Vehicle Tracking based on an Improved DeepSORT Algorithm and the YOLOv4 Framework

Vehicle tracking plays an important role in traffic surveillance systems in which efficient traffic management is the main objective. During the last several decades, with the rapid growth of the number of vehicles, the task of detecting and tracking vehicles efficiently and accurately has become challenging. Existing algorithms often fail to track vehicles continuously throughout the video stream due to the nonlinear nature of vehicular motion and vehicle occlusion in crowded scenarios. This paper proposes a vehicle tracking algorithm as an improvement on DeepSORT (Simple Online and Realtime Tracking with a Deep Association Metric), based on an optimized YOLOv4 (You Only Look Once version 4) detector, the Unscented Kalman filter, and AlexNet as the feature extraction network which ensures better performance in tracking the non-linear motion of vehicles and tracking through occlusions with a reduced number of ID switches compared to state-of-the-art vehicle trackers.