Research On Road Traffic Sign Detection

Traffic signs, the important road safety facilities, are of great significance in guiding and warning both pedestrians and drivers. With the rise of intelligent transportation technology, road traffic signs recognition has become one important part of application in automotive driver assistance and driverless cars areas. Therefore, research on road traffic signs detection and recognition becomes one of hot issues in machine vision, intelligent transportation areas and so on.For above issues, road traffic sign detection and classification technology using pattern recognition and image processing are studied in this paper. Based previous work, the traffic signs segmentation with fusion features of color and shape is proposed against the deficiency of the previous algorithm using only color information, thus improving the accuracy; then the paper proposes the learning-based traffic signs recognition algorithm with HSC (Histograms of Sparse Codes), which is built on the HOG (Histograms of Gradient)-based algorithm and improves the accuracy of road traffic signs recognition to some extent.The main contributions of this thesis are as follows:(1) Due to the difference of domestic and international traffic signs in design as well as complexity of background, the accuracy is also affected when detected in different data sets. And there’s no domestic bench mark for traffic signs detection, for which300pictures taken in Beijing which include502traffic signs are prepared for experimental data set.(2) Traffic sign segmentation algorithm with the fusion features of color and shape information is proposed. Based on our previous works, we improve the color-based segmentation algorithm and avoid the influence of noise which color is similar but excessive with signs to the adaptive threshold, which strengthens its robustness to interference in a large extent. Also shape segmentation algorithm based on Fourier descriptor is proposed for assisting segmentation, which can reduce the effect of noise and improve the accuracy to a large extent.(3) Traffic sign recognition algorithm based on learning-HSC feature is proposed, and results both of HSC feature-based and HOG feature-based algorithm are compared in experiment. Meanwhile the impact caused by different parameters is also tested and analyzed in both situations.