IN/OUT counting multiple times

[CodeWithEhtisham]

I am implementing functionality to track people entering and exiting a space. However, I am encountering an issue when a user stops at the door. The center point of the bounding box interacts with the line multiple times, causing the count to increase repeatedly. How can I resolve this? I want each person to be counted only once for entering or exiting.

POLYGON = np.array([[rect[0][0], rect[0][1]], [rect[1][0], rect[0][1]], [rect[1][0], rect[1][1]], [rect[0][0], rect[1][1]]])
print(f"Polygon Coordinates: {POLYGON}")

# Define the line from the user-drawn line
LINE_START = sv.Point(line[0][0], line[0][1])
LINE_END = sv.Point(line[1][0], line[1][1])
print(f"Line Coordinates: Start={LINE_START}, End={LINE_END}")

# Initialize the line zone for counting
LINE_ZONE = sv.LineZone(start=LINE_END, end=LINE_START, triggering_anchors=(sv.Position.CENTER,))

tracker = sv.ByteTrack(minimum_consecutive_frames=6)

# Initialize annotators
box_annotator = sv.BoxAnnotator(thickness=1)
label_annotator = sv.LabelAnnotator(text_thickness=2, text_scale=1)
polygon_zone = sv.PolygonZone(polygon=POLYGON, triggering_anchors=(sv.Position.CENTER,))
trace_annotator = sv.TraceAnnotator()
line_zone_annotator = sv.LineZoneAnnotator(text_scale=0.8, text_orient_to_line=True)
smoother = sv.DetectionsSmoother(length=5)

# Variables for in/out counting
in_count = 0
out_count = 0

# Process video frames
for frame in tqdm(frames_generator, total=video_info.total_frames):
    # Detect objects in the frame
    results = model(frame, imgsz=1280, conf=0.5, classes=[0], device='cuda:0', verbose=False)[0]
    detections = sv.Detections.from_ultralytics(results)
    
    # Apply the polygon zone to filter detections
    detections = detections[polygon_zone.trigger(detections=detections)]
    
    # Track the detections
    detections = tracker.update_with_detections(detections)
    detections = smoother.update_with_detections(detections)
    print(detections)
    # Trigger the line zone and count in/out events
    a=LINE_ZONE.trigger(detections=detections)
    # print(LINE_ZONE.in_count)
    # print(LINE_ZONE.in_count_per_class)
    # print(LINE_ZONE._in_count_per_class)
    # print(a)
    # Increment counters based on direction (in/out)
    # in_count += LINE_ZONE.in_count
    # out_count += LINE_ZONE.out_count
    lable = [
        f"{id}"
        for id in detections.tracker_id
    ]
    # Annotate the frame
    frame_copy = frame.copy()
    frame_copy = sv.draw_polygon(scene=frame_copy, polygon=POLYGON, color=sv.Color.RED, thickness=2)
    frame_copy = box_annotator.annotate(scene=frame_copy, detections=detections)
    frame_copy = trace_annotator.annotate(scene=frame_copy, detections=detections)
    frame_copy = sv.draw_line(scene=frame_copy, start=LINE_START, end=LINE_END, color=sv.Color.RED, thickness=2)
    frame_copy = line_zone_annotator.annotate(frame_copy, line_counter=LINE_ZONE)
    frame_copy = label_annotator.annotate(scene=frame_copy, detections=detections, labels=lable)

    # Display counts on the frame
    # cv2.putText(frame_copy, f"In: {in_count}", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
    # cv2.putText(frame_copy, f"Out: {out_count}", (10, 70), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)

    # Display the frame
    cv2.imshow("Processed Frame", frame_copy)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

# Release resources
cv2.destroyAllWindows()

[LinasKo]

Hi @CodeWithEhtisham 👋

I expect to have the solution in the next few days via #1540

[CodeWithEhtisham]

Thank you for getting back to me. Please let me know when it becomes available.

[LinasKo]

Hi @CodeWithEhtisham 👋

It turned out significantly harder than I thought, but it's merged.

To use the code immediately, first install the develop version like so:
!pip install git+https://github.com/roboflow/supervision.git@develop

Then, set crossing_acceptance_threshold in your LineZone to a value greater than 1.
Docs: https://supervision.roboflow.com/develop/detection/tools/line_zone/#supervision.detection.line_zone.LineZone-functions

Note: I expect this to be available in the standard supervision package on Friday night, when we launch the new version.

[CodeWithEhtisham]

thank you @LinasKo