Object tracking using improved deep sort yolov3 architecture

This paper proposes a new architecture for object tracking. This design is the improved version of the deep sort yolov3 architecture. The correlation tracker of the Dlib is also inserted into the proposed architecture to reduce the identity switches. In addition, the novel architecture is designed with a parallel approach to boost the operating speed. Experimental results with different videos showed that the proposed architecture obtained lower identity switches and higher operating speed compared to the conventional deep sort yolov3 approach.