Lane Boundary Annotation Using Aggregated LiDAR Polyline Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital map annotation methods for autonomous vehicles are inefficient, requiring manual correction of discontinuous lane boundary polylines generated from limited LiDAR scans, which increases annotation burden and reduces map accuracy.
Innovation Solution
A unified framework and tooling that extracts features from sensor data using a neural network to generate globally consistent polylines, allowing for automated semantic object generation and reducing the need for manual annotation by using convex hull algorithms and aggregation functions to create continuous road geometry representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual correction methods are used for lane boundary polylines, then annotation accuracy can be improved, but annotation burden and time consumption increase significantly
Solution Approach 1:
The system performs preliminary automated generation of lane boundary polylines using neural networks and convex hull algorithms before manual annotation. This preliminary action creates a draft annotation that is then refined through manual correction, significantly reducing the overall annotation time while maintaining accuracy.
Solution Approach 2:
The system enables self-service annotation by allowing annotators to automatically generate lane boundary polylines through convex hull algorithms based on selected road geometry points. The system automatically computes the polylines without requiring manual point-by-point drawing, reducing annotation burden while maintaining consistency.
2Device complexity
If limited LiDAR scans are used for polyline generation, then data processing complexity is reduced, but polyline continuity and global consistency deteriorate
Solution Approach 1:
The system merges multiple overlapping LiDAR scans and their corresponding feature maps into a unified representation. By combining information from multiple scans, the system generates continuous and globally consistent polylines that span across the entire map area, overcoming the limitations of individual scan coverage.
Solution Approach 2:
The system introduces an intermediary aggregation process that combines features from multiple LiDAR scans before generating polylines. This intermediary step ensures that the resulting polylines are continuous and consistent across scan boundaries, bridging the gaps that would exist if scans were processed independently.
3Productivity
If automated neural network feature extraction is implemented, then annotation efficiency improves, but dependency on machine learning models increases
Solution Approach 1:
The system replaces manual mechanical annotation processes with automated neural network-based feature extraction. The neural networks automatically identify and extract road geometry features from sensor data, eliminating the need for manual feature identification and significantly improving annotation efficiency.
4Ease of operation
If convex hull algorithms are used for semantic object generation, then manual annotation burden is reduced, but geometric precision requirements increase
Solution Approach 1:
The system enables self-service semantic object generation by automatically computing convex hulls from selected road geometry points. Annotators simply need to select key points, and the system automatically generates the semantic objects with precise geometric boundaries, reducing manual effort while maintaining high precision.
Data Source
AI summary
Provided are a system and methods for a unified framework and tooling for lane boundary annotation, which include obtaining sensor data along a trajectory corresponding to locations of a base map. Features are extracted from the sensor data. The features are input into a trained neural network that outputs overlapping rich feature maps comprising polylines. The overlapping rich feature maps are aggregated according to an aggregation function to obtain raster image. Vectorization is applied to the raster images to extract roadway geometry represented by globally consistent polylines.