A Portable Object Recognizing And Tracking System Based On Omnidirectional Vision

Machine vision is the most direct and most effective and intelligent scientificmethod of robot. The robot can get the rich external information analysis quickly andconveniently, making it realize the cognition of the world and the obstacles of escapeand other functions, by the use of focus technology based on image feedback, auto-mation technology and network communication technology. In the research field ofwalking robot, machine vision is an essential part of feedback, providing trafficinformation of obstacles for robot to make a judgment before or after the fall.High cost, unintelligent, strong pertinence and lack of real-time are the mainproblems in the traditional vision solutions. Across the omnidirectional field, de-spite the panoramic binocular cameras, the distortion is serious. However, the ef-fect of3D reconstruction error is big, which cannot meet the requirement of therobot perception of environmental information.In view of the above problem, the design scheme utilizes uniform distributioncameras splicing scheme to realize real-time omnidirectional vision. Firstly, fivecameras are common webcam with low cost. Two of them make the binocu-lar camera, realizing the3D reconstruction. Besides, the binocular camera and theremaining three cameras are respectively equipped with one degree of freedomwhich expands the vision information from panel to three dimension space. Sec-ondly, this system is equipped with an embedded operating system as the com-mand center of camera freedom and image processing center, realizing the intel-ligent scheme of the system. This system, which is able to process image data it-self and can be access to image data and control the position of camera by network,has the characteristics of strong portability. Finally, face detection,3D recon-struction, table tennis, locating and tracking algorithm, in this paper, are used toevaluate the performance of the system. Fortunately, the experimental results showthat this system has the ability to real-time detection and tracking human face,make binocular camera accurately at the target and be effective in3D reconstruc- tion. Therefore, this system can real-time provide detailed environmental infor-mation for humanoid robot.This paper introduces the overall design scheme and image processingknowledge related used in the following. Then elaborates the construction of thewhole system from three aspects: the mechanical structure design, control systemdesign, image processing algorithm. At the same time, experiments are conductedto test the performance of the system. Then it comes to the final results.