A Novel Lane Detection Algorithm Research For Severe Complex Conditions

Since 1980s,intelligent vehicle self-navigation technology based on computer vision has become a foremost researching topic in the field of vehicle aided and self-driving technology.Lane detection technology is the one of key parts of vehicle self-navigation technology.However,in the environment of unstructured road(such as rural or mountain road)and shadow,there are much weakness on the technology of lane detection,such as low-rate of recognition,poor real-time performance and vulnerance of lightness.In order to overcome the shortcomings above,this thesis focuses on the low recognition rate,influence of road shadow and inaccurate fitting for complex shadow conditions.Under the premise of guaranteeing the real-time performance of the system,the research is carried out according to image preprocessing,feature extraction,image segmentation and tracking and recognition.The main work of this thesis is showed as follow:(1)The image is segmented by support vector machine algorithm,the neural network algorithm and the improved OTSU algorithm,so that the system can successfully separate the road area from the road image with complex shadows.This thesis designs experiments of these three methods and compares their results.(2)The vertical projection method is used for recognize the different type of road direction.According to the different type of road direction,this thesis designs the different models respectively.The extracted road edges are fitted to mathematical function by least square method.However,in complex road conditions,many noise points may appear on the road edge line,which will affect the function fitting results.Therefore,this thesis focuses on emphatically the method of LMedSquare for improving the accuracy of least square method.(3)Kalman filter algorithm and particle filter algorithm is introduced to track and predict the road parameters specifically.Moreover,the improved Kalman filter designed by this thesis is used for tracking and predicting of road parameters.Experimental results show that the proposed algorithm is robust to unstructured roads in mountainous areas,effectively avoiding the interference of shadows and noise,which can still play an effective role in the absence of road line conditions.