Research On Vehicle Detection Algorithm Based On Monocular Vision

With the rapid increase in the number of cars,driving environment has become increasingly complex.In order to improve the safety and reliability of active driving and reduce the probability of traffic accidents,it is of great significance to research and develop vehicle safety aided driving system.At the same time,with the development of computer vision technology,vehicle detection and recognition based on visual information has become one of the hotspots of computer vision research.Compared with other objects,the appearance information of vehicle is rich and the influence of environment is relatively large,which demands high efficiency and robustness of vehicle detection algorithm.In this thesis,based on the analysis of the current situation of vehicle detection at home and abroad,the vehicle detection,feature extraction and other aspects of in-depth study.Firstly,several types of vehicle detection algorithms are analyzed.Then two kinds of classical vehicle detection methods,Haar feature + Adaboost classifier method and HOG feature + SVM classifier method,respectively,the vehicle detection test.Experimental results show that the latter method is superior to the former,but there is still a false detection rate and high detection rate of the problem.In order to solve the above problems,this thesis proposes an improved HOG feature algorithm(hereinafter referred to as PWHOG feature),which first solves the problem of missing spatial position information in the HOG feature,and then performs statistical analysis of the problem To improve,that is,for each pixel in the two adjacent bin to do a precise division.Experimental results show that the algorithm proposed in this thesis can improve the recognition rate and reduce the false detection rate and missed rate.Then,a new search algorithm based on Y-S linear model is proposed to solve the problem that the proposed PWHOG feature has too large dimensions and leads to a large decrease of detection speed.The Y-S linear model is obtained by least squares method,and then the NG distribution model is established according to the position of the vehicle in the image(The relationship between the number of vehicles and the position distribution).At last,the model is set up,the image is traversed and the model is updated in real time.Experimental results show that the algorithm proposed in this thesis improves the detection speed based on the accurate description of the vehicle size.At last,through a variety of databases,the algorithm proposed in this thesis and the classic algorithm were done a lot of experiments,and comparative analysis.The validity and superiority of the proposed algorithm in improving the detection accuracy and reducing the false detection rate,missed detection rate and speeding up the search are verified.In this thesis,the classical HOG feature is improved and the vehicle detection is realized by the proposed accelerated search algorithm.Finally,the effectiveness and superiority of the improved algorithm are proved by the comparative experiments.