Research On Methods Of Vehicle Detection Based On Machine Vision

In recent years, with the rapid development of computer vision technology,intelligent transportation technologies based on vision are used more and more widely in the life. Vehicle detection based on machine learning is one of the key problems of the intelligent transportation field, also is an important part in the object detection field.The accuracy of vehicle detection has an important influence on subsequent intelligent transportation research. This thesis mainly studies the single-view vehicle detection algorithm, multi-view vehicle detection algorithm, vehicle tracking algorithm, vehicle type recognition algorithm. The main contents are as follows:First, we study an aggregated channel feature extraction method. The feature not only contains the generalization of HOG channel feature, but also contains the color channel and gradient channel features; We discuss the vehicle detection feature selection problem under the multi-view, compare the discriminative sub-categorization algorithm with k-means and Latent SVM clustering algorithm.Second, we research on the effects of the other three type classifiers of soft cascade Ada Boost classifier on vehicle detection called soft cascade Real Adaboost, soft cascade Gentle Ada Boost, cascade Modest Ada Boost. We analyze the effects of parameters of the three classifiers on the performance of detection; Combined with the parameters, we put forward a better max pooling classifier.Third, we discuss an improved tracking algorithm called minimum output sum of squared error filter. Used in frame interval detection, we analyze the effects of interval numbers between two frames on detection performance. We apply HOG, LBP, DSIFT feature to extract the local feature of vehicles, project the high-dimensional feature to low-dimensional space combined with the random projection technology; We use SVM to divide the detected vehicles into different categories, and compare with mainstream recognition algorithms.Fourth, based on the Win Form framework we design a complete vehicle detection platform, including vehicle image detection, vehicle video detection, vehicle recognition,etc.