Salient traffic sign video detection based on hypercomplex frequency domain

Traffic sign detection under the dynamic video is one of the important research content in an intelligent traffic assistance safety system. The classic traffic sign detection methods adopted all kinds of visual features serial computation, failing to consider the interaction between various visual features, the detection precision need to be improved. Therefore, this paper proposed a novel multi-scale salient traffic sign video detection algorithm based on hypercomplex frequency domain. Firstly, hypercomplex vector is constructed by using the color visual and motion information of input video sequences. Then, the traffic sign location is detected by the normalized the multi-scale hypercomplex frequency domain visually salient image. Finally, the corresponding traffic sign is extracted by using the morphological operator in each video frame. The Experimental results indicate that compare with the traditional algorithm, the proposed method has higher precision and capacity of resisting disturbance to detect traffic sign under dynamic traffic background.