Vehicle Flow Extraction And Analysis Based On Automatic Lane Detection

Most proposed traffic parameter detector can not detect enough parameter information and has problem of installation and maintenance. Specially, in vehicle video sequences, the road image segmentation and lane detection methods are improved and vehicle detection based on automatic lane detection algorithm is performed in this paper. The research is significant both in computer vision and practical application field.The entropy maximization algorithm combined with Mean shift algorithm for road image segmentation is proposed in this paper. Firstly, the entropy maximization algorithm is used to segment the gray image for the initial segmentation area. In the updating stage, the parameters of the area is used in Mean shift algorithm as initial values. This method is faster than the Mean shift algorithm used alone and more precise than the entropy maximization algorithm used alone. The accuracy of this segmentation method is above 90%, which can satisfy the preciseness.To simplify road edge detection image, the method that using Mathematical Morphology on it is proposed. Then improved Hough transform is used to detect the lane according to its shape and color. At last virtual detection lines and areas are set on the image.Adapted background model and update algorithm and shadow detection method based on color space are used in this paper to improve the preciseness of vehicle detection. Firstly the background image will be established in real scene and keep updating with time. Moving vehicle will be extract by using Background subtraction. The number, speed and shape of the vehicle will be calculated. The background update algorithm is just used in interesting area to enhance the real time. According to the illumination and chrominance differences between moving shadow and vehicle, we use a shadow detection method based on color space to subtract the shadow area. Applying this method on the vehicle detection system, the precise is improved a lot.Research on vehicle flow extraction and analysis based on lane detection is made in this paper. Experimental results show that our method is effective with an average preciseness of 90% on road image segmentation, 93% on vehicle number calculation, 90% on vehicle catalog and 8 km per hour inaccuracy on vehicle speed detection. In the future, the research for comprehensive performance such as robustness, the accuracy, real time, etc, are the emphasis and direction.