Research Of Front Vehicle Recognition Technology Based On Vision

With the rapid development of economy and technology, Cars become the main transport, but they are accompanied by all sorts of traffic accidents and jams. Thus, with peoples' increasingly strong voice on smart car and self-driving, visual navigation technology have attracted much attention and research of scholars from home and abroad, because they are similar to human driving. Research of front vehicle recognition technology based on vision, as a core part of the visual navigation technology, is of great significance.According to the distribution of gray values of the road, establishing a Gaussian model. And then determining binary threshold value by the mean and variance of the model. Thus, the obtained binary threshold value will be changed by the type of road and light intensity, so that vehicle bottom shadow segmentation would have strong robustness in changing environment. After the image binarization and morphological filter, this paper adopted the way that obtaining the suspected vehicles according to the edge of shaded area at the vehicle's bottom to avoid that obtaining suspected vehicles area is insufficient according to rectangle features of the vehicle's bottom shadow area. Details methods list here. After determining edge of the shadow at the vehicle's bottom, then determining suspected area of the vehicle according to certain rules. In this situation, suspected area of the vehicle is greater than the area of the vehicle itself. And taking the Hough transform to get a further refinement of suspected area of the vehicle. Finally, getting accurately suspected area of the vehicle.After obtaining the suspected area of the vehicle, firstly using the gray scale symmetry to complete preliminary verification, then using the optimized SVM classifier to complete the secondary verification, thereby obtaining a final location area of the vehicle. The support vector machine(SVM) classification accuracy have great relationship with SVM parameters, In this paper, we use a genetic algorithm to optimize the parameters of SVM, so the SVM classifier can achieve the best classification results.This paper used the VS2008 development tools and visual processing open source library Opencv2.3.1 to realize front vehicle identification algorithm by C++ programming. After a lot of experimental verification, vehicle recognition rate of the algorithm can reach 90.6% in simple case of the road environment and reached 73.9% in complex case of the road environment.