Research On Region Of Interest Based Road Traffic Signs Detection Algorithm

In the face of diversified road environment and increasingly complex traffic safety problems,intelligent transportation system(ITS)and advanced auxiliary driving system came into being,and have received extensive attention.At the same time,traffic sign recognition system as the core technology and frontier field of ITS system,has gradually become a popular research topic of many domestic and foreign scientific scholars.In the complex road environment,traffic sign is susceptible to many factors such as illumination,weather,color degradation,shape distortion and dynamic blur.Therefore,the effective road traffic sign detection method has become a research hotspot.The main contents of thesis include the following points:(1)Since the road environment of natural scene is complex and changeable,Therefore,the collected traffic sign images need to be pretreated to better detection and recognition.In thesis,with purpose of enhancing image,Gamma correction and gradient sharpening are adopted to reduce the influence of light intensity change.Meanwhile,thesis not only compares the ability of noise suppression about median filtering and adaptive multilevel median filtering but also Gaussian filtering and adaptive Gaussian filtering.Finally,the threshold segmentation effects of RGB,HSV,Lab three different color spaces are compared.(2)A traffic sign detection method based on region of interest and random forest classifier is proposed,The detection method first uses the linear SVM classifier to color-transform the traffic sign image,combined with the idea of shape template matching to calculate the traffic sign matching score on the corresponding scale and position,and compares it according to matching score and previously set threshold value to obtain a final candidate ROIs region;Then,the PHOG feature descriptor of the candidate ROIs region is extracted and the random forest classifier is generated in combination with the structural model learning strategy to complete the removal of the interfering ROIs region,thesis improve the accuracy of the target area detection;Finally,three sets of different contrast experiments were carried out in the 380 traffic sign image test databases,which were taken in the different road environment of the surrounding schools,the results show that the method based on ROIs and random forest classifier has a good detection effect.(3)To address the problem of low recall and high error rate of the traditional traffic sign detection algorithm based on ICF and Adaboost,thesis proposes a traffic sign detection method based on integral channel feature and multi-class classifier,The detection method is based on the adaptive spectral clustering of the ICF feature and the Adaboost algorithm to obtain the ROIs region of the target,then histogram equalization is imposed on regions of interest,use of LLC encoded SIFT feature and linear SVM combined to generate four different shape classifier.Finally,through the verification of GTSDB database,the results show that the detection method of traffic cascade classifier based on SICF-Adaboost + LSIFT-SVM is robust to illumination variation,shape distortion,color degradation and dynamic blur.