Crowd monitoring and detection

Abstract

Covid-19 has impacted our life a lot ever since the virus emerged in early 2020. Due to its rapid spreading characteristics, several safety measures have been put into service for safe management. Singapore has implemented various laws and regulations regarding social gatherings, mask-wearing, safe distancing, etc.

Digital systems like Trace Together is widely used to facilitate contact tracing efforts. However, it requires a lot of manpower to check the scanning pass and vaccination status which always lead to a long queue and make social distance hard to maintain. Current trace together system is also not convenient to the elderly who cannot use a smartphone. The tapping-in device has a high faulty rate and cannot recognize a certain number of smartphones. In this case, a newly developed crowd monitoring system is eliminated to resolve the issues regarding manpower and inconvenience caused by the current trace-together system.

The project consists of two main parts: face mask detection and social distance tracking. The first part of the project focus on YOLOv5 CNN algorithm development and mask detection training. With thresholding setting and non-max suppression algorithm, a filter is developed to improve the accuracy of face mask detection. The second part aims at Bird’s Eye View distance calculation algorithm and centroid distance tracking. Bird’s Eye View area can be manually chosen, and the detection focuses on this area, which decreases the computation and improves the detection speed. Lastly, the combined output shows whether the object human is wearing a mask and the crowd is following social distance. A GPU of 1660Ti is used for model detection and OpenCV acceleration.