Visual Servo Detection Methods In Complex Environment And Its Application Research In Parallel Robot

At present, various problems involved in the complex environment of visual servo detection have been considered in the field of image processing and pattern recognition. It is very important to find the content of visual servo system when it is in uneven illumination or the illumination is decreased. In this dissertation, the visual servo detection task in complex environment is studied, and the main research productions are presented as follows.(1) In order to avoid interferences of complex environment, a good image enhancement approach based on retinex algorithm will be executed initially. After that, as to corresponding problem of visual focuses extracted by visual attention model, a multi-visual integrated model is proposed and applied for feature patch in complex scene. At first, an initial visual focus set is extracted by this integrated visual attention model. Through mutual information and structure similarity, a fusion visual focus label set can be created by a corresponding matrix. Then two visual focuses should be combined, and the fusion energy distribution between them should be calculated. In order to rapidly getting the combined visual focus, the two visual focuses area will be reset by us ing uniform energy distribution. At the end of visual focus fusion, the feature area will be integrity and robust. Applying this method to feature patch extration on parallel robot in complex scenes can get good effects and could be serve for camera calibration.(2) In complex environment, it is difficult to extract the edge information of the calibration reference. Here, two kinds of edge detection methods are designed. The first is the short step affine transform Sobel which can achieve the visual servo in the speed and effect. However, it cannot guarantee the edge closing. The direct result is that the edge moments and the edge of the affine invariant moments cannot be used. The second is saliency narrow-band active contour model. It will use visual attention model to extract the visual focus. Then the edge of the visual focus will be the initial boundary of the C-V model and a good smooth closed edge will be obtained by rapid evolution. Finally, through analyzing the experimental result, the effectiveness and robustness of the saliency narrow-band C-V model can be validated.(3) In view of the camera calibration problems in complex environment, an edge moment calibration method is proposed. Using affine invariant moment theory, the edge affine invariant moments are designed to distiguish calibration reference or other objects. In the definition of edge moments, the relation equation between edge moments increment and the camera position increment are established. By solving these equations, the camera parameters will be obtained. Through simulation experiment, binocular camera calibration based on e dge moment can ensure the accuracy and robust of calibration in complex environment of camera calibration.(4) As to difficult to find the feature points in weak texture image, a suitable method for stereo matching of visual servoing is proposed. SIFT algorithm can obtain many feature descriptors. A part of the feature descriptors will be extracted Meanshift based feature selection mechanism and then the Meanshift weak texture feature descriptor could be calculated. Those Meanshift weak texture feature descriptors will be used for stereo matching. The graph cut optimization algorithm makes the matching energy reaching minimum and the visual depth map value of same feature points will be piecewise discrete to integer interval so as to get the whole visual depth map. The experimental results demonstrate the weak texture feature descriptor is superior other descriptors including SIFT, SURF and NCC in stereo matching effect and adaptive in complex environment.Finally a conclusion is made for the whole contents of this dissertation both the disadvantages of the current work being noted, and the perspectives of this field for the next step have also been discussed.