A target localization method based on color recogni-tion and connected component analysis is presented in this paper. The raw image is converted to HSI color space through a lookup table and binarized, then followed by a line-by-line scan to find all the connected domains. By setting appropriate threshold for the size of each connected domain, most pseudo targets can be omitted and coordinates of the target could be calculated in the mean time. The main advantage of this method is the absence of extra filtering process, therefore real-time performance of the whole system is greatly improved. Another merit is we introduce the frame difference concept to avoid manually presetting the upper and lower bound for binarization. Thirdly, the localization step is combined with target enumeration, further simplified the implementation.
Experiments on our ARM system demonstrate its capability of tracing multiple targets under a mean frame rate of 15FPS, which satisfied the requirement of real-time video processing on embedded robot systems.