Online Multi-Object Tracking with Historical Appearance Matching and Scene Adaptive Detection Filtering

TL;DR

This paper introduces novel methods for multi-object tracking that utilize historical appearance matching and scene-adaptive detection filtering to improve identity consistency and handle occlusions and noisy detections effectively.

Contribution

The paper presents a historical appearance matching method and a joint-input siamese network trained in two steps, enhancing tracking robustness during occlusions and noisy detections.

Findings

Significant improvement in identity consistency during occlusions.

Effective removal of noisy detections based on scene conditions.

Enhanced robustness against temporal errors in multi-object tracking.

Abstract

In this paper, we propose the methods to handle temporal errors during multi-object tracking. Temporal error occurs when objects are occluded or noisy detections appear near the object. In those situations, tracking may fail and various errors like drift or ID-switching occur. It is hard to overcome temporal errors only by using motion and shape information. So, we propose the historical appearance matching method and joint-input siamese network which was trained by 2-step process. It can prevent tracking failures although objects are temporally occluded or last matching information is unreliable. We also provide useful technique to remove noisy detections effectively according to scene condition. Tracking performance, especially identity consistency, is highly improved by attaching our methods.