Fusion of Thermal and Depth Image to Improve Human Segmentation for a Mobile Robot

Abstract

In machine vision, surveillance systems are a kind of security that concentrates on the safety of the human and property. One of the main tasks of a surveillance system is the detection of humans. This paper presents a system of human detection and the development of a technique of human segmentation using a combination of information thermal and depth in a real indoor setting from a mobile robot. A novel fusion of thermal-depth information (FTDI) is introduced to enhance the efficiency of the segmentation process and expedite processing. In experimental studies, evaluation of the performance for the proposed system is carried out using Ground Truth (GT), in which the proposed system yield is compared to GT. The proposed system performs well with an approximate accuracy of over 90% for all data sets as illustrated in the quantitative results and even outperformed state-of-the-art algorithms. This paper presents the novelty of the work, in which the detection method can improve the classification of persons and their occlusion. The advantages, such as being computationally inexpensive and performs well even under severe occlusion and poor illumination, show that this proposed system is robust.