A spatiotemporal tensor-based multi-object matching algorithm

Abstract

A multi-object matching algorithm based on spatiotemporal tensor is proposed for stereo localization systems with binocular views. Multi-object matching is formulated as a graph matching problem, which forms an efficient and extensible framework for the various matching cases. Then, the spatiotemporal tensor, including appearance, geometric and temporal cues, is used to improve the reliability and accuracy rate of matching in various conditions. Also, a high-order graph matching algorithm is introduced and improved. The experimental evaluation shows that proposed algorithm is robust to various cases, accurate in terms of matching rate, while being time-consuming little.