3D point cloud segmentation method based on attention network

An attention and network technology, applied in the field of computer vision, can solve the problems of insufficient capture of global context information, blurred object boundaries, lack of context information, etc., to facilitate identification and segmentation, enhance spatial differences, and achieve good segmentation effects.

Active Publication Date: 2019-08-09
XIDIAN UNIV
View PDF5 Cites 15 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The encoder-decoder convolutional neural network architecture used in this method can upsample the low-resolution features derived from the pooling layer to the input resolution, but because the upsampling layer lacks context information, the upsampled object boundary becomes vague and irreversible
[0005] Most of the existing technologies for 3D point cloud segmentation reconstruct the 3D coordinate system of the point cloud into a 2D coordinate system. The calculation process is cumbersome, and at the same time, the capture of global context information is insufficient, resulting in blurred and irreversible object boundaries after upsampling.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • 3D point cloud segmentation method based on attention network
  • 3D point cloud segmentation method based on attention network
  • 3D point cloud segmentation method based on attention network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0021] At present, the wide use of various 3D scanning devices has produced a large amount of point cloud data. At the same time, the application environment of 3D printing, virtual reality, and scene reconstruction has put forward various requirements for the processing of point cloud data. The processing of point cloud data, especially point cloud segmentation is the basis of various applications or task processing such as 3D reconstruction, scene understanding and target recognition and tracking. The segmentation results are beneficial to object recognition and classification, which is a hot research issue in the field of artificial intelligence Difficult problems have attracted the attention of more and more researchers.

[0022] Existing point cloud segmentation networks, such as PointNet, PointNet++, PointSIFT networks, etc., directly input 3D point cloud data into the network for training, but still do not make full use of global context information to learn better featu...

Embodiment 2

[0032] The 3D point cloud segmentation method based on attention network is the same as embodiment 1, obtains the AMNet model file described in step 2, specifically comprises the following steps:

[0033] (2.1) Build a training network: the training network uses an attention network (ANet for short) and a multi-scale module (Multi-scale group model, MSG for short) to form a point cloud segmentation network, called AMNet for short; the AMNet backbone network includes a MSG Module, an ANet branch network, three downsampling layers (Res model, Re for short), and three upsampling layers (FP model, FP for short).

[0034] Among them, the attention branch network (Attention Network, referred to as ANet) includes two transposition units, two multiplication units, one addition unit, two convolutional layers, and the convolution kernel size of each convolutional layer is 1×1 with a step size of 1.

[0035]Among them, the multi-scale module (MSG module) includes MSG1, MSG2, and MSG3. M...

Embodiment 3

[0052] The 3D point cloud segmentation method based on the attention network is the same as embodiment 1-2, the stretching formula described in step 3, specifically:

[0053]

[0054] s=1-(tanh(z 1 )) 2

[0055] z 1 =(ln((1+threshold) / (1-threshold))) / (-2)

[0056] Among them: f(z) represents the new z value obtained after the z value of the point cloud data of the test set is processed by the stretching formula, the threshold controls the slope s value of the linear function and the intersection point z of the linear function and the tanh function 1 The size of the value, the value range of threshold is [1 / 2, 1].

[0057] The larger the threshold, the intersection point z of the linear function and the tanh function 1 The farther the distance from the origin, the z value of the point cloud data of the test set is smaller than z 1 When , the tanh function is used to stretch, and the z value of the point cloud data of the test set is greater than or equal to z 1 Values...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a 3D point cloud segmentation method based on an attention network, and solves the technical problem of insufficient utilization of global context information in existing semantic segmentation. The method comprises the following steps: preprocessing 3D point cloud data set data; constructing an AMNet segmentation network based on an attention network and a multi-scale module, and training the training set data; performing stretching processing on test set data; carrying out network performance evaluation by using an AMNet model file, and optimizing the result by a D-KNNmodule and outputting a final segmentation result. According to the method, global context information is fully utilized through the AMNet, an accurate segmentation result is obtained, the space consumption of point cloud data processing is effectively reduced, the space cost is reduced, and meanwhile the accuracy of the segmentation result is improved. The method is used for 3D point cloud semantic segmentation.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a method for 3D point cloud segmentation, in particular to a 3D point cloud segmentation method based on an attention network, which is used for 3D point cloud segmentation. Background technique [0002] The dense and high-precision three-dimensional point coordinates on the surface obtained by the airborne LiDAR (Light Detection And Ranging) system by emitting and receiving laser pulses are called LiDAR point cloud data. The processing of point cloud data, especially point cloud segmentation is the basis of various applications or task processing such as 3D reconstruction, scene understanding and target recognition and tracking. It is a hot research issue in the field of artificial intelligence, and it is also a difficult problem. researchers' attention. [0003] Dalian University of Technology proposed a large-scale dense scene point cloud semantic segmentation...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06T7/11G06T7/136G06K9/62
CPCG06T7/11G06T7/136G06T2207/20081G06T2207/10028G06F18/24147
Inventor 焦李成李玲玲贾美霞李艾瑾吴兆阳丁静怡张丹郭雨薇唐旭冯志玺张梦旋
Owner XIDIAN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap