Lane Detection Algorithm Based On Adaptive Region Of Interest

With the improvement of road environment and increase of ownership, automobile contributes significantly to daily convenience and economic interest, however, with unignorable personal casualtiesand property damage. Thus, industries focus on automobile safety measures in order to lower traffic incidents. In recent years, intelligent driving assistance is highly valued by researchers, and comes with rapid development and deployment; we especially are grateful to computer vision, vehicle sensing and communication control, which make avoiding traffic incidents technically possible. In such aspect, lane detection, as the fundamental and critical technique, gains popularity among researchers. This article aims to conduct further researching and improving based on recent algorithms.For the purpose of faster and better information extraction, it urges us to select and compare different pretreatment approaches so as to choose a particular method with higher speed and sharper output. From previous work, we choose Sobel edge detection operator to detect lane boundaries, meanwhile, make related amelioration according to lane characteristics. Then an optimal threshold is proposed to use for improving the adaptability of the algorithm to lane environment. Finally, morphological method is introduced for image dilation, erosion and eliminating isolated noise point, therefore, we can get skeletonized lane image.In the process of lane detection and tracking, we purpose an algorithm, starting from linear prediction of adaptive ROI. This algorithm takes the advantages in stability and precision of Hough transform with linear restrictions which are used in determining initial point of lane. Then lane detection can be achieved via linear least square fitting, which means we can determine adaptive region of interest and reduce noisy interference from non-lane area. Specifically, tracking is based on the changes of lane direction, and either straight line or curve can be accepted, viz. this method would guarantee real-time operation and overcome the drawback came by fixed region of interest. Furthermore, it complies the detection and trace of both straight and curved shape, simultaneously.Experiments show that the algorithm can detect lane effectively with an accuracy to 98.35%, and work well on difficult cases such as test field, curve and broken lanes in particular. It has satisfactory performance on speed, accuracy and robustness.