An Automatic Street Panorama Road Contour Extracion Method Via Multi-feature Fusion And Application Research

With the gradually development of information technology, the image-based street view technology has been widespread used in attractions show, path query, automatic navigation, etc. Among them, the image-based road contour detection method is the basic to realize the above applications. However, it is almost impossible to apply the traditional road extraction method directly on panorama. Therefore, the extraction of panorama road contour and lane markings has gradually become a urgent problems direction in this field.The main reason is that there is a significant difference between the street panorama and the traditional street image, especially the image deformation produced during the panorama mosaicing process. In allusion to the problems mentioned above, based on the characters of panorama and lane marking, the difference of color information between the urbanization road part and non-road part, texture direction and priori information, we propose an automatic road contour extraction method for street panorama via multi-feature fusion and an automatic lane marking extraction method by low-rank representation. The detail works list as follow:(1) Comparison and analysis of extraction methods for road contour is presented. We perform the experiment with OTSU, Genetic Algorithm and LASV. The experimental results show that these three kinds of method can not obtain satisfactory results. The main reason is because in the process of fusion of panorama, the road part is curved and deformed, which affect the color information in general method and model of the road.(2) We estimate road vanishing point by adopting Gabor wavelet and the Local Adaptive Soft-Voting algorithm(LASV). And then use the method proposed in this paper to extract the road contour of panorama. Considering the direction of panorama street road parts can along the way point to the vanishing point or road contour point, OCR(Orientation Consistency Ratio) characteristics was introduced to give the correct direction on the panorama road deformation areas to get a coarse road contour. When we get the preliminary result, we adopt the watershed method to optimize the coarse road contour of panorama which has no shadow effects on the road.(3) Based on the extraction of road contour, we can obtain a targeted area for the lane marking extraction. In order to extract the lane marking, this paper adopts the method based on low rank representation to separate the noise part which is lane marking part from the extracted image. The experimental results show that our novel method can extract contour and lane marking for complex street panoramic image automatically and precisely.In general, this paper realiz an automatic street panorama road contour extraction method via multi-feature fusion after analyze the traditional methods. The method can apply to different kinds of panorama and automatic complete the road contour and lane lines extraction.