Alvarez CVPR2010文献3D scene priors for road detection 中使用,kitti测试时也引用了
Another 3D cue is the layout of the scene. The layout is
analyzed using three major parts of the image: (1) sky pix- els, (2) vertical surface pixels and (3) ground pixels. With these 3D cues the road is limited to ground, non-sky image regions. Further, regions are avoided which are vertically orientated (i.e., buildings, vehicles, pedestrians or any other object present in the scene). The segmentation of the image in these 3D cues is computed by the method proposed by Hoiem et al. [15]. This method provides, for each pixel, a label and a confidence map for each class (Fig. 5). Hence, a pixel–wise road confidence map L is obtained considering those pixels which have higher probability of being ground areas than being vertical surfaces or being sky.
Road detection using scene layout is robust to different
types of asphalts, lane markings and pedestrian crossings. However, scene-layout for road detection may be sensitive to shadows as the algorithm uses superpixel segmentation.
利用图像中的3d信息
还原3d信息:Recovering surface layout from an image
Alvarez CVPR2010文献3D scene priors for road detection 中使用,kitti测试时也引用了