Human recognition in a cluttered indoor environment by sensor fusion

Abstract

The detection of human in a cluttered space is an incumbent requirement for a mobile robot functioning in an indoor environment. This paper addresses this issue in a threefold approach method to solve the human detection in an indoor environment by a mobile robot. Firstly, the distance information is obtained from the RGB-D camera and the LiDAR sensor and then it is evaluated with an extrinsic calibration method for an accurate tracking. Then, an object detector YOLO v3 is used for the human classification and detection from the image generated by the RGB-D camera. YOLO then returns the region of interest (ROI) which will be selected for the depth data for generating the human position. Lastly, the selected depth information of human is added to the position information of LiDAR sensor for more accurate localization of the human. The proposed approach shows effective human tracking result for small mobile robot interacting with human.