Vehicle Collision Avoidance Technology Research Based On Information Fusion

With the number of cars increases every year, road traffic turns increasingly complex, in order to reduce road traffic accidents, the automotive intelligence and security technology become more and more urgent demands. Nowadays, the intelligent transportation systems makes the combination of the advanced information technology, data communications transmission technology, electronic control technology and computer processing technology that applied in the whole traffic management system, makes the good integration of artificial intelligence and the vehicle applied into an orderly and efficient system. In order to reduce the accident occurred such as rear-end accident, intelligent transportation systems in the automotive collision avoidance technology is particularly important, which helps the driver to get the front traffic conditions and remind drivers to make the appropriate emergency treatment according the real circumstances.This paper presents a vehicle collision avoidance method based on vision sensors and laser radar sensor information fusion. Through analyzing the principle for the variety of sensors, the range of applications, advantages and disadvantages, using the vision and multi-sensor laser-based detection methods that are easier to combine and efficient fusion.In this paper, visual images are firstly single tested using the visual sensors and through the visual images we can get the preliminary confirmation for the front goal by the multi-feature combination method. After the pretreatment for the collected visual images, the lane line markings are detected and extracted by using the improved Hough transform to determine the effective area for the forward vehicles detection, in this effective area, use the statistical methods of block region to search the shadow characteristics that at the bottom of the vehicles and find out the possible location of vehicles; use edge detection to exclude non-vehicles areas; use the symmetry principle of vehicle to further detect forward vehicles.And then use the horizontal scanning of the laser sensor in the one-dimensional direction to the forward vehicles target, first select the growth starting point, use the region growing method to produce the range images, judge whether the target is vehicle based on the distance images.It is found that the vision sensors and laser sensors are deficient, so we adopt information fusion of laser and vision to detecting target, then the vehicle validation function can be build when got the width and length of anterior target. The vehicle validation function can be used to build Bayes decision classifier basing on the minimum risk, then it can confirm the vehicle in the vision and fuse the information of vision sensors and laser sensors, result of experiment show that it is a good method to detect exactly the anterior vehicle. Finally, the paper track the vehicle that is detected before, predict the vehicle target location with Kalman filter in vision images, use the NMI characteristics to verify and confirm the predicted target, and then combined with the distance measurement function of the laser sensor to track the forward vehicle, when the distance between the home side vehicle and the forward vehicle reached the safety critical range, the system would give out an alarm signal to alert the driver to make a crash emergency treatment.