Technology Of Road Sign Detection And Recognition Based On Machine Learning

The traffic sign recognition(TSR)system can capture the road scene using cameras for the real time detection and categorization,and these different types of traffic signs help driver to achieve safe driving.Therefore,traffic sign recognition system has a very real significance,playing an important role in assisting intelligent vehicle driving and improving driving quality.In this paper,we focus on the various issues of traffic sign recognition in natural scene,such as light conditions,the presence of similar color noise around traffic signs and partial occlusion,carry out extensive and depth research aiming at the accuracy of the entire system.The main work and the innovations of this paper are as follows:1)Aiming at the various problems which influence the accuracy of the traffic sign detection in natural scene,such as the dramatic changes in lighting conditions,the presence of similar color noise around traffic signs,we propose a novel traffic sign detection method based on edge enhancement MSER feature.Firstly,to reduce the effect of illumination change in RGB color space,we use gray world balance for preprocessing,followed by color enhancement and morphological filtering to clearly distinguish traffic signs from environmental background.And then,the traffic sign ROI candidate regions are extracted through the edge enhanced MSER feature.Finally,these candidate regions are further filtered using a shape analysis method based on Hough transform.At last,we perform comprehensive comparative experiment based on published German Traffic Sign Detection Benchmark(GTSDB)data sets,and results show that the accuracy rate of proposed method has improved by 20%,compared to the same type of detection methods.2)Aiming at the various problems which influence the accuracy of the traffic sign detection in natural scene,such as shooting angle,motion blur and partial occlusion,we propose a novel approach for the detection of traffic signs,which is based on the phase symmetry of traffic sign's shape type.The detection process has four stages,consisting of a color enhancement,a phase symmetry computation,a morphological filter and a maximally stable extremal regions(MSER)detection phase.We combine the color information of traffic signs using color enhancement with the symmetry of its shape type performing phase symmetry computation to highlight the Region of interest(ROI).At last,the candidate regions of traffic signs are detected as MSER and more reliable traffic sign candidate regions are got through empirical geometry constraints.The proposed system attains a highly accuracy up to 94%on the GTSDB.3)In the foundation of the above research work on the detection method,for the multi-classification problem of traffic signs,this paper deeply compare and analyze the recognition performance of real-time multi-class classifiers while histogram of oriented gradient(HOG)feature is in different dimensions,such as Random Forest(RF),support vector machine(SVM).And in order to enhance the performance of the model training time and classification efficiency,we proposed a feature space reduction method based on Fisher criterion.Experimental results show that the recognition accuracy of random forest classifier base on HOG feature has improved by 2.1%,compared to SVM classifier based on the same type HOG feature.