# Ultralytics YOLO 🚀, AGPL-3.0 license from collections import defaultdict import logging import cv2 import numpy as np from ultralytics.utils.checks import check_imshow, check_requirements from ultralytics.utils.plotting import Annotator, colors from shapely.geometry import LineString, Point, Polygon # create logger logger = logging.getLogger(__name__).addHandler(logging.NullHandler()) # need shapely>=2.0.0 check_requirements("shapely>=2.0.0") class ObjectCounter: """ A class to manage the counting of objects in a real-time video stream based on their tracks. """ def __init__(self): """ Initializes the Counter with default values for various tracking and counting parameters. """ # Mouse events self.is_drawing = False self.selected_point = None # Region & Line Information self.reg_pts = [(20, 400), (1260, 400)] self.line_dist_thresh = 15 self.counting_region = None self.region_color = (255, 0, 255) self.region_thickness = 5 # Image and annotation Information self.im0 = None self.tf = None self.view_img = False self.view_in_counts = True self.view_out_counts = True self.names = None # Classes names self.annotator = None # Annotator # Object counting Information self.in_counts = 0 self.out_counts = 0 self.out_counts_prev = self.out_counts self.in_counts_prev = self.in_counts self.counting_list = [] self.count_txt_thickness = 0 self.count_txt_color = (0, 0, 0) self.count_color = (255, 255, 255) # Tracks info self.track_history = defaultdict(list) self.track_thickness = 2 self.draw_tracks = False self.draw_boxes = False # added by steve.yin @ 3/1/2024 self.track_color = (0, 255, 0) # Check if environment support imshow self.env_check = check_imshow(warn=True) def set_args( self, classes_names, reg_pts, count_reg_color=(255, 0, 255), line_thickness=2, track_thickness=2, view_img=False, view_in_counts=True, view_out_counts=True, draw_tracks=False, draw_boxes=False, # added by steve.yin @ 3/1/2024 draw_reg_pts=True, # added by steve.yin @ 3/1/2024 count_txt_thickness=2, count_txt_color=(0, 0, 0), count_color=(255, 255, 255), track_color=(0, 255, 0), region_thickness=5, line_dist_thresh=15, ): """ Configures the Counter's image, bounding box line thickness, and counting region points. Args: line_thickness (int): Line thickness for bounding boxes. view_img (bool): Flag to control display the video stream. view_in_counts (bool): Flag to control display the incounts. view_out_counts (bool): Flag to control display the outcounts. reg_pts (list): Initial list of points for the counting region. classes_names (dict): Classes names track_thickness (int): Track thickness draw_tracks (Bool): draw tracks draw_boxes (Bool): draw boxes draw_reg_pts (Bool): draw reg_pts count_txt_thickness (int): Text thickness object counting display count_txt_color (RGB color): count text color value count_color (RGB color): count text background color value count_reg_color (RGB color): Color of object counting region track_color (RGB color): color for tracks region_thickness (int): Object counting Region thickness line_dist_thresh (int): Euclidean Distance threshold line counter """ self.tf = line_thickness self.view_img = view_img self.view_in_counts = view_in_counts self.view_out_counts = view_out_counts self.track_thickness = track_thickness self.draw_tracks = draw_tracks self.draw_boxes = draw_boxes # added by steve.yin @ 3/1/2024 self.draw_reg_pts = draw_reg_pts # added by steve.yin @ 3/1/2024 # Region and line selection if len(reg_pts) == 2: logging.info("Line Counter Initiated.") self.reg_pts = reg_pts self.counting_region = LineString(self.reg_pts) u = np.array([self.reg_pts[0][0], self.reg_pts[0][1]]) v = np.array([self.reg_pts[1][0], self.reg_pts[1][1]]) elif len(reg_pts) == 4: logging.info("Region Counter Initiated.") self.reg_pts = reg_pts self.counting_region = Polygon(self.reg_pts) u = np.array([ (self.reg_pts[0][0] + self.reg_pts[1][0]) / 2, (self.reg_pts[0][1] + self.reg_pts[1][1]) / 2, ]) v = np.array([ (self.reg_pts[2][0] + self.reg_pts[3][0]) / 2, (self.reg_pts[2][1] + self.reg_pts[3][1]) / 2, ]) else: logging.warning( "Invalid Region points, which can only be 2 or 4. " + "Using Line Counter Instead!" ) self.counting_region = LineString(self.reg_pts) u = np.array(self.counting_region.coords[0]) v = np.array( self.counting_region.coords[len(self.counting_region.coords)-1] ) # get line orientation, rotate ccw 90degrees, get line normal vector n = v - u nvec = np.array([-n[1], n[0]]) # print(f"v: {v}, u: {u}, n: {n}, nvec0: {nvec}") self.counting_region_nvec = nvec / (np.linalg.norm(nvec) + 1e-6) # print(f"nvec: {self.counting_region_nvec}") self.names = classes_names self.track_color = track_color self.count_txt_thickness = count_txt_thickness self.count_txt_color = count_txt_color self.count_color = count_color self.region_color = count_reg_color self.region_thickness = region_thickness self.line_dist_thresh = line_dist_thresh def mouse_event_for_region(self, event, x, y, flags, params): """ This function is