Research On The Lane Line Detection And Vehicle Detection In Drivable Area Based On Vehicle Visual Environment

With the development of intelligent transportation system,the cost reduction of image vision sensor and the improvement of shooting accuracy,the development of automatic driving and assisted driving system which take the actual road traffic image as the processing object is more and more rapid.As a key technology of automatic driving and assisted driving systems,lane line and vehicle detection technologies have become a research hotspot for scholars and enterprises at home and abroad.There are many lane line and vehicle detection algorithms available today,and they are constantly being optimized.Due to the complexity of traffic scene,different weather conditions and road surface shadows make it difficult to extract lane lines.The accuracy of vehicle detection is also affected by the diversity of vehicles in traffic scenes and the variability of illuminating environment.In view of the above problems,this thesis studies two key technologies of lane line and vehicle detection.This thesis proposes a lane detection algorithm under illumination invariant conditions.First,the algorithm uses the texture feature to extract the vanishing point of the road image through the local voting algorithm,and obtains the region of interest for lane line detection;then,the illumination invariant grayscale image of the region of interest is obtained through the chromaticity space transformation based on the sensitization function of camera and the reflection characteristics of the object surface;Finally,the target area of the lane line detection in the illumination invariant gray image is extracted by using the characteristics of the lane line in the road image,and the lane line detection result is obtained by detecting the lane line in the target area by the Hough transform based on the polar angle constraints.Considering one of the main security threats of the other vehicles in the drivable area,on the basis of lane line detection,this thesis proposes a vehicle detection algorithm in the drivable area.At first,the algorithm uses the characteristics of the vehicle bottom shadow to be darker than other areas of the road under any illumination condition,and the vehicle shadow in the vertical gradient intensity image of the road image is extracted initially by the gradient intensity threshold.Then,the vehicle shadows in the vertical gradient intensity image are further screened by using the vehicle width morphological method.At end,the drivable area is obtained by lane line detection results,and the detection of vehicles in the drivable area is realized by threshold method.In this thesis,the proposed lane detection and vehicle detection algorithm are tested on the dataset Alvarez dataset,iROADS dataset and the actual road image dataset,which contain the actual traffic scene images taken under different weather and illumination conditions.The experimental results show that the lane line and vehicle detection algorithms proposed in this thesis have higher accuracy in different weather and different illumination conditions,and have good robustness.The problem of low recognition rate of lane line caused by the change of illumination condition and the shadow of road surface is solved.The influence of the accuracy of vehicle detection caused by the diversity of vehicles in the traffic scene and the variability of the illuminating environment is avoided.