Point Cloud Labeling via Obstacle Recognition and User Correction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Manual labeling of point cloud data is inefficient due to the non-intuitive nature of three-dimensional object characteristics in point cloud images, leading to low labeling efficiency and visual fatigue, requiring large amounts of correctly labeled data for training and verification of obstacle recognition algorithms.

Innovation Solution

A method and apparatus that utilize an obstacle recognition algorithm to provide an initial labeling result for point cloud data, allowing users to correct and update the labeling by adjusting borders and checking object type information, with the updated results used to train and optimize the recognition algorithm.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual labeling is used to ensure correctness of point cloud data, then labeling accuracy is improved, but labeling efficiency deteriorates

Engineering Contradiction:

Improvelabeling accuracyVSAvoidlabeling efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The obstacle recognition algorithm performs preliminary labeling of point cloud data before manual review, generating initial labeling results that reduce the amount of manual work required while maintaining accuracy through subsequent user correction

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The obstacle recognition algorithm acts as an intermediary between automatic labeling and manual labeling, providing initial results that guide manual correction efforts and reduce the overall manual labeling burden

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If entirely manual labeling process is used to ensure correctness, then labeling accuracy is improved, but time consumption increases

Engineering Contradiction:

Improvelabeling correctnessVSAvoidtime consumption

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary automatic labeling using the obstacle recognition algorithm before manual correction, reducing the total time required while ensuring correctness through user verification of the initial results

Inventive Principle:

Principle #10Preliminary action

3Productivity

If obstacle recognition algorithm is used to generate initial labeling, then labeling efficiency is improved, but labeling accuracy may deteriorate

Engineering Contradiction:

Improvelabeling efficiencyVSAvoidlabeling accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system implements feedback through user correction of the algorithm's initial labeling results, where user corrections are used to retrain and improve the obstacle recognition algorithm for future labeling tasks

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS10460172B2Method and apparatus for labeling point cloud data

Publication Date: 2019.10.29 BEIJING BAIDU NETCOM SCI & TECH CO LTD

AI summary

The present application discloses a method and an apparatus for processing point cloud data. The method of an embodiment comprises: recognizing an object in a to-be-labeled point cloud frame by using an obstacle recognition algorithm, to obtain a recognition result; presenting the recognition result as an initial labeling result of the point cloud frame; and updating the labeling result in response to a correction operation by a user on the labeling result. According to the embodiment, the speed and accuracy of point cloud data labeling are improved.