A target recognition algorithm based on color clustering and seed filling

PAN Haipeng, HU Lihua, LIU Yu
(Institute of Automatic, Zhejiang SciTech University, Hangzhou 310018, China)

Abstract: Aiming at the problem that the target recognition is impressionable of illumination variation in humanoid robot soccer, conventional seed filling algorithm was improved, and a target recognition algorithm based on color clustering segmentation and seed filling was proposed. The HSI color space was employed in the target recognition algorithm. Based on intensity and saturation, the image was separated to two parts, which were color area and gray area, by the method of thresholding. Then the color area was sorted to different color classes by clustering method based on the hue histogram. Meanwhile, the improved seed filling algorithm was used to minimal expand the color regions and extract regions′ features. In this way, fast and precise target recognition was implemented. Experimental results show that the proposed method is higher free of disturbance, precise in segmentation and satisfies the realtime. It has certain practical value.
Key words: robot soccer; image segmentation; feed filling; target recognition