Research On Lane Detection And Recognition Algorithm Under Complex Road Image

With the development of science and technology,the number of motor vehicles is also increasing,which induce many people death in the traffic accidents caused by the careless or inattention driving every year.Lane Departure Warning System can effectively reduce the occurrence of traffic accident;the line detection technology which can apply to a variety of complex road lane is the most critical part of the system,which related to the success of the system.But there are still two main problems for the detection of the lane marking accurately,the first problem is that when the road image is relatively complex,how to improve the lane line detection robustness,the second problem is that when the robustness satisfies the system requirements,how to improve the real-time requirement of the algorithm.To solve above two problems,two real-time and effective lane mark detection algorithms are studied in this paper.After the complex road image is preprocessed,considering the problem that a single threshold cannot be used to segment the lane line feature points from the non-feature points,a new Otsu optimal threshold and image pixel coordinates is used in this paper,The local threshold is used to extract the lane feature points.It has been proved that the lane feature points can be extracted even for the uneven road and the road with shadow by this algorithm.At the same time,according to the inherent linear characteristics of the lane line,several Haar-like rectangular features are designed to train the weak classifier to detect the lane line feature points.The improved Haar-like rectangular feature is trained by the improved Adaboost algorithm,and the Haar-like weak learning algorithm is weightedly combined into a strong learning algorithm.Finally,the trained discriminant function is used to determine whether the point is belonging to the lane marking or not.Considering that there are still many noise points in the feature points by using the improved Adaboost algorithm,the post-treatment is added,which can decrease the error points for the strong classifier.In the procedure of lane fitting,the shape of the lane is determined firstly,if the lane is straight,the improved Hough transfer method is used to fit the line,otherwise,the weighted hyperbola curve model is used to fit the lane,which makes the fitting method is more adaptive to the real lane marking.