/Action_Recognition_Surveillance

Action_Recognition_Surveillance, for huamn fall down action recognition, helmet、smoke、cell-phone detection.

Primary LanguagePython

Action Recognition for Surveillance

Using YOLOv5s to detect each person in the frame and use AlphaPose to get skeleton-pose and then use ST-GCN model to predict action from every 30 frames of each person tracks.

Which now support 7 actions: Standing, Walking, Sitting, Lying Down, Stand up, Sit down, Fall Down.

Using modfied YOLOv5 to detect helmet, cell phone, and smoking.

Prerequisites

  • Python >= 3.6
  • torch >= 1.7.1
  • opencv-python >= 4.1.2

You can also use "pip install -r requirements.txt" directly.

Original test run on: i7-8750H CPU @ 2.20GHz x12, GeForce RTX 2060 6GB, CUDA 10.2

Data

Train with rotation augmented COCO person keypoints dataset for more robust person detection in a variant of angle pose.

For actions recognition used data from Le2i Fall detection Dataset (Coffee room, Home) extract skeleton-pose by AlphaPose and labeled each action frames by hand for training ST-GCN model.

Pre-Trained Models

Configure environment variables

  1. open ld.so.conf file

    sudo gedit /etc/ld.so.conf

  2. add HikVision SDK lib path in new line

    Project_path/HiKcamSDK/lib

  3. update .so library list

    sudo ldconfig

Basic Use

  1. Download all pre-trained models into ./Models folder.
  2. Run my_multi_thread.py or Run image_input.py
    python my_multi_thread.py

Reference