Research On Lane Departure Warning System Of Speedway Based On Machine Vision

Nowadays, with the rapid development of social economy and the increasingly explosion in auto inventory, traffic accidents happen more frequently. The main reason is that the vehicle deviates from the lane and collides with the car on the side lane when drivers are out of consciousness. As a result, the auxiliary system of intelligent driving is needed, which can avoid lane departure accidents and provide safe driving.Lane departure warning system is one of the researches in the auxiliary system of intelligent driving, and the extraction and recognition of lane mark line takes an important role in this system. This dissertation has a deep research based on monocular vision in the judgment of driveway type and the algorithm of subsection extraction.To begin with, in order to eliminate the irrelevant information and enhance the testability of useful information, the driveway image sequences collected by camera were preprocessed. Then a division of the preprocessed image was made to distinguish the road area and an image only contains driveway mark line was obtained. The driveway mark line is refined to reduce the amount of data. The principles of extraction straight line are analyzed by the least squares, Hough transform and the median intercept. The optimal algorithm of extraction is then selected by experiments in this dissertation.Decomposed the driveway mark line into two part ULM and LLM and analyzed their characteristics. Adopted Hough transform computing the parameters of LLM and least square computing the parameter of ULM, summed up the judgment criterion of driveway type according to the relationship between slopes. With a large number of experiments in straight, left and right corner driveway image, the reliability of the criterion is verified.The driveway mark line was extracted by segment according to the judgment of driveway type. If the lane is straight, the average of parameters of LLM and ULM was used as the parameters of lane. If the lane is a turn, a parabolic model was established for ULM and extracted by the least square method, in combination with the parameters of LLM lane was marked in the image to verify the reliability of extraction algorithm. According to the extraction results of the former frame, the current search area of seeds is limited and the real-time tracking of driveway mark line is realized.At last, by introducing the principles of the most common current model of lane departure warning and analyzing its advantages and disadvantages, the driveway mark line equation and slope parameters are used to confirm the position and direction of the vehicles in the driveway, and a kind of lane departure warning model without calibration is established.