Lane Marking Detection Algorithm Research Under Complex Environment Based On Computer Vision

Intelligent driving technology is the inevitable trend of future automobile development.Environmental perception is the foundation of intelligent driving and lane marking detection is a important part of environmental perception.It is difficult to extract the lane marking feature under complex environment and the dashed curve fitting is challenging.To solve the problem above,the following aspects are researched in this thesis.First of all,the Region of Interest(ROI)is determined and the color space of the ROI is converted.To solve the problem of lane marking feature extraction under complex environment,a split window threshold segmentation algorithm based on histogram of oriented gradients and support vector machine is proposed in the detection period.In the lane tracking period,a reduplicated threshold segmentation algorithm based on Otsu algorithm is proposed to extract the lane marking feature.The experiment shows that the algorithms proposed above segment the image better than other segmentation method with a better adaptability.Secondly,a lane marking detection algorithm based on lane width match is proposed.After the candidate feature point extraction and filtering the fake feature points,a field based least squares curve fitting algorithm is proposed,which can determine the fitting method adaptively according to the distribution of the feature points under different environment.To solve the problem of lane marking missing and dashed line fitting,a matching algorithm based on lane width is proposed.The lane width feature is extracted to optimize the curve fitteing.The experiment shows that the curve fitting algorithm proposed above gets a better accuracy than other fitting methods with a good real-time,and it can be applied to more scenes.Finally,a lane tracking method based on dynamic ROI is proposed.By determine the ROI dynamically according to the the similarity between the adjacent frame,the robustness of the algorithm is improved.Results of experiment shows that the algorithms proposed above can process a frame within 17 millisecond with an accuracy of 94%.The lane marking detection algorithm proposed in this thesis can be applied to complex road environment.It is of a certain theoretical and practical value for the development of lane marking detection system and environmental perception technology of intelligent vehicle.