Polygon Boundary Point Classification for Autonomous Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional maps and GPS systems lack the accuracy required for safe navigation of autonomous vehicles, and existing techniques for maintaining digital maps and identifying objects on these maps are insufficient for providing up-to-date and accurate data.
Innovation Solution
A segmentation model is employed to segment image data and identify objects on a map, generating boundaries for these objects. A second model predicts a confidence score for points along these boundaries, indicating their correct positioning or potential errors, allowing for human assessment in regions with low confidence scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional maps and GPS systems are used for autonomous vehicle navigation, then the system is simple to implement, but the accuracy of location data is insufficient (over 100 meters error)
Solution Approach 1:
The patent segments the map maintenance process into multiple components: automated segmentation models for boundary detection, confidence scoring systems for quality assessment, and selective human assessment for low-confidence regions. This segmentation enables high-precision location data (10 cm or less) by combining automated processing with targeted human verification, resolving the contradiction between accuracy and system complexity.
Solution Approach 2:
The patent introduces confidence scoring as an intermediary mechanism between automated segmentation models and human assessors. The confidence score determines whether human assessment is needed, creating a layered approach that achieves high accuracy without requiring all regions to be manually verified, thus balancing precision with manageable system complexity.
2Measurement precision
If human assessors manually trace all boundaries on maps, then the accuracy of map data is high, but the time and cost required are excessive
Solution Approach 1:
The patent applies partial action by having human assessors evaluate only the subset of boundaries with low confidence scores, rather than manually tracing all boundaries. The automated segmentation model handles high-confidence regions efficiently, while human assessors focus on problematic areas, dramatically reducing total assessment time while maintaining high boundary accuracy.
Solution Approach 2:
The automated segmentation model serves itself by generating confidence scores that automatically identify which regions require human assessment. This self-service mechanism prioritizes work efficiently, allowing the system to maintain high boundary accuracy without requiring exhaustive human evaluation of every map region.
3Productivity
If automated segmentation models are used for all map regions, then the productivity is high, but the reliability of boundary data decreases in some regions
Solution Approach 1:
The patent implements feedback through confidence scoring, where the segmentation model's own confidence levels determine whether human verification is needed. Regions with low confidence scores trigger human assessment, and this feedback loop ensures that reliability is maintained in critical areas while preserving high productivity in regions where the model is confident.
Solution Approach 2:
The patent applies local quality by treating different map regions differently based on their confidence scores. High-confidence regions receive automated processing only, maintaining high productivity, while low-confidence regions receive enhanced human review, ensuring reliability where needed. This localized approach optimizes both productivity and reliability across the entire map.
Data Source
AI summary
Methods and systems of classifying points on a polygon boundary. The method includes acquiring image data about a training region having a training object, generating a segmentation map based on the image data including a plurality of points and defining a training polygon boundary. The plurality of points are associated with a respective feature vector generated by the segmentation model. The method includes acquiring a ground-truth polygon boundary, determining a displacement indicative of a label for the given boundary point between the position and a closest edge from the sequence of edges, and training a second model. A training iteration includes using the position of the given boundary point, a feature vector associated with the boundary point, and a set of feature vectors of neighboring points for generating a predicted score and adjusting the second model based on a comparison between the predicted score and the label.