Saliency and Active Contour based Traffic Sign Detection

In this paper, we propose a new approach to detect salient traffic signs, which is based on visual saliency and auto-generated strokes for image segmentation. The proposed algorithm deals with two tasks on detecting traffic signs: auto-location and auto extraction. Firstly, inspired by recent work of visual saliency detection, we obtain the location of traffic signs in a natural image by multi-scale principle component analysis (MPCA). Secondly, in order to extract traffic signs, auto-generated strokes are used instead of drawing the strokes by the users, the sign board area is extracted using localizing Region-Based Active Contour. Extensive experiments on public datasets show that our approach outperforms state-of-the-art methods remarkably in salient traffic sign detection. Moreover, the proposed detection method has higher accurate rate and robustness to different natural scenes.