A Method of Depth Prediction Based on PDB-ConvLSTM
Yang Guanshui, Liu Licheng, Xiongzi Li · 2022 5th International Conference on Pattern Recognition and Artificial Intelligence (PRAI) · 2022
In order to solve the problems that motion features are ignored in the current deep learning methods for depth prediction and the coding module based on maximum pooling may lose details, a depth prediction method based on Pyramid Dilated Deeper ConvLSTM (PDB-ConvLSTM) is proposed in this paper. The method combines delated convolution with Convolutional LSTM(ConvLSTM) to learn the time information of image sequences and retains the details of image to a greater extent. Experiments on the KITTI standard dataset show that the proposed method improves the accuracy of percentage of right pixels of depth prediction by 5% on average compared with other existing depth prediction methods.