Fig. 1. Principle based on binocular reconstruction method

Fig. 2. Depth of field estimation and loss function based on left and right views

Fig. 3. Principle of multi scale unified operation

Fig. 4. Flow chart of our network

Fig. 5. Evaluation indicators of the four methods (KITTI dataset). (a) AbsRel; (b) RMSE

Fig. 6. Depth estimation results on the KITTI dataset. (a) Original image; (b) true value; (c) method of Ref.[14]; (d) method of Ref.[17]; (e) method of Ref.[20]; (f) our method

Fig. 7. Evaluation indicators of the four methods (Make3D dataset). (a) AbsRel; (b) RMSE

Fig. 8. Depth estimation results of outdoor roads.(a) Original image; (b) estimated result of depth of field