A RANSAC-Based Track Initialization Algorithm for Multi-Sensor Tracking System

Abstract

For multi-sensor tracking system, track initialization is a problem due to crossed data association between different frames and different sources. In this paper, a RANSAC-based track initialization algorithm for multi-sensor system is proposed. Considering the fact that target motion is contiguous in time-space domain, measurements of the target are homologous to the target motion model. Therefore, the proposed algorithm uses homology to associate all measurements regardless of frames and sources. RANSAC technology is used to recursively learn the target motion model, associating continuous measurements from all sensors. It is better to avoid missing measurement effects in some frames or in some sensors through a homology association mechanism. Compared with the traditional track initialization algorithm, the proposed algorithm maintains good robustness in noise and clutter environment, which is confirmed by simulation results in this paper.