Video Pedestrian Event Detection

English｜中文

Pedestrian tracking and gender recognition(red: female, green: male)

Pedestrian tracking and helmet wearing identification(red: not wearing helmet, green: wearing helmet)

Pedestrian tracking and face mask wearing identification(red: not wearing face mask, green: wearing face mask)

Pedestrian tracking and using phone identification(red: using phone, green: not using phone)

Pedestrian tracking and smoking detection(red: smoking, green: not smoking)

Quick Start

1. Clone this repository

   git clone https://github.com/thu-huangzc/VPED.git
   git pull

2. Install necessary python packages

   pip install -r requirements.txt

   The Python version is 3.9.19 and the cuda version is 11.8. Of course, other python or cuda versions may also be possible. You can have a try.

3. Download pre-trained model checkpoints

   a. Pedestrain detection and tracking -> yolov8m.pt

   b. Helmet detection -> helmet_head_person_epoch10.pt

   c. CLIP model -> clip-vit-base-patch16

   d. Cigarette detection -> cigarette_epoch20.pt

   e. Phone detection -> phone_detection.pth

   Then, place yolov8m.pt, helmet_head_person_epoch10.pt and cigarette_epoch20.pt under ./ckpt/yolo/, place clip-vit-base-patch16 under ./ckpt/, place phone_detection.pth under ./ckpt/classifier.

4. Run demo

   a. Upload videos for offline testing

   CUDA_VISIBLE_DEVICES=0 python demo.py --input TEST_VIDEO_PATH --event EVENT --draw-results

   b. Camera online test

   CUDA_VISIBLE_DEVICES=0 python demo.py --input webcam --event EVENT --draw-results

   • input: 'webcam' or input video file path;
   • event: 'gender', 'helmet', 'mask', 'smoking' or 'phone' (More behavioral recognition will be added later);
   • draw-results: whether to save the inference result video, which will be saved under ./inference/test_videos by default.

Notes

Related Projects: Smart_Construction, yolo, clip, Safety-Helmet-Wearing-Dataset, MiVOLO

If you find this project helpful, we would greatly appreciate it if you could give us a free star～