HD Map Training Data Generation From Labeled Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional maps for autonomous vehicles lack precision and accuracy, and the process of creating and updating high-definition maps is expensive, time-consuming, and unable to keep up with frequent road changes, making it challenging for autonomous vehicles to navigate safely and efficiently.
Innovation Solution
A system that generates high-definition maps by combining sensor data from autonomous vehicles to create a point cloud, labels objects, and uses this data to train deep learning models for accurate navigation, allowing for real-time updates and precise location determination within a safety threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional survey teams create high-definition maps using specialized survey cars, then map accuracy and precision are improved, but the cost and time required for map creation increase significantly
Solution Approach 1:
The patent uses conventional vehicle sensors to capture sensor data that copies the environmental information needed for HD maps, eliminating the need for specialized survey cars. The system processes this copied data through deep learning models to generate map features, achieving accurate maps without expensive survey equipment.
Solution Approach 2:
The patent replaces the mechanical surveying system (specialized survey cars with high-resolution sensors) with an automated computational system. Conventional vehicle sensors combined with deep learning algorithms substitute for the mechanical surveying apparatus, reducing both cost and time while maintaining map accuracy.
2Reliability
If survey teams frequently update maps to keep up with road changes, then map freshness is improved, but the cost and resources required increase
Solution Approach 1:
The system enables conventional vehicles to automatically contribute sensor data for map updates during normal operation. Vehicles self-service the map updating process by capturing and transmitting data without requiring dedicated survey operations, allowing continuous map freshness without additional survey fleet complexity.
Solution Approach 2:
The patent makes conventional vehicles multi-functional by enabling them to serve both their primary transportation purpose and the secondary function of collecting map data. This universal use of existing vehicles eliminates the need for specialized survey fleets while maintaining map freshness through continuous data collection from diverse sources.
3Area of stationary object
If GPS systems are used for location determination, then coverage area is improved, but location accuracy deteriorates to over 100 meters
Solution Approach 1:
The patent merges GPS data with deep learning-based map features and sensor data to create a hybrid localization system. This combination maintains the broad coverage of GPS while achieving centimeter-level accuracy through integration with HD map features, resolving the contradiction between coverage area and location precision.
Data Source
AI summary
According to an aspect of an embodiment, operations may comprise receiving sensor data from one or more vehicles, determining, by combining the received sensor data, a high definition map comprising a point cloud, and labeling one or more objects in the point cloud. The operations may also comprise generating training data by receiving a new image captured by one of the vehicles, receiving a pose of the vehicle when the new image was captured, determining an object having a label in the point cloud that is observable from the pose of the vehicle, determining a position of the object in the new image, and labeling the new image by assigning the label of the object to the new image, the labeled new image comprising the training data. The operations may also comprise training a deep learning model using the training data.