Research Of Vehicle Recognition Technology Based On Traffic Video Image Processing

Today, developing smart city makes people’s work, study and life becomes more intelligent. Being an important part of people’s daily lives, transportation desperately needs the help of technological means to manage better and reduce traffic accidents. Vehicles are the main elements of the transportation system, so the management of vehicles can be a good drive of entire transport system’s management. Vehicle identification has a very wide range of applications and rich application value in a variety of traffic management systems. Due to the complexity of the vehicle and traffic scenes, vehicle recognition technology has not achieved good results and breakthroughs in practice, so this paper carried out the method to improve vehicle recognition rate.Taking into account multiple interfere in complex traffic scenarios, the video image pixel is modeled object, and algorithm model of detection of moving target vehicle based on Gaussian mixture is proposed under complex traffic scenes. Selecting the multi-dimensional model describe the target to increase the reliability of vehicle detection background model. To shadow which the movement of vehicles produced in the course of the campaign, it gives a shadow suppression method based on color space conversion. While taking into account other small interference noise in the background of the model, it gives interference noise suppression methods based on the largest connected domain.To ensure the accuracy and integrity of the vehicle extraction, the experimental data of Guangzhou Wushan Road transport video demonstrate the effectiveness of this method.By the distribution of gray projection of the vehicle in the horizontal and vertical directions, the paper proposed a method of coordinate positioning of vehicle rearview mirror. With the coordinates of the vehicle rearview mirror, it gives the coordinate of the vehicle face and completes segmentation of the car face. To divided face of car, this paper focus on extracting HOG and fast edge detection based on structured forests, and stitch the two characteristic features to recognize vehicle. The algorithm can be made to effectively identify models that can adapt to various changes in the angle of the vehicle and has good robustness.In this paper, different extracted characteristics were further processed, SE edge detection and HOG feature are integrated effectively based on the PCA algorithm. In this paper, it gives the vehicle classification method based on support vector machine classifier. In order to adapt to the situation of uneven distribution of sample data, we give AdaBoost-SVM classifier which is comprehensive iterative algorithm, consist of iterative algorithm AdaBoost and Support Vector Machine. Through continuous training and iteration of samples, it can achieve the final strong classifier by weighted combination. The classifier combines the advantages of two algorithms, not only to overcome the dimension and local minimum problems, but also have higher accuracy and stability. Finally, it verify the effectiveness of the algorithm using BIT-vehicle identification database which is the standard-model.