Research On The Embedded Lane Departure Warning System Based On Machine Vision

The driver error or distraction always causes the vehicle off track on the freeway,which becomes an important factor for major accidents.The lane departure warning device will give an alarm when the car deviates from the origin lane and the traffic accident due to lane departure will be greatly reduced.The method based on machine vision is used for the research of the embedded lane departure warning system and the main research work as follows:(1)The testing platform is designed for the embedded Linux system.A clear road image is got by utilizing the wide dynamic range and high light suppression function of the SONY EFFIO-P simulation camera.Image data from the TVP5150 video decoding module is transmitted to the S5PV210 processor based on the ARMV7 architecture and the computing result is displayed on LCD.A buzzer sending the warning signal is controlled by the processor.(2)Each device driver is accomplished in the embedded Linux system.Firstly,the work principle of each device is introduced.Then,the selective analysis of their driven architecture in the embedded Linux system is given.At last,the driven development process is designed.On the basis of accomplishing the driven development,Applications are tested to realize the beep at different frequency,collect image data from TVP5150 and display on LCD.(3)The noise interference from different driving environments is eliminated by image preprocessing.According to the mathematical model of the lane cross-section,the correlation filter operators with different scales are used to enhance lane markings in the local region of the images under different scenes.Moreover,the background noise is further reduced by making use of the top-hat filter and lane markings under different environments are extracted by the maximum variance threshold segmentation.(4)Achieving the rapid and accurate lane recognition and establishing the effective warning mechanism.Firstly,the lane of the initial frame is accurately extracted by Hough transform from the image preprocessing result in the paper.Then,the lane boundary fitting is rapidly done based on Box Plot and Random Sample Consensus for the tracking detection.Furthermore,during the lane tracking phase,the database of lane parameters is established,and current lane parameters are matched with those in the database to determine the confidence and the database gets update dynamically,thus detecting lanes under different typical scenes.On the basis of getting lane parameters,the warning decision based on the angle offset and the rate of distance offset is established,and its effectiveness is verified.The experiment under the embedded system platform shows that the embedded lane departure warning system has the advantage of real time,accuracy and stability.