A Self-Adaptive Algorithm for Traffic Sign Detection in Motion Image Based on Color and Shape Features
Ka Zhang,Yehua Sheng,Zhijun Gong,Chun Ye,Yongqiang Li,Cheng Liang
DOI: https://doi.org/10.1117/12.760485
2007-01-01
Abstract:As an important sub-system in intelligent transportation system (ITS), the detection and recognition of traffic signs from mobile images is becoming one of the hot spots in the international research field of ITS. Considering the problem of traffic sign automatic detection in motion images, a new self-adaptive algorithm for traffic sign detection based on color and shape features is proposed in this paper. Firstly, global statistical color features of different images are computed based on statistics theory. Secondly, some self-adaptive thresholds and special segmentation rules for image segmentation are designed according to these global color features. Then, for red, yellow and blue traffic signs, the color image is segmented to three binary images by these thresholds and rules. Thirdly, if the number of white pixels in the segmented binary image exceeds the filtering threshold, the binary image should be further filtered. Fourthly, the method of gray-value projection is used to confirm top, bottom, left and right boundaries for candidate regions of traffic signs in the segmented binary image. Lastly, if the shape feature of candidate region satisfies the need of real traffic sign, this candidate region is confirmed as the detected traffic sign region. The new algorithm is applied to actual motion images of natural scenes taken by a CCD camera of the mobile photogrammetry system in Nanjing at different time. The experimental results show that the algorithm is not only simple, robust and more adaptive to natural scene images, but also reliable and high-speed on real traffic sign detection.