Neural network searching method and equipment for binocular vision matching

Abstract

The invention provides a neural network searching method and a device for binocular vision matching, and the method comprises the steps: obtaining training data which comprises a binocular image and corresponding parallax data; searching structures of a feature extraction network and a three-dimensional matching network in a binocular vision matching model in a unit-level search space and a network-level search space by utilizing the training data; optimizing a first weight parameter set, a second weight parameter set and a network weight according to the difference between the parallax data obtained in the search process and the parallax data in the training data until the search process converges; and reserving at least part of the operation according to the value of the first weight parameter set, reserving at least one path according to the value of the second weight parameter set, and obtaining an optimized feature extraction network and a three-dimensional matching network basedon the reserved operation and path.

Description

technical field [0001] The invention relates to the technical field of neural network search, in particular to a neural network search method and device for binocular vision matching. Background technique [0002] Humans have the ability to observe, reconstruct and understand the three-dimensional world through binoculars. Accurate perception and reconstruction of scenes is crucial for human decision-making. Binocular vision matching (Stereo matching) is committed to giving computers the ability similar to human eyes, and its research goal is to calculate the disparity map (disparity map) from the two-dimensional color images obtained by the binocular camera. After obtaining the disparity information, the depth information and 3D information of the original image can be easily obtained according to the projection model, so this technology has a wide range of applications in the fields of 3D scene reconstruction, robotics, and automatic driving. Binocular vision matching is ...

AI Technical Summary

Problems solved by technology

[0005] In addition, in the binocular vision matching method based on deep learning, the cost of 3D matching needs to repeatedly use a large number of floating-point operations, and cascade layers must be used to instantiate these calculations to ensure the effect, resulting in a large number of calculations. power consumption

Images