Forward Vehicle Detection,tracking And Ranging Based On Monocular Vision

With the increasing number of private cars,the traffic safety problem has been paid more and more attention.Therefore,it is the current research hotspot to study the vehicle-assisted safety driving system which can improve traffic safety.Consider the expressway and urban loop as the typical working conditions,we study and implement the forward vehicle detection,tracking and ranging algorithm based on monocular vision.And then realize the development of the vehicle collision warning system based on monocular vision,which can improve driving safety significantly.Specific work as follows:1.We study,improve and propose a vehicle detection algorithm based on vehicle shadow fusion features,which improves the adaptability of the scene and the speed of the algorithm.Based on the assumption above,We propose a vehicle detection algorithm based on the gradient direction histogram feature and the support vector machine(SVM),which further improved the vehicle detection accuracy.We study and implement a vehicle detection algorithm based on the Haar-like feature and cascade classifier based on the active learning framework.Through the introduction of active learning framework iteration to improve the performance of the classifier,the algorithm detection accuracy rate reach to 92.7% with real-time.2.In this paper,we study and compares the performance between the MeanShift vehicle tracking algorithm based on kernel function weighted color histogram and the vehicle tracking algorithm based on real-time compressive.In contrast,real-time compressive vehicle tracking algorithms are not sensitive to changes in lighting,stably tracking multiple vehicle targets in complex driving scenarios.3.In this paper,we research and implementation a forward vehicle ranging algorithm based on the inverse perspective projection principle.Through the self-made calibration template and reference object,we realize the accurate and real-time distance estimation of the multi-target in the projection coordinate system.According to the distortion difference of different camera acquisition devices,we use the distortion correction processing based on the camera calibration technology to improve the accuracy of the follow-up distance calculation.The overall vehicle distance measurement algorithm satisfies the design error of less than 10% absolute error,the test results shows high calculation accuracy and real-time.4.Based on the above-mentioned algorithm,this paper proposes two kinds of forward vehicle collision warning software design based on monocular vision,and uses the MATLAB environment to quickly prototype the algorithm,batch test and optimization iteration,and then realize the portable C / C ++ code generate.The experimental results show that the proposed vehicle collision warning system based on monocular vision can perform real-time and stable detection and tracking of the vehicles in the typical situation,the whole system could improve the safety of the traffic significantly.The use of the above-mentioned workflow automatically generated engineering source code is highly readable,efficient implementation,strong code robustness.Corresponding workflow have important reference value for similar engineerning requirements.