3D Object Annotation Using Point Cloud Projection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for 3D object image annotation in machine learning are inefficient and costly, as they require manual annotation and lack accuracy in determining the actual location of objects in a 3D space.

Innovation Solution

A method and apparatus that use a combination of image sensors and 3D space sensors to obtain point cloud data, project 3D annotations onto a 2D image, and perform image recognition to determine parameter information for accurate movement control of mobile platforms, such as autonomous vehicles, by generating control instructions based on the annotation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If manual annotation methods are used for 3D object image annotation, then annotation accuracy can be maintained, but annotation efficiency is low and costs are high

Engineering Contradiction:
Improveannotation efficiencyVSAvoidannotation accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The system performs automatic 3D annotation by having the annotation model process point cloud data and images autonomously to generate bounding boxes and parameter information, eliminating the need for manual annotation while maintaining high accuracy through the synergistic combination of multiple data sources and modalities

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The annotation model serves multiple functions simultaneously: it processes both point cloud data and images, performs 3D bounding box annotation, extracts parameter information, and supports various object types, making it a universal solution that replaces multiple specialized manual annotation tasks

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Measurement precision

If only 2D image data is used for object annotation, then the annotation process is simple, but the accuracy in determining actual 3D location is insufficient

Engineering Contradiction:
Improve3D location accuracyVSAvoiddata processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system merges point cloud data from 3D sensors with 2D image data from cameras, combining their respective strengths to achieve accurate 3D location determination. The point cloud provides depth and spatial information while images provide texture and visual context, and their integration enables precise annotation that neither modality could achieve alone

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The system transitions from 2D image annotation to 3D spatial annotation by incorporating point cloud data that adds the depth dimension. This dimensionality change enables the annotation model to determine actual 3D locations, bounding box dimensions, and spatial relationships that cannot be obtained from 2D images alone

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Measurement precision

If 3D point cloud data is used for annotation, then 3D location accuracy is improved, but annotation cost and processing complexity increase

Engineering Contradiction:
Improve3D annotation accuracyVSAvoidannotation cost
Core Design Contradiction:
Measurement precisionVSEase of manufacture

Solution Approach 1:

The system replaces manual mechanical annotation processes with an automated annotation model that processes point cloud data and images computationally. This substitution eliminates labor costs associated with manual 3D annotation while maintaining high accuracy through the model's ability to automatically extract bounding box parameters and spatial information from the data

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3852064B1Object labeling method and apparatus, movement control method and apparatus, device, and storage medium
Publication Date: 2024.07.10 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3852064B1 patent drawingFigure 1~2
  • EP3852064B1 patent drawingFigure 3
  • EP3852064B1 patent drawingFigure 4~5

AI summary

Embodiments of the present disclosure disclose an object annotation method and apparatus, a movement control method and apparatus, a device, and a storage medium. The method includes: obtaining a reference image recorded by an image sensor from an environment space, the reference image comprising at least one reference object; obtaining target point cloud data obtained by a three-dimensional space sensor by scanning the environment space, the target point cloud data indicating a three-dimensional space region occupied by a target object in the environment space; determining a target reference object corresponding to the target object from the reference image; determining a projection size of the three-dimensional space region corresponding to the target point cloud data and the three-dimensional space region being projected onto the reference image; and performing three-dimensional annotation on the target reference object in the reference image according to the determined projection size. By performing three-dimensional annotation on the target reference object in the reference image according to the target point cloud data, the annotation accuracy may be improved, an actual location of the target reference object may be accurately determined, and annotation does not need to be performed manually. Embodiments of the present disclosure can thus improve the annotation efficiency, and reduce the cost of annotation.