Physical Distancing Monitoring with Background Subtraction Methods
Hendra Adinanta, Edi Kurniawan, Suryadi Suryadi, Jalu Ahmad Prakosa · 2020
World Health Organization (WHO) has confirmed that the spreading of coronavirus disease 2019 (COVID-19) could be avoided by keeping the physical distance at least 3 feet (1 meter). Then, we have a motivation to employ computer vision techniques to monitor social distancing violations. The principle of the works are to detect persons, then to assess the physical distancing violation from their distance. Most of the researchers have tried to utilize object detection methods such as faster RCNN, Yolo, and SSD to detect persons from the frame. Those methods rely on, the support of Graphics Processing Unit (GPU) to execute their heavy computation. In this works, we propose social distancing monitoring by applying background subtraction methods based on Gaussian Mixture Models (GMM) i.e. Geo-metric Multigrid (GMG), k-Nearest Neighbor (KNN), Mixture of Gaussian (MOG), and Mixture of Gaussian 2 (MOG2). These methods have been used to filter persons from the frame with computational process. Some parameters evaluation measures have been determined to check the best method suitable for this works. In terms of performance, better methods are ranked as KNN, MOG, MOG2, and GMG.