Research And Implementation Of Image Unwrapping And Target Detection And Tracking Method Of Omni-Directional Vision

Human obtain information surrounding environment depends mainly on visual perception and it accounts for about70%of the total information of human access to. Compared with the traditional vision system, omni-directional vision technology is a emerging visual perception technology and it has been applied popularly in military, aviation and civil facilities due to its advantage of wide field. This paper mainly focused on the interpolation algorithm and target detection and tracking algorithm has been studied and then developed an omni-directional vision software.Due to the omni-directional image has distortion problem caused by concentric circles and omni-directional vision doesn’t conform to the observation habit of human eye, so it is necessary to unwrap the omni-directional image. The paper introduced the unwrapping method and then analyzed the interpolation method used in the process of unwrapping. The performance of nearest interpolation and bilinear interpolation were compared and the improved algorithm of partial interpolation method based on lookup table was proposed lastly. The experimental results showed that the improved algorithm could satisfy the requirement of real time system.The paper also studied the commonly used algorithm in target detection and compared the advantages and disadvantages of Optical flow, Frame difference and Background subtraction. Then the results of target detection adopting Frame difference and Gaussian mixture model were compared. Combined with the complexity of actual environment, such as leaves shaking, water wave etc, and the complexity of object motion condition, the paper pointed out that the Gaussian mixture model was better than the Frame difference.Then the paper introduced the algorithm of target tracking including CamShift algorithm and Kalman filter algorithm. CamShift algorithm is based on target color tracking, so the target search box will expand and even lost when target is covered or similar color object appears around the target. Kalman filter is a kind of optimal estimation method and can minimize the error caused by noise in the system according to the forecast and update process constantly, and then obtained the optimal state estimation. The paper proposed Kalman filter combined with CamShift algorithm to solve the problems in the CamShift algorithm and could realize the target tracking effectively.Lastly, the paper implemented an omni-directional video software on Visual Studio2008platform combined with OpenCV. The software could realize the real-time unwrapping, target detection and target tracking of the omni-directional video.