Detection Algorithm Based On The Vision Of All-time Vehicle Safety Research

In the field of intelligent vehicle, Vision-based vehicle safety system has becomes an important topic. There are two parts in the system. One is a system which analyses the environment of the outdoor. The other is a system which gets the status of driver. The lane detection and the vehicle detection are prerequisites of the outdoor system. Find a stable and accurate detection algorithm is the basis for detection. In this paper, the algorithms which are for all-days lane detection and the vehicle detection at night were studied.Based on the rain interference as typical environment, a second search method was designed for noise suppression in the lane detection algorithm. At the same time, we built a dynamically update model for search threshold of intensity and suppression strength threshold, so that the algorithm of vehicle detection could automatically adapt to the environment change. Because of system's requirement for the detection stability, we designed a smoothing model among frames, making the algorithm results more stable.Two classifiers are set for vehicle detection algorithm. The first classifier, which is designed for the features of vehicle lights, is used for searching region-of-interest. The second classifier, which is designed for the edge features of vehicle, is the main classifier. It is a cascade classifier which trained by AdaBoost method. The method can effectively solve the problem of the reflected light from road which is superior on adaptability whereas the traditional approach is dependant on the detecting the lights of vehicle.Using images from reality road environment for test, the experiment result shows that the algorithm could finish the detection mission well, and could be replied in real project.