Lane Auto-Labeling with Spline Fitting for Consistent Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing labeling processes for training data in artificial intelligence models are time-consuming and lack uniform quality due to variations in labeling proficiency among individuals and tasks, leading to inconsistent results.
Innovation Solution
A method and server system that provides auto labeling assistance by allowing labelers to select suitable lane detection models and generate spline models using semi-auto labeling and line pose regression techniques, optimizing spline points through meta-heuristic algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual labeling tasks are performed by human labelers, then labeling can be completed with current technology, but the process is time-consuming and quality is inconsistent
Solution Approach 1:
The system enables self-service labeling by allowing the labeling task to service itself through automated spline model generation and regression. The computer automatically performs lane detection, generates spline models, and optimizes parameters without requiring continuous human intervention, thereby improving both speed and consistency.
Solution Approach 2:
The patent replaces the mechanical human labeling process with an automated computer-based system that uses lane detection models, spline curve generation, and regression algorithms. This substitution eliminates human variability and significantly increases labeling speed while maintaining consistent quality standards.
2Measurement precision
If multiple different lane detection models are applied to target images, then more accurate lane detection results can be obtained, but the complexity of the labeling system increases
Solution Approach 1:
The system performs preliminary actions by pre-processing target images through multiple lane detection models before the actual labeling task. By detecting lanes in advance and generating preliminary spline models, the system reduces the complexity of the subsequent labeling process while maintaining high accuracy.
Solution Approach 2:
The patent segments the labeling process into distinct stages: lane detection using multiple models, spline model generation, and regression optimization. This segmentation allows each component to be independently optimized and managed, reducing overall system complexity while achieving high detection accuracy.
3Loss of time
If semi-auto labeling with spline model generation is used, then labeling time is reduced, but the complexity of the automated system increases
Solution Approach 1:
The system applies partial automation by generating spline models automatically for portions of the labeling task while allowing human review and adjustment. This partial action approach reduces labeling time significantly without requiring a fully complex automated system, achieving a balance between speed and manageable complexity.
Solution Approach 2:
The spline model serves as an intermediary between raw lane detection results and final labeling output. This intermediary structure simplifies the automated system by providing a standardized intermediate representation that can be easily generated and adjusted, reducing overall system complexity while maintaining fast processing.
Data Source
AI summary
A method for providing auto labeling assistance service is disclosed. The method includes steps of: (a) providing a labeling interface to a labeler terminal, to thereby cause a labeler to select a specific lane detection model suitable for a target image among a first lane detection model to an n-th lane detection model by referring to at least one lane detection result among a first lane detection result to an n-th lane detection result, by the labeling interface displayed on the labeler terminal; and (b) in response to acquiring first point information to k-th point information on a specific lane selected to label the specific lane among lanes located in the target image, causing a semi-auto labeling module to generate a first spline model having m spline points corresponding to the specific lane and to display the first spline model, to thereby provide the auto labeling assistance service.