Summary: | 碩士 === 南台科技大學 === 電機工程系 === 99 === This thesis presents an adaptive object background segmentation scheme for soccer robot system. The proposed system consists of three subsystems such as image preprocessing, adaptive object background segmentation, and image object recognition and classification. The image preprocessing subsystem aims to do color space conversion and image filtering for global image sequences of field. The adaptive object background segmentation subsystem integrates two algorithms of self-organizing map neural network (SOMNN) and temporal differencing (TD). In which, the SOMNN provides adaptive computations for segmentation of foreground from background; besides, the TD aims to calculate dynamic objects. To do union operation of these two outputs, it could segment out the static or dynamic foreground from the background. Based on the above results and morphological operations, in the image object recognition and classification subsystem, a learning vector quantization neural network (LVQNN) algorithm is applied to classify the color patches of the robots and the ball, and then to determine the details for real-time controls. Finally, the experimental results in 3-vs-3 android soccer competitions demonstrate the feasibility of the proposed system.
|