There are two elements in road traffic accident scene graph, base map and other elements in the site of traffic accident. The base map mainly includes traffic sings and traffic markings. Drawing traffic accident scene graph is a necessary link during disposing road traffic accident, while drawing the base map costs most of the time of drawing traffic accident scene. In order to save processing time of dealing with the traffic accident, the concept of road base map is proposed. The symbol design and data organization of the base map are analyzed. All this provides a new way to make the traffic accident scene graph.

Abstract-With the rapid increase of vehicle holdings, urban transport is facing a severe test. Driver-assistance systems (ADAS) can effectively avoid happening of traffic accidents. Real-time detection of moving vehicle based on vision has become a research focus of ADAS. A method of multiple features fusion for front-view vehicle detection is proposed in this paper. Firstly, in RGB color space, license plate position is located using color conversion and Otsu threshold segmentation method is improved to confirm the vehicle's candidate area. Secondly, geometrical characteristics of the license plate is adopted to eliminate distraction regions and to verify the extracted license plate. Finally, rear lamps of vehicle are detected and matched in candidate license plate regions, so vehicle area is further determined. The experimental results show that the proposed method in this paper can be used to detect vehicle in real-time and false detection rate is low. Keywords-rear-view vehicle; RGB color space; Otsu threshold segmentation I INTRODUCTIONDriver-assistance systems require real-time and reliably to detect vehicle and pedestrian around the vehicle driving direction, to remind the driver or vehicle to automatically take corresponding measures to avoid potential dangers. In ADAS, front-view vehicle detection is one of the ways to improve security, it can effectively avoid the rear-end collisions of traffic accidents due to lack of safe distance between vehicles.Vision-based vehicle detection method mainly consists of vehicle models [1], the optical flow methods [2] and feature methods [3]. In [4], Zeng zhihong established a 'U' model and a rectangular model, which were used to detect the long distance and short distance vehicle, respectively. A fine vehicle model, which has seven parameters, was built according to the features of car body [5] to search and match of vehicles by using genetic algorithms and energy function. Model-based vehicle detection has a great dependent on model, so that this scheme is not good for real-time detection. Vehicle detection based on optical flow can not detect slow or stationary vehicle. Feature-based vehicle detection method using significant features of vehicle, such as color, edge symmetry and vehicle's shadow, to segment vehicle area [6,7]. When there is a large impact on image from surroundings light and noise, single feature detection will be weakened, so multi-feature combination is able to improve the accuracy of vehicle detection.Detecting the candidate area of vehicle based on color feature is a popular method. Cheng et al.[8] used a new model for the color conversion, transformed the three color components to two components, one of which represents the color model of vehicle, and the other represents a color model of non-vehicle, the method is appropriate for detecting aerial images, which reduce the effect on billboards and buildings of the road. Yan et al. [9] made binary image processing by setting H and S components threshold, this man-made threshold does not have un...