Research On Preceding Vehicle Detection And Warning Algorithm For Intelligent Vehicle

In the field of intelligent transportation, vehicle detection, dri ver fatigue detection, lane detection and pedestrian detection constitute the safety collision avoidance system for vehicle auxiliary driving together. Preceding vehicle detection and warning is one of the key technologies. It's used to detect whether there are vehicles ahead, and provide driving condition for safety system.This thesis mainly studies preceding vehicle detection and early warning algorithm based on the monocular vision. With the comparative analysis of vehicle detection and warning algorithm at home and abroad, we put forward an algorithm based on shadow hypothesis and la yered HOG symmetry characteristics verification for preceding vehicles detection and an algorithm of fuzzy warning for obstacles, then, we realize the preceding vehicle detection and warning system. The two main research contents are as follows:Firstly, the algorithm research of preceding vehicle detection. A new kind of layered HOG symmetry Feature is proposed in this paper, and a method based on shadow hypothesis and layered HOG symmetry characteristics verification is used to detect the vehicles in front of the camera. The whole process is divided into hypothesis stage and verification stage. In the hypothesis stage, considering the characteristics of the shadows below the preceding vehicles, the image sub-region, which would be the hypothetic sub-images of the vehicles, was extracted. In the verification phase, first, instead of the higher dimensional HOG, we used multilayer lower dimensional HOG to compute the HOG symmetrical vector, and mix the layered HOG vector and the symmetrical vector together as a Layered HOG Symmetrical Feature vector. Then, the Layered HOG Symmetrical Feature vectors of samples were used for ELM(Extreme Learning Machine) classifier training. At last, we verified the sub-images with the ELM classifier to get the test results. A lot of multi-scale change operations is been avoided in this algorithm, and it also improved the detection speed and accuracy.Secondly, the algorithm research of fuzzy warning for obstacles. This thesis uses the fuzzy warning algorithm to give warning information for the obstacles in front of the vehicle. According to the size, location and the position change in front and rear frame of the obstacles, the warning results are determined. And the warning results are divided into three levels: safety, notice and danger. The fuzzy warning algorithm has small calculating amount and high warning speed, which can meet the requirement of real-time warning, and can be applied not only to the front vehicle warning, but also to the pedestrians and other obstacles warn ing.In the hardware implementation of the system, we designed a special camera device mounted on the top of the vehicle, to collect the environment information of the vehicle. As to software, this thesis uses the Visual Studio 2008+OPENCV 2.3.1 as the compiled environment, and chooses C++ as the programming language. Software system mainly includes two modules: the detecting module and the warning module. Analysis of the experimental results show that: the vehicle detection and warning system can detect the front vehicle and make the corresponding decision accurately, and can satisfy the real-time requirements of the system.