BP neural network based image segmentation technique for small sized league soccer robot team system

ZHU Yan-li, LI Xiao-ming
(Faculty of Mechanical Engineering & Automation, Zhejiang Sci-Tech University, Hangzhou 310018, China)

Abstract: BP neural network was adopted in the small sized league(SSL) robot soccer team for image segmentation in order to fulfill the requirements of real time processing and environmental light adaptation. The theoretical relationship between the image segmentation and BP neural network was studied, and the model of the BP neural network was established. The array of each component values(H, Cb, Cr) of all the pixels in the image were taken as the input of neural network, and the arrays of target pixel categories as the output. The convergence rate was improved by optimizing the learning ratio parameters and the error function definition. The background-subtraction method and normalization algorithms were applied to speed the segmentation process. The image segmentation experiment shows that the model and algorithm of BP neural network can solve the image segmentation problems for small size soccer robot vision subsystem, and demonstrate good performance in real timing and adaptability in different lighting conditions.
Key words: BP neural networks; image segmentation; soccer robot