Online Multiple Object Tracking using a Detection Mean Confidence-based Track Management Method

JIITA, vol.4 no.3, 2020, pp.380-386, DOI: SOON

Online Multiple Object Tracking using a Detection Mean Confidence-based Track Management Method

Young Chul Lim 1,*), Minsung Kang 1)
1) Research Division of Future Automotive Technology, Daegu Gyeongbuk Institute of Science & Technology, Daegu, Republic of Korea

Abstract: In this paper, we propose a track management scheme based on the detection mean confidence to create correct tracks and to terminate false tracks efficiently and reliably. The detection mean confidence (DMC)-based tracking method follows a detection-by-tracking framework which utilizes the output of an object detector as the observation model of a Bayesian filter and updates the target state with a motion model and an observation model. In online multiple object tracking methods, detected objects are associated with current tracking objects in each frame. Accordingly, these methods are more vulnerable to noisy detections than offline methods. In this paper, we propose a reliable track management method to tackle these problems. Our experimental results show that our approach improves the accuracy of multiple object tracking while minimizing the numbers of false positives and false negatives.

Keyword: multiple object tracking; track management; object detection;