Segmented Line Object for Image Annotation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional methods for creating ground truth data for machine learning algorithms, such as image labelling for seat belt routing detection in vehicular applications, are inefficient and laborious, making it difficult to generate training data quickly and accurately.

Innovation Solution

A method involving a visual output component where a segmented line object is established by creating data points and labels simultaneously, allowing for efficient tracing of path objects within an image, with interactions determining visibility and positioning within threshold distances, thereby speeding up the annotation process.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If traditional labelling methods (image labelling, bounding box labelling, point labelling) are used for path objects, then the labelling process is simple, but they are ineffective for evaluating path routing

Engineering Contradiction:

Improvepath routing evaluation accuracyVSAvoidlabelling method applicability

Core Design Contradiction:

Manufacturing precisionVSEase of manufacture

Solution Approach 1:

The patent segments the path object into multiple line segments connecting sequential data points. This segmentation enables precise representation of the path's geometry and routing, allowing effective evaluation of path objects while maintaining ease of implementation through systematic point-by-point labelling.

Inventive Principle:

Principle #1Segmentation

2Manufacturing precision

If pixel-wise segmentation labelling is used, then path routing can be evaluated, but the process is laborious and time consuming

Engineering Contradiction:

Improvepath routing evaluation accuracyVSAvoidannotation speed

Core Design Contradiction:

Manufacturing precisionVSProductivity

Solution Approach 1:

The patent replaces the mechanical pixel-by-pixel labelling process with a point-based coordinate system approach. By defining paths through discrete data points connected by line segments, the system achieves accurate path routing evaluation while dramatically reducing the time and effort required compared to pixel-wise segmentation.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Productivity

If segmented line objects with simultaneous data point and label creation are used, then annotation speed increases, but system complexity increases

Engineering Contradiction:

Improveannotation speedVSAvoidinteraction component complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent merges the creation of data points and their associated labels into a single simultaneous operation. This integration allows annotators to define path geometry and semantic information in one unified process, increasing annotation speed while the modular interaction component design keeps system complexity manageable.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentEP4138046A1Method and system for creating training data

Publication Date: 2023.02.22 APTIV TECHNOLOGIES AG

Figure 1
Figure 2
Figure 3~4

AI summary

A vehicle can be fitted with a visual seat belt sensor that detects whether a seat belt is worn when a seat is occupied. Such systems can be implemented by using artificial neural networks or machine learning algorithms, which requires a large amount of training data. A method for preparing training data includes providing an image containing a path object. Starting at a first end of the path object, a segmented line object is established, the segmented line object consisting of a plurality of data points and line segments. Data points and labels are created simultaneously.