Extracting Vehicle Track Information from Unstabilized Drone Aerial Videos Using YOLOv4 Common Object Detector and Computer Vision

Several applications are common in providing vehicle detection and tracking in videos. Unfortunately, these applications are most effective in stationary surveillance situations where background subtraction yields phenomenal result for object detection. Machine learning models, such as Single Shot Detector like You Only Look Once (YOLO), do not depend on motion to detect objects of interest in images; however, they suffer from high misdetections due to varying sizes of objects particularly in drone footage of urban areas. In this paper, we present a method for extracting temporal vehicle track information from urban area drone surveillance videos that out-performs state of the art YOLO + deepSORT multi object tracker and achieves vehicle counts comparable to manual human counts.