Main structures of code are brought from people-counter and tiny-yolo object detector

This sub-project is one of modules for Anti-COVID19 robot which can detect cough person and record social distancing violations. This module is for the latter which traces the close contact made by infectee in the public space with a raspberry pi 4B- based surveillance camera. For this purpose, we utilize the pedestrian detection and object tracing to execute computing-burden jobs in the limited hardware environment. Contact information of observed people is then collected using Intel Realsense Depth Camera D435. The close contacts which last for a certain time or longer will be detected and recorded as a screenshot. In the last, an examiner only need to search and check the saved screenshots, instead of inspecting a long record of the surveillance camera.

• Intel Realsense Depth Camera D435 (Any depth camera will be fine)
• Raspberry pi 4B

1. Install the intel realsense python sdk
2. Install the OpenCV, NumPy, dlib, imutils

$ python3 main.py -s 30 -pd 150 -md 0.5 -se 10

• -s: Number of frames skipped between detections
• -pd: Minimum pixel euclidean distance between pedestrians for the contact detecting
• -md: Minimum meter distance between pedestrians and the camera for the contact detecting
• -se: Minimum time (seconds) for contact to be detected

The output will be stored as ./capture/{date-time of the contact}

We utilized the depth information from depth-cam to calculate 3d distance between people.

Tracking mode

output screenshot (capture/20210108-003020.png)

Overall design

Cough Detector design

Social Distancing Recorder design

Robot