A New Automatic Vehicle Tracking and Detection Algorithm for Multi-Traffic Video Cameras

Abstract

Vehicle tracking systems are a vital tool in modern-day law enforcement and security operations. With the increasing threats of terrorism, organized crime, and illegal trafficking, monitoring and tracking suspicious vehicles has become a top priority for security agencies around the world. In this study, a target vehicle, which was described as suspicious, was tracked using the proposed vehicle tracking method that contains Gaussian Mixture Model (GMM) and Blob analysis. The same target vehicle was then detected using the Regions with Convolutional Neural Networks (RCNN), Faster RCNN, and You Only Look Once (YOLO) deep learning object recognition algorithms. In these applications, public traffic surveillance system images from the internet are used. Tracking is performed on images taken from more than one traffic surveillance system on the same road or route. The results from these methods were compared with each other, and the highest mean Average Precision (mAP) value was observed as 89.20% for the Faster RCNN algorithm using the Resnet101 deep learning architecture.