Study On The Vehicle Safety Pre-warning Method Based On Computer Vision

Because of the increase of the urban traffic and the irregulation of driver’sperformance, a large number of traffic accidents occur at presents. According tothe statistical results, drivers own main responsibility for major accidents.Vehicle active safety warning systems can process real-time analysis of roadvehicle environmental information, when there is a dangerous situation, earlywarning systems will give some warning signals to alert the driver to avoidtraffic accidents.The computer vision technology has been used for lane detection and laneobstacle detection, and a fore-warning system of decision-making has beenestablished by using motion models. The main work of this paper is summarizedas follows:A method on lane detection based on the significance feature of directionand color is proposed. By testing the commonly method of image segmentationbased on region and boundary-based, the difficulties and problems of lanedetection process analyzing has been obtained. Combined with the inherent ofdirection and color, the new method has been proposed, the lane detectionmethod used in this paper mixed the direction of lane feature maps and weightedcolor (white and yellow) saliency map. Simulation results show that the result ofthis algorithm has better robustness for road strong light, shadows, etc.A method on lane obstacle detection based on optical flow field is given.Because of the simultaneously movement among vehicle, background andcameras, optical flow method has been used. In order to reduce the interferencefrom moving target in other lanes, the picture has been divided into two regions，which can reduce the interference of the moving target of other lane. Thismethod owns a certain validity and versatility.A personalized warning decision method with adjustable parameters ismade. The sports model of vehicle displacement also been established, whichshows the value of velocity, acceleration and other related parameters of different motion state. Based on the motion models, the smallest followingdistance to avoid collision is calculated when the distance between two carsreduce. In the above model, driver’s reaction time and change parameters basedon road conditions has been considered. The model gives a personalized warningdecision, so this warning system can be more humane and higher reliability.The simulation results prove that the lane detection algorithm and obstacledetection algorithm proposed in this paper are more robustness and adaptable,and the warning decision method is more reliable.