Multi-objects real time recognition based on color information

Multi-objects real time recognition encounters several difficulties, such as the large quantity of data that have to be processed, the fast speed of recognition that guarantees the real time and the disturbance from the change of environmental illuminating level. In this paper, we present a new technique to pre-sampling and real-time segmenting, which combined two image segmentation techniques: the simplified nearest neighbor classification and color space thresholding. This method simplifies sampling and result in good segmentation and real-time performance. We also describe the bit mark method used to represent object´s color and the real-time recognition arithmetic based on grid sequential lookup. Experiments in real applications showed that the sensitivity in pixels classification is decreased, anti-jamming and real-time performance are increased, and it is robust to the change of lighting.