Identification method of PointNet for complex scene

Abstract

The invention discloses an identification method of PointNet for complex scene. According to the method, a pyramid pooling module is used for optimizing the recognition capability of Point Net in a complex scene. The feature vector obtained after Point Net processing is accessed to a pyramid pooling module. The pyramid pooling module can improve the performance of open vocabulary object and filling identification in complex scene analysis. The capability of context aggregation based on different regions with global text information is leveraged. The global prior representation is effective forgenerating a high-quality result in a scene analysis task; for a complex scene, the pyramid parsing module provides good description for overall scene interpretation, the knowledge graph depends on prior information of scene contexts, and the pyramid pooling module can aggregate context information of different regions, so that the capability of obtaining global information is improved. Therefore, due to the addition of the pyramid module, tasks such as object classification, partial segmentation, semantic segmentation and the like in a three-dimensional space can have higher accuracy.

Description

technical field [0001] The invention belongs to the field of image retrieval and relates to a recognition method for optimizing PointNet for complex scenes. Background technique [0002] With the rapid development of the Internet and deep learning, we have done a lot of research on two-dimensional images, but for 3D point clouds, it is still a challenge for us. 3D point cloud data is an unordered set of points. Typical convolution structures require a highly regular input data format, so we need to convert these data into regular 3D voxel grids or image collections (such as views), and then Feed it to a deep network architecture. However, this data representation transformation makes the resulting data unnecessarily large. It also introduces quantization artifacts that can obscure the natural invariance of the data. The PointNet network directly takes point clouds as input and outputs class labels for the entire input or per point segment / part labels for each point of the...

AI Technical Summary

Problems solved by technology

Unobtrusive class scenes contain objects of arbitrary size. These details are also important for the recognition of complex scenes. Objects with fine details may not be resolved without the aid of prior information.

Images