Depth Estimation Method for Monocular Images Combined with Position Estimation

In this paper, we propose a monocular image depth estimation method based on ResNet and bitmap estimation. Traditional monocular depth estimation methods usually utilize convolutional neural networks (CNNs) for end-to-end training, but often require large amounts of labeled data. To overcome the problem of difficult data labeling, we use a competitive cooperative learning approach to jointly train depth estimation networks, camera motion estimation, optical flow estimation, and motion segmentation networks. We use ResNet as the backbone network for depth estimation, which achieves efficient feature transfer through residual concatenation. In addition, we introduce a bit-position estimation module to assist the depth estimation task by estimating camera motion and bit-position information. Experimental results show that our proposed method achieves good performance in the monocular depth estimation task and still achieves satisfactory results in the absence of labeled data.