Application Research Of Traffic Sign Recognition System Based On Embedded

With the rapid increase of vehicle ownership,the smoothness and safety of road traffic has become the primary concern of ITS.In the field of active safety of vehicles,traffic sign recognition systems can provide road traffic information to drivers in time and reduce the occurrence of traffic accidents.So it is extremely important to design a traffic sign recognition system that can be applied to automobiles.The research object of this paper is the three common traffic signs,which are mainly detected,tracked and recognized.Then the system algorithm is transplanted to the embedded platform for application implementation.In the traffic sign detection phase,in order to quickly detect traffic signs in complex backgrounds or disturbances,this paper uses the dual characteristics of color and shape to detect them.By comparing the color separation effect of RGB color space and HSV color space,the HSV color space is selected for color threshold segmentation to eliminate most of the background.Then the color segmentation image is combined with two algorithms to improve the detection accuracy,which are based on improved The Hough transform's circular detection algorithm and polygon approximation rectangle and triangle detection algorithms.In the traffic sign tracking phase,in order to accelerate the detection time of the video sequence,Kalman filter is used to track the video.At the same time,an improved Kalman filter tracking algorithm,which based on BP neural network,is proposed to improve tracking accuracy.It mainly uses the learning ability of the network to improve the estimation ability of the Kalman filter.In the traffic sign recognition stage,this paper adopts FHOG feature as the feature extraction method of traffic sign,which combines the advantages of both HOG feature and PCA,can extract features well,and A multi-class classifier,which based on SVM,can achieve better recognition results.Finally,this paper chooses AM5728 as the embedded platform for system algorithm application,and its can well meet the system requirements with heterogeneous multi-core video processing mode.Through the development of AM5728,the operating environment,which can support the traffic sign recognition system work,is compiled.Then the system algorithm is cross-compiled and transplanted to the AM5728 platform,and optimized to some extent.When the system is tested in the laboratory and the car respectively,it can achieve a good real-time and applicability.