A scene flow estimation method based on fusion segmentation and redistribution for autonomous driving

Abstract

Abstract A novel approach is presented here to solve the problem of motion occlusion and motion edge blurring in the existing scene flow estimation. First instance segmentation and superpixels are combined to segment the target and other regions in fusion segmentation. The pixels in each block are then redistributed by the optical flow to ensure the motion of pixels in the subblocks is consistent. Moreover, the 3D motion of subblocks with sufficient pixels is estimated by the energy function, and the others are considered outliers. Finally, the Driving and the KITTI benchmarks are used to evaluate the proposed method. The results demonstrated that the fusion of segmentation and redistribution is positive for the estimation, and this method outperforms the other state‐of‐the‐art methods both qualitatively and quantitatively.

Keywords

• autonomous driving
• region growth
• scene flow estimation
• spatial motion estimation
• superpixel segmentation