Maintaining Social Distancing Using Artificial Intelligence

Abstract

Listen

Translate article

I spearheaded the development of a model that can detect whether people have breached the six-foot distance rule and alert each user who has. The software is installed in the hardware of CCTV cameras and can perform effective contact tracing. The application goes through the video footage and does the following process for every frame in the video. It uses Python libraries such as Keras-RetinaNet to detect which objects are human and where they are located in the frame. Then, it uses the law of similar triangles to find the distances between people, whose accuracy is improved by 30% using the law of cosines to find a more accurate distance. The application then uses Multi-task Cascaded Convolutional Neural Networks to grab people’s faces from an image and then uses a face recognizer to recognize the names of people in the frame. The application then alerts every user that broke the social distancing rule through email and logs the instance into a database from that frame. For every frame, the application uses cv2 to create an edited image of the original that displays which people broke the social distancing rule, by writing names on top of their bounding boxes and drawing red lines between the people who broke it. In the end, the application outputs an edited video clip of the original, which in reality is a collection of all the edited frames. Already it has begun to show promise, with an 85 percent accuracy rate.