Research Of Adaptive Recognition In Theambient Light Based On Reinforcement Learning

Machine vision is an important research direction in the field of robotics. In the process of identifying specific target by the visual system,it is susceptible for the robot in variable illumination conditions. In order to solve those problems, the paper proposes a method which can recognize the target in variable illumination. Using reinforcement learning algorithm, the goal of adaptive picking up target can be achieved by automatically adjusting the parameters under different lighting conditions.The paper, first of all, sums up ways of picking up the target in variable illumination conditions. And the illumination model, the color space model and principle of camera are analyzed. On this basis, the image is converted from RGB model to HSV model and is separated three channels. Threshold method is applied to identify the parameters of the target information. After the implementation of morphological image processing, body contour information of target can be obtained by edge detection algorithm, then using contour to recognize the target.Discrete segmentation is used for light intensity. For the situation of the color chromaticity range constant, but saturation and brightness vary under the same object and different light, we use reinforcement learning to achieve adaptive turning about target body color saturation and brightness parameter values. Many reinforcement learning algorithms are compared with analysis and Sarsa algorithm is selected. The relevant parameters, strengthen function and search strategy are determined.Using Open CV library and VC++ develop related learning software, the algorithm is validated by the experimental platform, which uses computer as upper computer and microcontroller as the lower computer. Experimental results show that the method of the paper is correct and feasible and in variable illumination conditions to pick up the target body has good adaptability and robustness.