Vision Sensor-Based Shoe Detection for Human Tracking in a Human–Robot Coexisting Environment: A Photometric Invariant Approach Using DBSCAN Algorithm

Abstract

Human-tracking problems are considered as one of the major problems in a human-robot coexisting environment. Such systems often employ a sensor fusion mechanism involving a vision sensing-based system along with ultrasonic or IR sensing systems. In this paper, we consider a very important sub-problem, where vision sensing is employed in such a human-robot coexisting environment for shoe detection purposes, by approximating the target person's positions in successive frames captured by the onboard camera of a robot, during human tracking. This paper considers a more challenging scenario of shoe detection under photometric changes and proposes a novel variant of the CFAsT-match algorithm, called photometric-invariant CFAsT-match (PICFAsT-match) algorithm. Both the CFAsT-match and PICFAsT-match have been implemented using the DBSCAN algorithm, a density-based clustering approach. Performance evaluations carried out for real-life tracking of human shoes show the supremacy of the proposed PICFAsT-match algorithm.