Abstract: A novel codebook model combined with the superpixel segmentation method is proposed in this paper to improve the efficiency of object detection. The original pixels are clustered based on the similarities of both color and location information to reduce the processing cost. Our revised codebook model based on the superpixel could not only suppress the effect of local noise, but also reduce the redundancy of codebook. Simulation results indicate that, our proposed algorithm could reduce the memory consumption and improve the processing speed significantly without sacrificing the detection precision. Our algorithm could implement the object detection in real time in the embedded video processing system based on the DM6437 processor.