Research Of Image Unwrapping And Real-time Target Tracking Based On The Omnidirectional Vision

The large field of view (FOV) of360°provided by the omnidirectionalview sensor, which has gradually emerged as a popular and effective visionsensor in the field of robotics, allows simultaneous monitoring of thesurrounding in different view angles and therefore enables a more flexibleand responsive algorithmic behavior. And real-time tracking of movingtargets in omnidirectional vision is the core issue of vision. To solvetemporary disappearance or interference from objects with same color whenusing traditional camera to obtain the interesting target under large-scaleand complex scenes, the target tracking algorithm based on cylinderunwrapping was investigated and the method was validated by theexperimental investigation with mobile robot.Firstly, to solve the distortion problem of panoramic image, areasonable panorama was obtained by using a360°camera to unwrap theomnidirectional image with cylinder unwrapping method. In this process, inorder to solve the pixel distortion problem, interpolation algorithm was usedto supplement the pixel. A local bilinear interpolation unwrapping algorithmwas improved in order to guarantee the image reducibility and real-timecharacteristic, by which clear panoramic video was obtained. Secondly,under unstructured complex scenes, CamShift combined with Kalman filteralgorithm is used to track the moving target effectively. Lastly, thisalgorithm was validated by the experimental investigation in Frontier-Ⅱautonomous mobile robot platform in Shanghai Jiao Tong University.Experimental results demonstrate that the omnidirectional vision system iscapable of real-time monitoring a large range of scenes around it the proposed method realize an real-time and robustness target tracking underlarge-scale and complex scenes, which contains moving target occluded,temporary disappearance or interference from objects with same color.