We've Moved!
Visit SDSU’s new digital collections website at https://digitalcollections.sdsu.edu
Description
The purpose of this study is to develop a robotic system that is able to follow a particular person in a real-world environment without losing track due to objects or other people obscuring her. The study focuses on developing image-recognition and motioncontrol techniques for robotic person-following. It uses Microsoft's Kinect device, which combines infrared imaging with video imaging for robust recognition, and thus offers the possibility of identifying and following a particular person in a crowd. The proposed system first constructs skeleton images of every person in the frame using the joint coordinate locations provided by an onboard depth camera. To identify its target, it performs poserecognition based on the skeleton image of the person and learns about her physical characteristics such as height and body-part dimensions. In addition, hue and saturation histograms for the person's outfit are used to perform color-based recognition. A significant feature of the study is the introduction of a new histogram matching technique that adapts to changes in environmental light conditions. This, together with skeleton model matching improves the performance of recognition. The second important aspect of the study is the development of software for the robot that allows the cameras to stay focused on the person of interest. The effort also encompasses research techniques that allow the robot to maintain a relatively constant distance from the person, and for graceful and gradual motion transitions for the robot during person-following. The shortcomings of the infrared-based depth sensor and the robotic vehicle limit the system's applicability to indoor environments. Nonetheless, the improvements in reliability through the use of the image-processing algorithms developed over the course of this research mean that, if these technical limitations are overcome, the scope of the system's applicability will widen significantly
