A Fast Multi-Object Tracking Method using DDR Descriptor
Kasama Leewan, Nipat Jongsawat, Anucha Tungkasthan · 2021
In the visual intelligence system, the two main processes for visual analysis are object detection and object tracking. At present, the first process is overcome by using the YOLO-based frameworks. It shows very good results as the object detection tool using deep learning technique. However, many researchers realize that there are lots of obstacles and challenges in the second process. Therefore, the objective of this paper is to propose the Distance, Direction, and Aspect Ratio (DDR) method that can be used to improve the effectiveness of object tracking process. The main contribution of this paper is to present the technique that achieves greater than Deep SORT algorithm in term of speed for tracking objects. The experimental results show that our proposed method can be used to track object more accurately compared to a Deep SORT algorithm but consume less processing time than a Deep SORT algorithm at a frame rate, 25 frames per second on average.