Human Moving Target Detection And Its Application Based On RGB-D Information

Along with the continuous renewal and development of software and hardware technology, detecting human target with computer has become a research focus. In visual application such as robot, monitoring system and intelligent vehicle, human target detection not only has direct application value but also provides foundation for the research on computer’s perception of human target’s behaviors, thus, it has broad theoretical research value.As a reform for visual field, the emergence of Kinect enables more researchers to access to a broader visual field – depth information. When no key breakthrough has been made for the research on single image information detection, the study of RGB-D information detection has gradually become a research focus in visual field.This paper mainly uses Kinect for color-depth information(RGB-D) collection, develops research on RGB-D-based human moving target detection technology, and applies the research content to practice. The research is composed of three parts.(1) By improving human detection method based on RGB image and combining depth information, this research designs a more efficient RGB-D-based human target joint detection method. In specific, it makes use of polymerization channel characteristic and linear SVM cascade classifier as well as the assistant localization of depth prior information to finally complete the task of rapid human target detection. Experiments have shown that this method can reach the detection rate of 30 fps under the premise of losing some accuracy, thus meeting the detection system’s real-time requirement.(2) With the advantages of Kalman filter and particle filter, the research has achieved a tracking method integrating particle filter and Kalman filter. Apart from making up for the deficiencies of single filter tracking method, this method can meet the actual requirement of tracking moving target. Experimental comparative analysis has proved that the filter-integrated tracking method is superior to single filter tracking method.(3) According to above detection and tracking method, a detection tracking model is established and a kind of application software – real-time human counting system is completed, so that users can experience through software interface the research results of human moving target detection based on RGB-D information.