Vehicle Detection and Tracking in UAV Imagery via YOLOv3 and Kalman Filter

Unmanned aerial vehicles (UAVs) can be used to monitor traffic in a variety of settings, including security, traffic surveillance, and traffic control. Numerous academics have been drawn to this topic because of the challenges and the large variety of applications. This paper proposes a new and efficient vehicle detection and tracking system that is based on road extraction and identifying objects on it. It is inspired by existing detection systems that comprise stationary data collectors such as induction loops and stationary cameras that have a limited field of view and are not mobile. The goal of this study is to develop a method that first extracts the region of interest (ROI), then finds and tracks the items of interest. The suggested system is divided into six stages. The photos from the obtained dataset are appropriately georeferenced to their actual locations in the first phase, after which they are all co-registered. The ROI, or road and its objects, are retrieved using the GrabCut method in the second phase. The third phase entails data preparation. The segmented images' noise is eliminated using Gaussian blur, after which the images are changed to grayscale and forwarded to the following stage for additional morphological procedures. The YOLOv3 algorithm is used in the fourth step to find any automobiles in the photos. Following that, the Kalman filter and centroid tracking are used to perform the tracking of the detected cars. The Lucas-Kanade method is then used to perform the trajectory analysis on the vehicles. The suggested model is put to the test and assessed using the Vehicle Aerial Imaging from Drone (VAID) dataset. For detection and tracking, the model was able to attain accuracy levels of 96.7% and 91.6%, respectively.