Smart surveillance system to monitor the committed violations during the pandemic

Abstract

The world now faces a medical crisis which needs to be resolved. COVID-19 is a disease which spreads between people mainly when an infected person is in close contact with another. To decrease the virus spreading, World Health Organization suggests some rules to follow such as wearing masks, social distance, and quarantining the infected people. In this work, we propose a surveillance system to monitor public spots and make sure those infected people don’t leave quarantine sites and to make sure that people wear a mask and practice social distance. Our proposed Research is designed to detect, track a moving person in video sequence and help to specify his profile when he is entering or leaving a special region. The system detects if they are wearing a mask or not, respecting social distance or not. Additionally, the proposed system informs the persons concerned in the field when the infected people enter in the monitored area or when they are found unmasked by performing a face recognition. The proposed system is used for tracking people in real time. It extracts frames from the video sequence and it performs the detection process using a deep learning pre-trained model and tracks people so we can analyze their behaviour and create profiles of the moving people in the area. We believe that the proposed system can minimize the jeopardy of the pandemic and it is definitely the optimal solution to contain the risk of the virus spreading, especially when manual human monitoring is almost impossible to cover the entire globe. Furthermore, we use transfer learning techniques to train a deep learning model for masked-unmasked face classification. The main novelty in the proposed system is threefold: (1) Adopting an efficient face detector to detect masked faces. (2) Synthesizing masked-face dataset to train masked-unmasked face classifier to decide whether people are wearing masks or not. (3) Adjusting the recognition algorithm to recognize the masked faces. The combination of these aspects gives excellent results. The experimental results show that the proposed system can perform very well in the real time execution.