The cctv-prototype-deepsort from gitqinxinyu

cctv-prototype-deepsort's Introduction

About

A small 1-day prototype of cctv camera tracking cars and humans.

Run script for different video_id (representing different cctv camera)
Store in Redis database (should be easy to change)
Visualization dashboard prototype for summaries
- filtering on different objects, cctv and time intervals
- example map of where cctv is and its occurences
- can add conditions for warnings on specific video_id (should be coded in). e.g. if car passed through show warning in dashboard

Details

Python 3, PyTorch
Tracking from pytorch deepsort: https://github.com/ZQPei/deep_sort_pytorch (with a few edits)
Pre-trained net: YOLOv3
Dash/Plotly for visualization dashboard

Setup and Run

1: Pytorch deepsort:

Download YOLOv3 parameters

cd detector/YOLOv3/weight/
wget https://pjreddie.com/media/files/yolov3.weights
wget https://pjreddie.com/media/files/yolov3-tiny.weights
cd ../../../

Download deepsort parameters ckpt.t7

cd deep_sort/deep/checkpoint
# download ckpt.t7 from
https://drive.google.com/drive/folders/1xhG0kRH1EX5B9_Iz8gQJb7UNnn_riXi6 to this folder
cd ../../../

Compile nms module

cd detector/YOLOv3/nms
sh build.sh
cd ../../..

2: Database and Dashboard

Run Redis server
Run dashboard app.py (/dashboard)

3: Run

usage: python cctv_run.py VIDEO_PATH
                                [--help]
                                [--frame_interval FRAME_INTERVAL]
                                [--config_detection CONFIG_DETECTION]
                                [--config_deepsort CONFIG_DEEPSORT]
                                [--video_id CCTV_VIDEO_ID]
                                [--ignore_display]
                                [--display_width DISPLAY_WIDTH]
                                [--display_height DISPLAY_HEIGHT]
                                [--save_path SAVE_PATH]          
                                [--cpu]