Tracking without a Tracker. A Computer Vision Algorithm for Person Counting

Person counting is essential for applications such as store analytics, crowd management, and COVID-19 prevention. This paper aims to develop a tracking approach that isn’t tied to a particular tracker but instead depends solely on the number of detections. The findings reveal that the initial person detected is assigned ID=1, with subsequent individuals receiving increasing IDs with each new detection. An algorithm is suggested to sustain these IDs despite occlusions that may lead to confusion. Particular emphasis is placed on addressing situations where a person exits as another enters simultaneously, ensuring consistent and accurate ID assignment. We aim to use this work to camera prototypes that count persons for the prevention of getting common places overcrowded, by getting their occupancy, in order to prevent airborne diseases and especially the COVID-19 virus.