils()
display_utils = DisplayUtils()
def apply_color_filter(img, lower=[5, 0, 60], upper=[20, 180, 255]):
blurred = cv2.medianBlur(img, 5)
hsv = cv2.cvtColor(blurred, cv2.COLOR_BGR2HSV)
lower_brown = np.array(lower)
upper_brown = np.array(upper)
brown_mask = cv2.inRange(hsv, lower_brown, upper_brown)
brown_res = cv2.bitwise_and(img, img, mask=brown_mask)
brown_cleaned = cv2.morphologyEx(brown_res,
cv2.MORPH_OPEN, (5, 5),
iterations=1)
brown_blurred = cv2.GaussianBlur(brown_cleaned, (3, 3), 0)
### Temp Displaying ###
# grid = display_utils.create_img_grid(
# [
# [brown_res, brown_dilated, brown_blurred],
import target_detection_img, color_filtering_img from center_prediction import CenterPredModel # TODO ideas # find contours after each mask, then blacken everything under a certain area. # crop into brown target after using contours to find a rectangle that is large on brown mask # add threshold that reverts to previous known centriod if over it rather than predicting # calulate depth # try expanding roi on point_tracking then running goodFeaturesToTrack on expaned roi # accesible at drive folder: https://drive.google.com/drive/folders/11khSnQNsxnt0JStAec8j-widmBLOzPsO?usp=sharing # video name (can use others too): vision-video-trees-white-notape-lowres-0.mp4 utils = GeneralUtils() shape_detector = ShapeDetector() display_utils = DisplayUtils() # keep last 150 frames or last 5 seconds at 30fps centroids_deque = deque(maxlen=150) NO_TARGET_THRESH = 15 frames_since_target = 0 CONSTANT_CENTROID_THRESH = 6 FRAMES_FOR_TRAIN = 3 MODEL_DEGREE = 1 center_pred_model = CenterPredModel(degree=MODEL_DEGREE) frame_count = 0 truth_centroids = [] pred_centroids = []
# video_path = Path(
# "E:/code/projects/frc-vision/datasets/target-dataset/vision-videos/orange-teal-target-1080p.mp4"
# )
# video_path = Path(
# "E:/code/projects/frc-vision/datasets/target-dataset/vision-videos/vision-video-trees-white-notape-lowres-0.mp4"
# )
video_path = Path(
"E:/code/projects/frc-vision/datasets/target-dataset/vision-videos/distance-measuring/all_dists.mp4"
)
RESOLUTION_SCALE = 1
KNOWN_DEPTH_WEIGHT = 0.5
utils = GeneralUtils()
shape_detector = ShapeDetector()
display_utils = DisplayUtils()
depth_model = DepthPredModel()
depth_model.load_from_json(meta_path, pixel_to_dist_path)
depth_deque = Deque(maxlen=5)
# Parameters for lucas kanade optical flow
lk_params = dict(
winSize=(27, 27),
maxLevel=5,
criteria=(cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 1000, 1.01),
)
def get_hexagon_points(frame):
# equalize color and brightness, 2000+ fps
import cv2
import numpy as np
from img_utils import DisplayUtils, GeneralUtils, ShapeDetector
from depth_prediction import DepthPredModel
# zero for completely silent, 1 for just console logs, 2 for displaying frames
VERBOSE_LEVEL = 2
RESOLUTION_SCALE = 0.5
pixel_to_dist_path = Path().cwd() / "target_points.json"
meta_path = Path().cwd() / "meta.json"
utils = GeneralUtils()
shape_detector = ShapeDetector()
display_utils = DisplayUtils()
depth_model = DepthPredModel()
depth_model.load_from_json(meta_path, pixel_to_dist_path)
# Parameters for lucas kanade optical flow
lk_params = dict(
winSize=(25, 25),
maxLevel=2,
criteria=(cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 0.03),
)
def get_hexagon_points(frame):
cv2.normalize(frame, frame, 0, 255, cv2.NORM_MINMAX)