A 3D target detection method that does not need to be processed after processing

A target detection and 3D technology, applied in the field of computer vision, can solve problems such as structural incompatibility, and achieve the effect of saving time and expense

Active Publication Date: 2022-08-09
ZHEJIANG UNIV
View PDF5 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, current mainstream 3D object detectors are not structurally adapted to this strategy

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
  • A 3D target detection method that does not need to be processed after processing
  • A 3D target detection method that does not need to be processed after processing
  • A 3D target detection method that does not need to be processed after processing

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0030] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be pointed out that the following embodiments are intended to facilitate the understanding of the present invention, but do not have any limiting effect on it.

[0031] like figure 1 As shown, a 3D object detection method without post-processing operation includes the following steps:

[0032] Step 1, in the target area, initialize K 3D candidate boxes and 1 object embedding feature. Both candidate boxes and object embedding features are learnable. As the training progresses, the object embedding feature encodes the general features of the object to be detected, denoted as E ∈ R 1×C , where C represents the feature dimension;

[0033] Step 2, using the existing feature extraction model PointNet adopted by VoteNet to perform feature extraction on the input point cloud samples. For details, please refer to "Deep Hough Voting for 3D Obj...

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 target detection method without post-processing operation, comprising: (1) initializing K 3D candidate frames and 1 object embedding feature; (2) performing feature extraction on input point cloud samples to obtain point features ; (3) Extract K 3D candidate frame features on point features; (4) Use object embedding features to screen and extract 3D candidate frame features to obtain K features; (5) Use self-attention model to make K features Exchanging feature information to obtain K proposed features; (6) Predicting K prediction results according to the proposed features, and training after one-to-one matching with the annotation information; (7) Using the 3D candidates of the K prediction results predicted in step (6) The frame replaces the K 3D candidate frames in step (1), and the feature proposal obtained in step (5) replaces the object embedding in step (1); repeat steps (3) to (7) multiple times to obtain detection results. The present invention can solve the problem of redundant prediction in the existing 3D target detector.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a 3D target detection method without post-processing operation. Background technique [0002] 3D object detection is a technology widely used in unmanned driving, indoor object detection, robot navigation, etc. The input of the 3D object detection task is point cloud data, and the output prediction results include the position of the 3D box, the category of the object in the 3D box, and the confidence of the 3D box. [0003] In recent years, the detection accuracy of 3D object detectors has been greatly improved. Typical works include: "DeepHough Voting for 3D Object Detection in Point Clouds", "A Hierarchical GraphNetwork for 3D Object Detection on Point Clouds", "H3dnet: 3d object" detection using hybrid geometric primitives", "Mlcvnet: Multi-level context votenet for 3d object detection". [0004] However, the predictions of these detectors have a lot of redundancy...

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 Patents(China)
IPC IPC(8): G06V20/64G06V10/774G06K9/62
CPCG06V20/64G06V2201/07G06F18/214
Inventor 刘子立蔡登徐国栋杨鸿辉何晓飞
Owner ZHEJIANG 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