The “floor-wall” traffic scenes construction for unmanned vehicle simulation evaluation

A new framework to construct 3D traffic scenes for unmanned vehicle simulation evaluation is proposed in this paper. The “floor-wall” geometry is applied for traffic scenes construction process which consists of two stages. The first stage is road planes specification implemented by support vector machine (SVM) and Markov random field (MRF) using superpixels. In the second stage, control nodes of road boundaries are specified to construct background scenes. Foreground traffic elements are supplemented, which are assumed to stand perpendicularly to the road planes. New viewpoint images can be generated according to the virtual vehicles´ positions in model space. Based on the scene models, traffic incidents can be simulated by properly organizing the background scenes and the supplemented foreground traffic elements. Furthermore, the abilities of unmanned vehicles can be evaluated.