Monocular depth estimation method based on CNN and Transform feature fusion

The invention belongs to the field of computer vision, and discloses a monocular depth estimation method based on CNN (Convolutional Neural Network) and Transform feature fusion, which comprises the following steps of: firstly, constructing a mixed depth data set, preprocessing the mixed depth data set, inputting a preprocessed picture into a network to extract a CNN feature map, and secondly, extracting a CNN feature map from the CNN feature map; a feature conversion module is used for obtaining a similar image feature map, the scale of the similar image feature map is adjusted, four Transform feature maps are obtained, then the two feature maps are input into a guide fusion module, four-level feature maps are obtained and input into a decoding fusion module, and a decoded output feature map is obtained; and finally, inputting the output feature map into a depth output module to generate a depth map, and carrying out loss value calculation on the depth map label and the depth map to obtain a trained model. A