Inpainting Model Corrects Broken Lane Geometry in Map Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital maps often contain incorrect or missing lane geometry data, which can impair route guidance and autonomous vehicle control due to inaccurate representations of road networks.
Innovation Solution
A machine learning model is employed to detect and correct lane geometry errors by using an inpainting model with a generator network and discriminator, processing representations through dilated convolution layers, and updating map databases with corrected lane line geometry based on quality scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional map data collection methods are used, then map coverage area is large, but lane geometry accuracy deteriorates due to missing or erroneous data
Solution Approach 1:
The patent introduces an inpainting model as an intermediary component that processes map data representations. This model includes a generator network with dilated convolution layers that fills in missing or corrects erroneous lane geometry data by learning from surrounding context and patterns in the map representation, thereby recovering lost information without requiring additional physical data collection
Solution Approach 2:
The patent replaces traditional mechanical/data collection methods for verifying and completing lane geometry with a machine learning-based inpainting system. Instead of physically traveling roads to collect and verify lane data, the system uses neural networks to automatically infer and correct lane geometry from existing map representations, substituting computational intelligence for conventional data gathering approaches
2Measurement precision
If manual correction of lane geometry errors is performed, then lane geometry accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent implements a self-service correction mechanism where the inpainting model automatically detects and corrects its own processing of map data. The generator network autonomously identifies missing or erroneous lane geometry and generates corrections without human intervention, while the discriminator provides automatic quality assessment. This eliminates the need for manual correction processes while maintaining high accuracy
Solution Approach 2:
The patent incorporates a feedback loop through the discriminator network that evaluates the quality of generated lane geometry corrections. The discriminator provides feedback signals that guide the generator network to improve its corrections, creating an iterative self-improving system that automatically refines lane geometry accuracy without manual intervention or time-consuming review processes
3Productivity
If simple inpainting methods are used, then processing speed is fast, but lane geometry correction accuracy deteriorates
Solution Approach 1:
The patent transitions from simple 2D image inpainting to processing map data in multiple representational dimensions. The system uses a multi-dimensional approach where map data is represented not just as images but also as vector data and structured formats, allowing the inpainting model to leverage geometric relationships, topological constraints, and semantic information across different data dimensions to achieve accurate corrections efficiently
Solution Approach 2:
The patent employs a composite architecture combining multiple neural network components (generator network with dilated convolution layers, discriminator network) and multiple data representations (image-based map data, vector lane geometry, structured attributes). This composite approach integrates the strengths of different methods and data types to achieve both high processing speed through efficient neural network computation and high accuracy through multi-faceted validation and correction
Data Source
AI summary
A method is provided to using a machine learning model to predict lane geometry where incorrect or missing lane line geometry is detected. Methods may include: receiving a representation of lane line geometry including a masked area representing an area of broken lane line geometry; processing the representation using an inpainting model including a generator to produce a corrected representation, where the masked area is inpainted to correct the broken lane line geometry, where the generator processes the representation through one or more convolution layers and at least eight dilated convolution layers to generate inpainting of corrected lane line geometry in the corrected representation; processing the corrected representation through a discriminator of the inpainting model, where the discriminator discerns a score reflecting a quality measure of the corrected representations; and updating a map database based on the corrected representation in response to the score satisfying a predetermined threshold.