Visual Multiple-object Tracking Algorithm Based on Motion Consistency

Abstract

The visual multiple-object tracking module is a key component of an active onboard obstacle detection system. However, the most of currently used visual multiple-object tracking algorithms rely on offline calculation for object detection, without adequately considering the adverse effect on tracking attributed to the time consuming nature of offline calculation in actual applications. This study presents a visual multiple-object tracking algorithm that leverages multi-characteristic fusion within a multi-thread asynchronous system framework for visual obstacle detection, designed in consideration of actual application environments. The proposed algorithm reflects an optimized cascade matching strategy, primarily depending on the motion consistency characteristic index developed in this study based on the target motion vector, and incorporating some common cues in object tracking research, such as the appearance model and Mahalanobis distance. The algorithm aims to improve the stability in tracking multiple objects with similar appearance features and motion patterns, while maintaining stability in normal cases. The proposed multiple-object tracking algorithm integrated into the above-mentioned framework was verified in experimental analysis. According to the experimental results, the proposed method demonstrates the capability to stably track various targets in actual application environments, outperforming the baseline method in terms of longer-lasting stable tracking performance.

Keywords

• multiple target tracking
• obstacle detection system
• motion consistency
• cascade matching
• object detection