Curvature feature recurrent neural network-based three-dimensional target identification method

Abstract

The invention relates to an image identification technology, and provides a curvature feature recurrent neural network-based three-dimensional target identification method for the problem of image noises in a three-dimensional target identification process in order to effectively describe features of a three-dimensional target under different view angles. The method comprises the steps of firstly,obtaining a joint curvature of a target three-dimensional model by calculating a local average Gaussian curvature and an average mean curvature of the target three-dimensional model, forming a curvature sketch of the three-dimensional model by extracting a local maximum value of the joint curvature, and generating a 360-degree two-dimensional image sequence by utilizing transmission projection transform to serve as an input of a trained recurrent neural network; and secondly, obtaining an identification type with a maximum correct probability by utilizing a softmax function in a softmax layerthrough utilizing a bidirectional recurrent neural network (BRNN) as a three-dimensional model multi-view sequence feature learning method. According to the method, common features of the three-dimensional target and a two-dimensional image can be automatically extracted; and relatively good robustness and relatively high target identification rate can be kept under image noise conditions.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a three-dimensional object recognition method based on a curvature feature recursive neural network. Background technique [0002] 3D object recognition refers to the process of automatically detecting, locating, and recognizing the specified object pattern from any given 2D image scene, which is one of the key issues in computer vision research. With the continuous development of computer vision technology, 3D object recognition is more and more widely used in industrial inspection, augmented reality and medical imaging and other fields. However, due to factors such as illumination changes, image noise, and object occlusion, it is difficult to extract the common features of 3D objects and 2D images from different viewing angles, which has become an urgent problem to be solved in 3D object recognition. [0003] The key to 3D target recognition is to find the 2D represe...

AI Technical Summary

Problems solved by technology

The method based on artificial marking points needs to manually initialize the feature points in the two-dimensional image. Due to the need for human interaction, such methods are not repeatable; the method based on geometric features achieves the goal by extracting information such as the midline skeleton and contour shape of the target. recognition, but the recognition effect of such methods is poor when there is noise in the image; methods based on deep learning use deep neural networks to fuse low-level image features into high-level features with semantic information, which can well solve three-dimensional The image noise problem of two-dimensional images in the target recognition process, but the commonly used deep convolutional neural network cannot express sequence attributes, and cannot effectively describe the characteristics of three-dimensional targets under different viewing angles

Images