Considering the requirement of high accuracy and robustness for traffic sign detection in real-world environments, this paper proposed a novel method for automatic detection of traffic sign. There are three main stages in the proposed algorithm:1)segmentation by adaptive threshold in HSI color space to find the region of interest; 2)eliminate the noise present in the binary image and the small objects with morphological operations and mean filter;3)contour extraction and curve fitting for a better contour by the least square method. Experimental results show a high success rate and demonstrate that the proposed framework is invariant to illumination, deformation, and partial occlusions.