Depth Recovery from Single Monocular Image in 3D Reconstruction and Visual Precise Positioning

The depth information of the image is essential for accurate positioning. However, for the scale of the model, stricter standards are inevitable. This paper proposed Lower Layer Efficient Neural Network (LL-ENet) on the problems of requirements with light model, mostly using on the mobile devices. Firstly, The model reconstruct encoder by changing the convolution for less calculation, then classify the depth for efficient predictive ability. Finally, the result by experience shows that LL-EN et has a better effect on depth recovery with single monocular image in the outdoor scenes. Moreover the further experience shows that the proposed LL-ENet has better performance than other algorithms.