Study Of The Front Obstacle Detection And Lane Detection

The research of the safety assistant driving system of intelligent automobile hasalready been the cutting-edge research, and it plays an important role in the guarantee oftraffic safety. This article mainly discusses the detection of obstacles in front of the carand the detection and fitting of lane line. To deal with the image through characteristicsof lane line and obstacles. Using Hough transformation method to detect lane line andfitting segment of lane line. Using the shadow method to detect vehicles, and then usingthe exact characteristics of the vehicle to determine the position, then using the Kalmanto predict position of the obstacle, and finally through distance measuring formula todemarcate the camera parameters，measuring the distance between the vehicle andobstacles through the secondary camera calibration parameters. The main contents ofthe article are as follows:(1) Firstly, collecting color data image by the CCD camera and proceeding thecolor image in grayscale, then filtering the image. Using edge detection operator toenhance lane edge, and then stepping on the image binarization of the image, after thatusing adjacent method to remove noise,and finally through DLD algorithm to furtherscreening lane area.(2) Using Hough transformation method to detect lane line, for the near regionusing the method of after both sides of the first intermediate.for far field using a regionof interest for hypothesis testing lane to reduce the Hough transform computation.Determination of driving cars as to whether pressure line can be carried out after thecompletion of detecting lane line in the near zone, with the method of fitting a straightline in the near region and parabola fitting in the far region.(3) Extracting the interesting part of the vehicle detection area to improve the real-time of the detection. Using adaptive dual-threshold detection method to detect thevehicle bottom shadow in the interesting region, then using Sobel to enhance thevertical region of the vehicle to improve detection accuracy. If the mutational mean grayrow exists vehicle obstacles, setting up an interesting identification region to establish arough calibration of the position of the vehicle, determining whether there is a vehicleby means of a partial entropy in the vehicle candidate region. Enhancing the verticaledges of the ROI image through the characteristics of the vehicle, and then calculatingsymmetry measure in the ROI region, the maximum peak is the center of the vehicle, to achieve precise position of the obstacle. Using Kalma to predict vehicle position toimprove the real-time.(4) This article introduces the basic theory of camera calibration, the method ofranging transformation deduced camera calibration. Using the quadratic calibrationmethod and selecting four calibration points artificially to analyze results rangingabsolute and relative errors. Using quadratic and cubic fitting for distance and using thecoordinates to correct calibration points, but it is difficult to meet the accuracyrequirements. Analyzing the influence to the results through the accuracy of the imageitself, and that indicates the second calibration method has good performance.