HAAR(Human Abnormal Activity Recognition) DeepModel (v.0.9)
Open Source Algorithm For Detecting Ship Passengers’ Abnormal Behaviors And Fall Accidents
Preventing accidents on ships and alerting for help is the goal of the detection algorithm. Prevent more accidents, save more lives.
Table of Contents
About The Project
Open Source CCTV-based AI algorithm which detects the abnormal behaviors of passengers on the ship to predict possible accidents and warn the onboard sailors. When the CCTV catches any abnormal behavior, the algorithm will detect the incidents and alert the current accident location to nearby coast guards in real-time to increase the rescue rate for the fallen passengers.
We defined activity of ship-passenger. (normal, abnormal)
[Walking, Lean-railing, Sit-down, Smoking, Move-Over, Standing ... ]
Demo Reel of the Model
Check out list of categories of behaviors of the passengers here
- Walking
- Move Over #1
- Move Over #2
Requirements
- Cuda, Cudnn : Cuda support GPU Device (We implemented RTX 3090)
- Detectron 2
- Linux or macOS with Python ≥ 3.7
- PyTorch ≥ 1.8 and torchvision that matches the PyTorch installation. Install them together at pytorch.org to make sure of this
- OpenCV is optional but needed by demo and visualization
- See Detectron Install.md
- AdelaiDet
- Detectron2 base
- See AedlaiDet Install.md
- FCPose, FCOS-Detection, Boxinst
Download pretrain Model : Key-Point
Download pretrain Model : Faster-RCNN
Download pretrain Model : Retinanet
Getting Started
The model can be started by executing haar_demo.py in /HAAR_Demo directory.
[Sample Run Script] (use custom model)
python HAAR_Demo/haar_demo.py \
--video-input ./HAAR_Demo/cctv_demo.mp4 \
--opts MODEL.WEIGHTS ./models/mymodel.pth