Fusion of thermal and depth images for occlusion handling for human detection from mobile robot

Abstract

In this paper we present a vision-based approach to detect multiple persons from a mobile robot in real world scenarios using thermal and depth images. First, human upper-body is detected from thermal image to obtain the bounding box coordinates (BBC). This is used as an initial estimate of the position of a person. Then, depth image is fused with the region of interest obtained from the thermal image. Occlusion detection is then performed using the BBC to obtain the actual position of detected persons. The proposed method is able to detect persons and is able to resolve occlusions during the detection process to correctly obtain all the persons in the view.