designed to move region with mouse events in a real-time video stream. Args: event (int): The type of mouse event (e.g., cv2.EVENT_MOUSEMOVE, cv2.EVENT_LBUTTONDOWN, etc.). x (int): The x-coordinate of the mouse pointer. y (int): The y-coordinate of the mouse pointer. flags (int): Any flags associated with the event (e.g., cv2.EVENT_FLAG_CTRLKEY, cv2.EVENT_FLAG_SHIFTKEY, etc.). params (dict): Additional parameters passing to the function. """ if event == cv2.EVENT_LBUTTONDOWN: for i, point in enumerate(self.reg_pts): if ( isinstance(point, (tuple, list)) and len(point) >= 2 and (abs(x - point[0]) < 10 and abs(y - point[1]) < 10) ): self.selected_point = i self.is_drawing = True break elif event == cv2.EVENT_MOUSEMOVE: if self.is_drawing and self.selected_point is not None: self.reg_pts[self.selected_point] = (x, y) self.counting_region = Polygon(self.reg_pts) elif event == cv2.EVENT_LBUTTONUP: self.is_drawing = False self.selected_point = None def extract_and_process_tracks(self, tracks): """ Extracts and processes tracks for object counting in a video stream. """ boxes = tracks[0].boxes.xyxy.cpu() clss = tracks[0].boxes.cls.cpu().tolist() track_ids = tracks[0].boxes.id.int().cpu().tolist() # Annotator Init and region drawing self.annotator = Annotator(self.im0, self.tf, self.names) # self.annotator.draw_region( # reg_pts=self.reg_pts, # color=self.region_color, # thickness=self.region_thickness # ) # Extract tracks for box, track_id, cls in zip(boxes, track_ids, clss): # Draw bounding box [modified by steve.yin @ 3/1/2024] if self.draw_reg_pts: self.annotator.draw_region( reg_pts=self.reg_pts, color=self.region_color, thickness=self.region_thickness ) if self.draw_boxes: self.annotator.box_label( box=box, label=f"{track_id}:{self.names[cls]}", color=colors(int(cls), True) ) # Draw Tracks track_line = self.track_history[track_id] track_line.append(( float((box[0] + box[2]) / 2), float((box[1] + box[3]) / 2) )) if len(track_line) > 30: track_line.pop(0) # Draw track trails if self.draw_tracks: self.annotator.draw_centroid_and_tracks( track=track_line, color=self.track_color, track_thickness=self.track_thickness ) prev_position = self.track_history[track_id][0] \ if len(self.track_history[track_id]) > 1 else None # Count objects if len(self.reg_pts) == 4: if ( prev_position is not None and self.counting_region.contains(Point(track_line[-1])) and track_id not in self.counting_list ): self.counting_list.append(track_id) obj_track_vec = np.array([ track_line[-1][0] - prev_position[0], track_line[-1][1] - prev_position[1] ]) if np.sign( np.dot(obj_track_vec, self.counting_region_nvec) ) < 0: self.out_counts += 1 else: self.in_counts += 1 elif len(self.reg_pts) == 2: if prev_position is not None: distance = Point(track_line[-1]) \ .distance(self.counting_region) if ( distance < self.line_dist_thresh and track_id not in self.counting_list ): self.counting_list.append(track_id) obj_track_vec = np.array([ track_line[-1][0] - prev_position[0], track_line[-1][1] - prev_position[1] ]) logging.info(f"obj_track_vec: {obj_track_vec}") if np.sign( np.dot(obj_track_vec, self.counting_region_nvec) ) < 0: self.out_counts += 1 else: self.in_counts += 1 self.outcounts_updated() self.incounts_updated() self.out_counts_prev = self.out_counts self.in_counts_prev = self.in_counts incount_label = f"In: {self.in_counts}" outcount_label = f"Out: {self.out_counts}" # Display counts based on user choice counts_label = None if not self.view_in_counts and not self.view_out_counts: counts_label = None elif not self.view_in_counts: counts_label = outcount_label elif not self.view_out_counts: counts_label = incount_label else: counts_label = f"{incount_label} | {outcount_label}" if counts_label is not None: self.annotator.count_labels( counts=counts_label, count_txt_size=self.count_txt_thickness, txt_color=self.count_txt_color, color=self.count_color, ) def display_frames(self): """Display frame.""" if self.env_check: cv2.namedWindow("Ultralytics YOLOv8 Object Counter") # only add mouse event If user drawn region if len(self.reg_pts) == 4: cv2.setMouseCallback( "Ultralytics YOLOv8 Object Counter", self.mouse_event_for_region, {"region_points": self.reg_pts} ) cv2.imshow("Ultralytics YOLOv8 Object Counter", self.im0) # Break Window if cv2.waitKey(1) & 0xFF == ord("q"): return def start_counting(self, im0, tracks): """ Main function to start the object counting process. Args: im0 (ndarray): Current frame from the video stream. tracks (list): List of tracks obtained from the object tracking process. """ self.im0 = im0 # store image if tracks[0].boxes.id is None: if self.view_img: self.display_frames() return im0 self.extract_and_process_tracks(tracks) if self.view_img: self.display_frames() return self.im0 def incounts_updated(self): if self.in_counts_prev < self.in_counts: yield f"{self.in_counts}" def outcounts_updated(self): if self.out_counts_prev < self.out_counts: yield f"{self.out_counts}" if __name__ == "__main__": ObjectCounter()