Binocular stereo matching method based on joint up-sampling convolutional neural network

Abstract

The invention discloses a binocular stereo matching method based on a joint up-sampling convolutional neural network. The method comprises the following steps: firstly, carrying out feature extractionon an input three-dimensional image pair by utilizing a two-dimensional convolutional neural network based on joint up-sampling; constructing an initial three-dimensional matching cost amount of thematching cost by splicing the features of the three-dimensional image pairs, further performing cost aggregation on the matching cost amount by adopting three cascaded three-dimensional convolutionalneural networks based on joint up-sampling, and finally obtaining a dense disparity map with sub-pixel precision by utilizing a regression method. Compared with an existing binocular stereo matching deep neural network, the convolutional neural network based on pyramid joint up-sampling is adopted in the decoding stage of the feature extraction and cost aggregation steps; and by fusing multi-leveland multi-scale context feature information, more detail textures can be effectively reserved in the up-sampling process, the calculation efficiency of the method is improved by adopting depth separable convolution with low calculation complexity, and the disparity map quality of binocular stereo matching is improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a binocular stereo matching method based on a joint upsampling convolutional neural network. Background technique [0002] Binocular stereo matching is a research problem that has attracted much attention in the field of computer vision. It has been widely used in various systems such as 3D reconstruction, automatic driving, autonomous robot navigation, and industrial inspection. In particular, current applications have an urgent need for real-time binocular stereo matching with high precision, high resolution, and large parallax. This undoubtedly poses a higher challenge to the computational efficiency and accuracy of the technology. In recent years, artificial intelligence based on deep learning technology has developed rapidly, and breakthroughs have been made in the fields of object detection, image classification, and speech recognition. Binocular stere...

AI Technical Summary

Problems solved by technology

Although the encoder can obtain rich semantic information by successively downsampling the spatial resolution, in the decoding process, the currently commonly used deconvolution upsampling tends to lose fine image structure information, resulting in parallax of object boundaries or small-sized objects Inaccurate prediction of outcome

Images