Research On Road Surface Condition Video Image Identification Based On Support Vector Machine

ABSTRACT:Bad road conditions are an important cause of road traffic accidents,when the vehicle in the rain,snow,fog and other bad weather driving which would prone to vehicle rollover and other major traffic accidents.Therefore,in the case of poor road conditions,accurate and rapid identification of road conditions has great practical significance for the efficient and safe operation of the highway.With the development of video image processing technology and the popularity of expressway video surveillance system,surveillance cameras have become the main traffic monitoring facilities.So,the use of video image technology to identify the road surface condition has become the current research hotspot.However,the identification of road surface conditions which under different lighting conditions and mixed road surface conditions are two urgent problems that need to be addressed.This paper proposes the use of video image technology to detect road surface status,and the development of road surface detection algorithm which can meet the needs of different road.The specific research contents are as follows:(1)In this paper,the road surface image is divided into 5 types:dry,wet,snow,ice and water.According to the original image size,the image block principle is established,and the image block of the single state is extracted to construct the road surface condition image library,which ensures that the image sample’s quality and purity.(2)The nine-dimensional color eigenvector is extracted by the third-order color moment method.The gray level co-occurrence matrix method is used to extract the energy,entropy,correlation and contrast texture features.Based on the 13-dimensional image feature vector,the pavement state characteristic database is formed.(3)In order to improve the identification accuracy of the algorithm,the kernel search algorithm is used to optimize the kernel function factor C and the penalty factor g in SVM.In this paper,a new method of road surface video identification based on SVM(support vector machine)is proposed.(4)Firstly,the parametric optimization SVM classifier is used to train multiple sets of different sample sizes to obtain the multi-group road surface condition image classification model.Then,the multi-group classification model is used to identify and select the best classification effect model.(5)Based on the experimental system and the video samples which from different acquisition methods,the original image of the actual road surface condition with large number of different environments and untrained training is verified by block identification.The experimental results show that the method based on SVM optimization classifier and video block identification is scientific and feasible,and the road surface classification model under grid search optimization algorithm is more than 90%accurate for single sample identification.The identification accuracy of the mixed road samples is more than 85%.This effectively solves the problem of the identification about mixed road surface condition and different lighting conditions.