Object Detection and Tracking Algorithms for Vehicle Counting: A Comparative Analysis

The rapid advancement in the field of deep learning and high performance\ncomputing has highly augmented the scope of video based vehicle counting\nsystem. In this paper, the authors deploy several state of the art object\ndetection and tracking algorithms to detect and track different classes of\nvehicles in their regions of interest (ROI). The goal of correctly detecting\nand tracking vehicles' in their ROI is to obtain an accurate vehicle count.\nMultiple combinations of object detection models coupled with different\ntracking systems are applied to access the best vehicle counting framework. The\nmodels' addresses challenges associated to different weather conditions,\nocclusion and low-light settings and efficiently extracts vehicle information\nand trajectories through its computationally rich training and feedback cycles.\nThe automatic vehicle counts resulting from all the model combinations are\nvalidated and compared against the manually counted ground truths of over 9\nhours' traffic video data obtained from the Louisiana Department of\nTransportation and Development. Experimental results demonstrate that the\ncombination of CenterNet and Deep SORT, Detectron2 and Deep SORT, and YOLOv4\nand Deep SORT produced the best overall counting percentage for all vehicles.\n