An Improved Chinese Chessman Recognition Method for Robot in Natural Environment
Muhammad Ilyas Raza, Vachiraporn Ketsoi, Jianhui Zhao · 2017
Computer vision is the prerequisite of chess playing robot in a natural environment. In this paper, a new method is presented for recognition of Chinese chessman with higher precision and less computing expense. Firstly, the noise is removed from the captured image by Gaussian filter. Secondly, the chess circles are detected by Hough circle transform, and the chess color is detected using the HSV color model, then the detected red and black character pixels are expanded by dilation. Thirdly, the positions of chessman are estimated based on detected circles and dilated regions, then chess positions are adjusted to their proper places. Finally, scale-invariant feature transform (SIFT) and nearest neighbor matching are both employed, together with sorting matching numbers, to recognize the chess characters. We have tested 100 sets of chessman and each set contained 12 different Chinese characters with random rotation, then compare with the traditional approach. Experimental results indicate that our proposed method outperforms in both consuming time and recognition accuracy.