Real-Time Human Detection and Tracking System: A Novel Comparative Study of Centroid Tracking, Single Shot Detection and YOLO Algorithms

This study presents the development of a device capable of implementing social distancing monitoring using Centroid Tracking Algorithm, Single Shot Detector (SSD), and YOLO which run on powerful computers or servers with dedicated GPU s due to their complex computations and resource-intensive deep learning models. This is to support the government's continuous policy on social distancing in public places in Metro Manila and regions that still have the highest COVID-19 cases. Using as few people as possible due to restrictions on physical spaces, this study also developed a novel test that would compare the two said algorithms for optimal an Intel NUC hardware-based object detection and object tracking system with a focus on counting people. Using real-time footage from a CCTV camera with OpenCV and Python, the data from the simulations was then sent and analyzed to the web server cloud platform, ThingSpeak. Overall, it was determined that YOLO algorithm has less errors than the Centroid Tracking + SSD Algorithm and was thus more compatible with the developed system.