Research On Detection Technology Of Obstacles In Front Of Construction Vehicles Based On Binocular Vision

The intellectualization of engineering equipment is a hot topic in current research.It requires that vehicles can work autonomously,such as completing operations,driving automatically,and avoiding obstacles.The critical premise of automatic driving is to be able to detect obstacles that affect driving and measuring the surrounding environment in real-time.At present,obstacle detection usually uses lidar and ultrasonic radar sensors.Such active detection methods will affect each other due to the signals between the devices.Compared with this kind of method,the binocular stereo vision has higher accuracy,cost performance,and real-time performance,and there is no interference between binocular devices.Therefore,the specific research in this paper is as follows:For the distortion of the binocular camera in the image acquisition process,the stereo correction algorithm is studied,and the "Zhang Zhengyou" calibration method is adopted to design experiments to calibrate binocular checkerboard photos.The internal and external parameter matrices with accurate accuracy are obtained by verifying the errors.Bouguet algorithm is used to correct the stereo image.At the same time,due to the influence of illumination and camera electromagnetic wave,noise interference,and uneven brightness of the picture,a histogram equalization combined with the bilateral filtering method is proposed to process the image.The accuracy of stereo matching is improved.The existing stereo matching algorithms are studied,among which the local stereo matching SAD(Sum of Absolute Differences)algorithm meets the real-time requirement,but the false matching rate is high.Therefore,this paper studies an improved stereo matching algorithm based on the characteristics of global matching.The accuracy is improved by constructing an energy function,and the computational complexity is simplified by dynamic programming.By comparing the results of the local matching and the improved algorithm,it is proved that the improved algorithm improves the matching accuracy,ensures the real-time performance,and has better parallax effect.The relationship between parallax and depth is used to construct a reprojection matrix to recover the target distance information.Experimental analysis shows that the improved algorithm also improves the accuracy of distance measurement.Research on obstacle detection method.The obstacle detection method based on the combination of disparity map and V disparity map is studied.Random Sampling Consistency(RANSAC)algorithm and Hough transform are used to fit the diagonal line of the road,which can effectively eliminate the interference of the road surface on the obstacle detection and then accurately detect the obstacle.At the same time,aiming at the problem of movement obstacle,the method of speed estimation and direction prediction of movement obstacles is studied.Through tracking and matching the images of two adjacent frames,the movement speed and direction are calculated by using the pose changes of the same points on the obstacle.Design outdoor simulation experiments to verify the reliability of the method...