Point cloud completion method based on visual angle generator

Abstract

The invention discloses a point cloud completion method based on a visual angle generator. The method comprises the following steps: firstly, preprocessing a data set by using an invisible point removal algorithm; secondly, designing a point cloud complementation network PV-Net based on a visual angle, wherein the network comprises an encoder E and a decoder D; wherein the encoder E is responsible for extracting a global feature f of an input incomplete point cloud, the decoder D is responsible for decoding the global feature f to obtain a complete point cloud, and the decoder adopts a visual angle-based strategy from part to whole and is realized through the visual angle generator. According to the point cloud completion method provided by the invention, an intermediate point cloud under an adjacent view angle is generated by fully utilizing view angle correlation between incomplete point clouds; the incomplete degree of the incomplete point cloud is gradually reduced by expanding the visual angle of the incomplete point cloud, so that the complete shape of the incomplete point cloud is deduced better, mutual supervision is performed by using the incomplete point cloud, a decoder is promoted to pay better attention to local details, and a better completion effect is achieved.

Description

technical field

[0001] The invention belongs to the technical field of computer vision; in particular, it relates to a point cloud completion technology based on a viewing angle generator. Background technique

[0002] In recent years, point cloud completion tasks have attracted extensive attention in the fields of computer vision and robotics. Raw point cloud data acquired by devices such as depth cameras or lidars are often sparse and incomplete due to limitations such as resolution or occlusion. Incomplete 3D objects greatly limit its application in real life, such as object detection, scene segmentation and other downstream tasks. Therefore, it is very necessary to predict or infer the missing structure from the partially observed residual defect cloud, which has very important potential and value in the field of 3D vision, especially in application scenarios such as autonomous driving and robotics.

[0003] Many works have attempted to solve the point cloud completion...

Images