Research On Road Information Recognition Based On Vision

The road information recognition based on vision is an important part in the field of driver assistance systems.The vehicle collects the road environment information by vehicular camera.Getting the lane’s position and the vehicle’s driving direction correctly is irreplaceable for driving safety and vehicle navigation.This thesis mainly focuses on lane detection and pavement arrow markings recognition,and eventually constructs the system for road information recognition.The major works of this thesis are as follows:1.Image pre-processing algorithm for the intelligent vehicle video.According to the characteristics of the video,making the pre-processing steps as follows,image resizing.setting the ROI(Region of Interest),image gray processing,bilateral filtering,image binarization by OTSU and image edge detection by Canny.With the pre-processing algorithm,the redundant information outside the lane is weakened and the useful information is enhanced.The image is more suitable for lane detection and pavement arrow markings recognition.2.Lane detection based on restricted Hough transform and dynamic ROI.Firstly,the algorithm restricts the area of the polar radius and polar angle,and then locates the lane with the restricted Hough transform.Secondly,in accordance with the last image lane’s information establishing the ROI,then realizing the lane tracking in the ROI.Finally,updating the ROI based on the result of lane detection in order to avoid invalid ROI and realize valid lane detection.3.Pavement arrow markings recognition based on SVM(Support Vector Machine)and template matching.Firstly,the algorithm locates the pavement arrow markings based on lane’s location and pavement arrow markings’ prior knowledge.Secondly,using the SVM algorithm and invariant moment features primarily identifies the pavement arrow markings.Finally,using the template matching algorithm to further detects mirror images.The MFC frame construction in Visual Studio 2012 and the C++ programming language and the OpenCV library are used for constructing the system of road information recognition.and a series of experiments have been performed.The result illustrates that the average accuracy of lane detection is 93.1%,the average accuracy of pavement arrow markings detection is 92.1%,the average accuracy of pavement arrow markings recognition is 98.0%,the average rate of lane detection is 17.3ms per frame,the average rate of pavement arrow markings recognition is 39.7ms per frame.The algorithm of this thesis has higher accuracy and faster processing speed,which can also satisfy the real-time requirements.