Research On Lane Recognition Based On Vision

With the rapid development of automobile industry,automobile brings convenience to people,but also causes potential dangers.Automobile safety has become an international topic.As a research hotspot in automobile field,intelligent vehicle driving assistant system can effectively improve vehicle safety and reduce traffic accidents.Lane recognition based on vision is an important part of intelligent vehicle assistant driving system.It plays a key role in adaptive cruise control system,lane departure warning system and lane keeping assist system.When the vehicle deviates,an effective lane recognition system will warn and remind the driver to correct or automatically adjust the driving behavior immediately to ensure the safe driving.Therefore,the recognition of the front lane and the front vehicle targets on this lane based on visual information is studied.(1)Preprocessing the road image.Firstly,the region of interest(ROI)is extracted from the road image,and 1/2 of the region below the image is taken as the ROI.Then,several gray processing methods including component method,maximum method,average method and weighted average method are analyzed,and the weighted average method is adopted in the paper.Secondly,the mean filter and median filter are realized,and the improved median filter method is used to filtered for the preprocessed image.Otsu method is used to extract lane contour information for binary image.Finally,three kinds of edge detection operators are compared,Canny operator is used to extract the edge of the image.(2)Research on Lane recognition.Considering that hough transform may only recognize the lane on one side,hough transform with polar radius and polar angle constraints can detect the location of left and right lanes accurately.By translating the detected left and right position straight lines,the upper and lower boundaries are established,and the complete lane feature points are included.The lane feature points of the lane constraint area are extracted according to the lane width restriction conditions and color jump characteristics.Finally,the least squares method is used to fit the feature points which meet the limit conditions of lane width and the color jump characteristics.The experimental results show that the lane recognition algorithm in the paper can identify straight,curved and virtual lanes well.(3)Recognition of the front vehicles on this lane based on HOG features and SVM.It mainly includes two parts: the hypothesis of the possible vehicle areas and the verification of the vehicle areas based on HOG features and SVM.Considering the computational speed of the algorithm,the possible search area of the vehicle on this lane is limited.For this area,binary operation and morphological operations of corrosion and expansion based on the shadow underneath the vehicle are used to find the hypothetical area where the vehicle exists.The vehicle sample library is constructed,the HOG features of vehicle samples and non-vehicle samples are extracted to train the SVM model.The test accuracy of the model is 98%.Finally,the HOG feature of the vehicle hypothetical area is counted to input into the SVM model to verify whether it is a vehicle or not.The results show that the recognition method of the front vehicle target on this lane based on the HOG feature and SVM of the shadow underneath the vehicle can recognize the vehicle target well.