Efficient Multi-Object Tracking Techniques using DAC Feature for Video Processing Applications

Object Tracking (OT) is still a challenging area of research, especially Multi-Object Tracking Accuracy (MOTA) in complex scenes. In the past, most popular methods used the global bounding box feature to represent an object because these methods calculate very quickly. However, it is inefficient of accuracy when applied to multi-object tracking processes where objects in a scene have similar box characteristics, such as cars and people. Therefore, this paper presents a technique for increasing the efficiency of tracking multiple objects in video files by adding global low-level visual content to the global bounding box feature. The experimental results show that the proposed method can increase tracking accuracy and maintain reasonable processing performance.