Robust and efficient people detection with 3-D range data using shape matching

Abstract

Information about the location of a person is a necessity for Human-Robot Interaction (HRI) as it enables the robot to make human aware decisions and facilitates the extraction of further useful information; such as low-level gestures and gaze. This paper presents a robust method for person detection with 3-D range data using shape matching. Projections of the 3-D data onto 2-D planes are exploited to effectively and efficiently represent the data for scene segmentation and shape extraction. Fourier descriptors (FD) are used to describe the shapes and are subsequently classified with a Support Vector Machine (SVM). A database of 25 people was collected and used to test this approach. The results show that the computationally efficient shape features can be used to robustly detect the location of people. 1