MB

• Home
• About
• Portfolio
• Archive
• CV

Crowd Detection and Counting

Code

import os
from ultralytics import YOLO
import supervision as sv
import numpy as np

# Choose your video path
CROWD_VIDEO_PATH = f"./data/Crowd_Activity.mp4"

# choose the model
model = YOLO('yolov8s.pt')

Initial Detection

# extract video frame
generator = sv.get_video_frames_generator(CROWD_VIDEO_PATH)
iterator = iter(generator)
frame = next(iterator)

# detect
results = model(frame, imgsz=1280)[0]
detections = sv.Detections.from_yolov8(results)

# annotate
box_annotator = sv.BoxAnnotator(thickness=1, text_thickness=1, text_scale=0.3)
frame = box_annotator.annotate(scene=frame, detections=detections)

%matplotlib inline
sv.show_frame_in_notebook(frame, (16, 16))

Ultralytics YOLOv8.0.35 🚀 Python-3.11.7 torch-2.2.0 CPU
YOLOv8s summary (fused): 168 layers, 11156544 parameters, 0 gradients, 28.6 GFLOPs

0: 960x1280 14 persons, 2 skateboards, 1 vase, 444.0ms
Speed: 3.0ms pre-process, 444.0ms inference, 17.4ms postprocess per image at shape (1, 3, 1280, 1280)

Filtering

Since our goal is to detect only people, we will alter the code to filter our all other classes, We also add the class label here.

# extract video frame
generator = sv.get_video_frames_generator(CROWD_VIDEO_PATH)
iterator = iter(generator)
frame = next(iterator)

# detect
results = model(frame, imgsz=1280)[0]
detections = sv.Detections.from_yolov8(results)
detections = detections[detections.class_id == 0]

# annotate
box_annotator = sv.BoxAnnotator(thickness=1, text_thickness=1, text_scale=0.3)
labels = [f"{model.names[class_id]} {confidence:0.2f}" for _, confidence, class_id, _ in detections]
frame = box_annotator.annotate(scene=frame, detections=detections, labels=labels)

%matplotlib inline
sv.show_frame_in_notebook(frame, (16, 16))

0: 960x1280 14 persons, 2 skateboards, 1 vase, 461.5ms
Speed: 1.4ms pre-process, 461.5ms inference, 1.2ms postprocess per image at shape (1, 3, 1280, 1280)

Polygon Zone

Now we can add a polygon zone to the scene and visualize it.

sv.VideoInfo.from_video_path(CROWD_VIDEO_PATH)

VideoInfo(width=640, height=480, fps=29, total_frames=769)

# initiate polygon zone
polygon = np.array([
    [300, 125],
    [510, 125],
    [580, 430],
    [300, 430]
])
video_info = sv.VideoInfo.from_video_path(CROWD_VIDEO_PATH)
zone = sv.PolygonZone(polygon=polygon, frame_resolution_wh=video_info.resolution_wh)

# initiate annotators
box_annotator = sv.BoxAnnotator(thickness=1, text_thickness=1, text_scale=0.3)
zone_annotator = sv.PolygonZoneAnnotator(zone=zone, color=sv.Color.white(), thickness=2, text_thickness=1, text_scale=0.5)

# extract video frame
generator = sv.get_video_frames_generator(CROWD_VIDEO_PATH)
iterator = iter(generator)
frame = next(iterator)

# detect
results = model(frame, imgsz=1280)[0]
detections = sv.Detections.from_yolov8(results)
detections = detections[detections.class_id == 0]
zone.trigger(detections=detections)

# annotate
labels = [f"{model.names[class_id]} {confidence:0.2f}" for _, confidence, class_id, _ in detections]
frame = box_annotator.annotate(scene=frame, detections=detections, labels=labels)
frame = zone_annotator.annotate(scene=frame)

%matplotlib inline
sv.show_frame_in_notebook(frame, (16, 16))

0: 960x1280 14 persons, 2 skateboards, 1 vase, 382.7ms
Speed: 1.2ms pre-process, 382.7ms inference, 0.9ms postprocess per image at shape (1, 3, 1280, 1280)

Detect only in the Polygon zone

# extract video frame
generator = sv.get_video_frames_generator(CROWD_VIDEO_PATH)
iterator = iter(generator)
frame = next(iterator)

# detect
results = model(frame, imgsz=1280)[0]
detections = sv.Detections.from_yolov8(results)
mask = zone.trigger(detections=detections)
detections = detections[(detections.class_id == 0) & (detections.confidence > 0.5) & mask]

