| As an emerging subject,traffic scene analysis plays a very important role in the automobile automatic driving system.The performance of traffic scene analysis under complex background has a direct impact on the development of video-based automatic driving technology.In this paper,a new type of traffic scene analysis is proposed,which is used for labeling each pixel in an image with its semantic category,and it aims at being widely applied to complex scenes with multi-calss objects.The system combines region-level feature analysis and sliding window detection to analyze traffic scene,and it mainly covers the following aspects:Based on the analysis of image superpixel,it is possible to accurately classify the background scene categories(such as roads,sky,trees,buildings,etc.)that have no fixed shape but relatively consistent texture information.We mainly select the following superpixel features:texture feature,color feature,appearance feature and geometric feature of the image subregion(size,shape,location,etc.).The object detection based on sliding window can obtain high detection precision of object scene categories(such as pedestrian,vehicle,non motorized vehicle,etc.).In this paper,the feature of object scene is described by fusing the LUV color feature and gradient feature in multi-scale space of an image,and an AdaBoost classifier is used to detect and recognize the object.The superpixel-based classification results of traffic image are fused with the multiple object detection results,and in order to improve the classification accuracy,the image context information is considered through the Markov random field.Thus,the framework of the scene analysis system is constructed.Finally,comparative experiments have been conducted on the challenging Sift Flow dataset and CamVid dataset,and the experimental results show that the proposed method can achieve better performance compared with other existing methods.Through the research in this paper,traffic scene analysis technology is enriched,and the practical process of automatic driving system is promoted. |
