Performance Analysis of YOLOv4-Based Multi-Object Tracker Using SORT

Jelynelle G. Bastasa,

Daryl I. Cerina, Lester C. Tubo,

and *John A. Bacus

College of Engineering Education, University of Mindanao, Philippines

Abstract

Object tracking has become fundamental in the technology that greatly contributed to and is widely used in security, health, and many fields in the industry. With computer vision, the technology develops and improves its performance. This study aims to develop an automated system that tracks multiple people simultaneously utilizing the combination of YOLOv4 and SORT algorithms. Moreover, the resulting data were gathered to conclude that the network size of the algorithm used in testing is directly proportional to the accuracy and indirectly proportional to the frame rate. Using NVIDIA GTX 1650 GPU, the system attains 6.67 FPS at 608 network size with Multi-Object Tracking Accuracy (MOTA) and Higher-Order Tracking Accuracy (HOTA) of 27.22% and 22.71%, respectively. It is also expected to improve performance when utilizing a more powerful device and algorithm.

Keywords

artificial intelligence, computer vision, human tracking, multi-object tracking, YOLO.

How to cite?