# annotate
labels = [f"{model.names[class_id]} {confidence:0.2f}" for _, confidence, class_id, _ in detections]
frame = box_annotator.annotate(scene=frame, detections=detections, labels=labels)
frame = zone_annotator.annotate(scene=frame)

%matplotlib inline
sv.show_frame_in_notebook(frame, (16, 16))

0: 960x1280 14 persons, 2 skateboards, 1 vase, 390.9ms
Speed: 1.0ms pre-process, 390.9ms inference, 0.8ms postprocess per image at shape (1, 3, 1280, 1280)

Multiple Polyzones

colors = sv.ColorPalette.default()
polygons = [
    np.array([
        [300, 125],
        [510, 125],
        [580, 430],
        [300, 430]
    ], np.int32),
    np.array([
        [110, 125],
        [300, 125],
        [300, 430],
        [25, 430]
    ], np.int32)
]
video_info = sv.VideoInfo.from_video_path(CROWD_VIDEO_PATH)

zones = [
    sv.PolygonZone(
        polygon=polygon,
        frame_resolution_wh=video_info.resolution_wh
    )
    for polygon
    in polygons
]
zone_annotators = [
    sv.PolygonZoneAnnotator(
        zone=zone,
        color=colors.by_idx(index),
        thickness=2,
        text_thickness=1,
        text_scale=0.5
    )
    for index, zone
    in enumerate(zones)
]
box_annotators = [
    sv.BoxAnnotator(
        color=colors.by_idx(index),
        thickness=1,
        text_thickness=1,
        text_scale=0.3
        )
    for index
    in range(len(polygons))
]

# extract video frame
generator = sv.get_video_frames_generator(CROWD_VIDEO_PATH)
iterator = iter(generator)
frame = next(iterator)

# detect
results = model(frame, imgsz=1280)[0]
detections = sv.Detections.from_yolov8(results)
detections = detections[(detections.class_id == 0) & (detections.confidence > 0.5)]

for zone, zone_annotator, box_annotator in zip(zones, zone_annotators, box_annotators):
    mask = zone.trigger(detections=detections)
    detections_filtered = detections[mask]
    frame = box_annotator.annotate(scene=frame, detections=detections_filtered)
    frame = zone_annotator.annotate(scene=frame)

%matplotlib inline
sv.show_frame_in_notebook(frame, (16, 16))

0: 960x1280 14 persons, 2 skateboards, 1 vase, 404.9ms
Speed: 1.2ms pre-process, 404.9ms inference, 1.3ms postprocess per image at shape (1, 3, 1280, 1280)

Processing the Whole Video

colors = sv.ColorPalette.default()
polygons = [
    np.array([
        [300, 125],
        [510, 125],
        [580, 430],
        [300, 430]
    ], np.int32),
    np.array([
        [110, 125],
        [300, 125],
        [300, 430],
        [25, 430]
    ], np.int32)
]
video_info = sv.VideoInfo.from_video_path(CROWD_VIDEO_PATH)

zones = [
    sv.PolygonZone(
        polygon=polygon,
        frame_resolution_wh=video_info.resolution_wh
    )
    for polygon
    in polygons
]
zone_annotators = [
    sv.PolygonZoneAnnotator(
        zone=zone,
        color=colors.by_idx(index),
        thickness=2,
        text_thickness=1,
        text_scale=0.5
    )
    for index, zone
    in enumerate(zones)
]
box_annotators = [
    sv.BoxAnnotator(
        color=colors.by_idx(index),
        thickness=1,
        text_thickness=1,
        text_scale=0.3
        )
    for index
    in range(len(polygons))
]

def process_frame(frame: np.ndarray, i) -> np.ndarray:
    print(i)
    # detect
    results = model(frame, imgsz=1280)[0]
    detections = sv.Detections.from_yolov8(results)
    detections = detections[(detections.class_id == 0) & (detections.confidence > 0.5)]

    for zone, zone_annotator, box_annotator in zip(zones, zone_annotators, box_annotators):
        mask = zone.trigger(detections=detections)
        detections_filtered = detections[mask]
        frame = box_annotator.annotate(scene=frame, detections=detections_filtered, skip_label=True)
        frame = zone_annotator.annotate(scene=frame)

    return frame

sv.process_video(source_path=CROWD_VIDEO_PATH, target_path="./runs/crowd_result.mp4", callback=process_frame)

from IPython import display
display.clear_output()