Abstract: For the accuracy and robustness of the existing scene flow estimation models toward to the complex background, weak-rigid motion and motion occlusion, this paper proposed a 3D scene flow computing method based on the optimized layering and segmentation of depth image. Firstly, the depth information of the moving objects and background in the image sequence are acquired by using the initial optical flow to optimize the layering and segmentation of the depth image. Secondly, the scene flow is iteratively computed by employing the coordinate descent method based on the result of depth layering. Finally, the test image sequences of RGBD, Middlebury and SRSF databases are chosen to evaluate the accuracy and reliability of the depth layering and scene flow estimation of the proposed method. The comparison results indicate that the proposed method has the better accuracy and robustness for scene flow computing under the complex background, weak-rigid motion and motion occlusion.