SODV: Social Distancing Violation System YOLO version 3

This project is a social distancing violation detection system implemented using Python. The previous development of this project used MobileNet SSD pre-trained on MS-COCO as the person detection algorithm. After the previous project finished in July 2020, I decided to further improve the detection algorithm by changing from MobileNet SSD to YOLOv3 to increase the accuracy. The program uses OpenCV for the image processing and utilizing the DNN module which solely tested on CPU. The system accuracy tested on Oxford Town Centre CCTV video-dataset (266 frames). This project was submitted to Malaysia Technology Expo (MTE) 2020 Special Edition COVID-19 International Innovation Awards under Faculty of Electrical Engineering of Universiti Teknologi MARA.

1. Prerequisites and configurations

All the requirements can be installed via the command:

$ pip3 install -r requirements.txt

The default input video is located in videos folder. To change the program to use camera stream as input, you need to change the configuration from CAMERA_FLAG : false to CAMERA_FLAG : true.

Note: All configurations can be changed in the config.json file.

2. Run project

Run:

$ python social_distancing_violation_system.py

3. Program structure

4. Program output

Output frame 10 to 250:

FRAME 10	FRAME 50	FRAME 100

FRAME 150	FRAME 200	FRAME 250

5. Overall accuracies

6. Accuracy for person detection

Dataset	TP	TN	FP	FN	%
Oxford Town Centre	29	0	0	11	72.5

7. Accuracy for social distance violation detection

Dataset	TP	TN	FP	FN	%
Oxford Town Centre	23	15	9	2	77.5

8. Project drawback

• No camera calibration for intrinsic parameter

9. References

Previous project
Person Detection for Social Distancing and Safety Violation Alert based on Segmented ROI

Output video

YOLO Pre-Trained Model

Dataset
MegaPixels: Origins, Ethics, and Privacy Implications of Publicly Available Face Recognition Image Datasets

LICENSE

This project is licensed under the terms of the MIT license.