Real-time abnormal behavior classification using AI-based CCTV

Clone the repository recursively.

$ git clone --recurse-submodules https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch.git

Install dependencies for ai-cctv.

pip install -r requirements.txt

cd Yolov5_DeepSort_Pytorch
python3 track.py --source <Path for the video file> --yolo_weights yolov5n6.pt --classes 0 --save-txt

or using script file.

./scripts/track.sh

if permission denied,

chmod 755 ./scripts/trach.sh

output text file will be generated at Yolov5_DeepSort_Pytorch/inference/output.

About XML Meta data format: here

Parse XML file.

python3 utils/xml_parser.py <filename>

use image_getter.py to get only images with abnormal action by using XML meta data.

python3 utils/image_getter.py <video name> Yolov5_DeepSort_Pytorch/inference/output/<result name> <output name>

with some restrictions,

1. at abnormal actions.
2. detect 2 people. (not 1 or 3)

API server for sending the detected abnormal actions.

uvicorn main:app --reload

Client for the abnormal detection.

npm i
npm run start

AI Hub

yolov5

Yolov5_DeepSort_Pytorch