The Research On Auxiliary Driving Complex Road Scene Classification Based On Single Image

With the rapid development of the rapid growth in automobile usage and road traffic, the driver assistance has been an important research topic of Traffic Safety. The navigation and the intelligent warning of driver assistance systems are depended on the understanding of the road scene. Vision-based road information extraction technology is the key technology of road scene understanding. In this paper, the following information extraction and intelligent warning methods from single road image have been studied extensively.Firstly, based on multi-feature fusion complex road scene classification method, the core idea is effective multi-feature fusion. Local color histogram feature of the global image feature extraction method,the gradient histogram features and improvements, and the features and integration of PCA dimension reduction,to obtain a low-dimensional integration features. The use of low-dimensional training sample fusion feature classifier training, the training of the test sample classification good classifier.Secondly, the road scene classification method based on pre-trained CNN. Use this design CNN network model training sample images to extract image features. Then use the extracted features training for road scene classification of SVM classifier, reuse classifier feature image scene classification.Finally, the road scene classification image gallery road type and number of species present in the application of intelligent transportation,building containing snow, village, street, highway, desert tunnels and 6 class road scene common database. In the database the trained classifier to the above two methods and experiments, a large number of experimental and comparison results show that the multi-feature fusion complex road scene classification average accuracy rate of 91%,based CNN pre-trained road the average rate of correct classification of the scene up to 91.27%, the validity of the method.