Research On Human Tracking For Robot Based On The Double-layer Cooperative Locating Mechanism

With the rapid development of robot technique, robot has become a majormethod for helping person’s daily life. As a key issue for a robot, human tracking hasattracted more and more attentions. Numerous approaches have been proposed totrack person. However, there still are some technical problems, such as occlusion,limited view field and distinguish people. To solve these problems, a human trackingsystem based on the double-layer cooperative locating mechanism was proposed. Inaddition, a visual interactive interface was developed. The proposed method wasverified by experiments conducted on the Pioneer3with cameras and RFID detectionsystem. The research work mainly included the following aspects:(1)Adaptive template matching based on head-shoulderIt is inevitable that the detected person is often occluded by other objects incomplicated environment. As a prominent feature of upper-body, the head andshoulder silhouette can be easily identified and is rarely sheltered. So it is an idealcharacteristic model for human detection. An adaptive template matching algorithmbased on head-shoulder is presented. In the algorithm, the person’s silhouette wasextracted through image processing in the images. The extracted silhouette haddifferent scales due to different depth. So the extracted head-shoulder silhouette andthe stored template were normalized. The Euler distance was adopted to measure thesimilarity between the normalized template and silhouette.(2)Adaptive multi-feature mean-shift algorithmThe traditional mean-shift algorithm cannot effectively handle inadequaterepresentation of target’s color distribution, invariable target model and occlusion.Aiming to resolve these problems, an adaptive multi-feature mean-shift algorithm wasproposed. In this algorithm, an improved kernel function was obtained by combiningthe silhouette of the target. This method could eliminate the affection of thebackground. Furthermore, the color-texture histogram was extracted to achieve targetmodel. A selective update strategy was proposed to adaptively update the target model.The update scheme can not only eliminate the accumulated error, but also avoiddrifting away.(3)Double-layer cooperative locating mechanismThe detection range of the RFID system is larger than that of the camera. On theother hand, the locating of stereo-vision is more accurate than that of the RFID.Therefore, a double-layer cooperative locating mechanism was proposed toeffectively take advantage of the RFID system and stereo-vision. The tracking results,obtaining from the coarse locating and fine locating, were assembled to locate thetarget robustly. When the target was occluded, the EKF was used to estimate the person’s motion between consecutive frames. In addition, when the target was out ofview due to quick turning, the coarse location was utilized to control the robot andfind target as soon as possible.(4)Intelligent gear shift algorithm based on the fuzzy controlA motion control algorithm, controlling the robot follow target and remain in thesafe distance from the target, was essential for human tracking system. Therefore, afuzzy rule based intelligent gear shift controller was presented. In this algorithm,considering the state of the target and robot, the reference velocity and turning-gainwere adjusted by two fuzzy controllers. The velocity of the robot adaptively changedaccording to computed reference velocity and turning-gain. The control algorithmenabled the robot to follow the target continuously and stably.The human tracking algorithm described above was verified by experimentsconducted on the Pioneer3-DX robot platform. Experimental results showed that theadaptive template matching was effective for locating the human which was in frontof camera view. The adaptive multi-feature mean-shift could handle inadequaterepresentation of target’s color distribution, invariable target model and occlusion.The double-layer locating mechanism which was based on the data of vision andRFID could estimate the person’s position when the target was occluded, and guiderobot by coarse locating processing when the target was out of view. The mechanismeffectively improved the robustness of tracking system. The intelligent gear shiftalgorithm based on fuzzy logic could remain the distance between robot and givenperson in a safe range.