Low-Body-Part Detection using RGB-D camera

Abstract

The reliable perception of a human in a dynamic environment is the most critical issue for interactive human-robot services. In human-robot interaction, a camera on a robot naturally captures the low-body-part of human because robots are usually shorter than the human. Conventionally, a two-dimensional laser range finder is used in low-body-part detection [1, 2]. However, these methods may cause errors when there are similar structures with legs. This video demonstrates a low-body-part detection scheme that not only exploits three-dimensional characteristics and but also the RGB features of the low-body-part. We build the low-body-part candidates by clustering from the legs to the heap. In the results, spurious candidates are eliminated by the proposed method.