Object-Point Localization for Incomplete HD Map Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing high-definition maps often have incomplete and inaccurate object information, making it difficult to precisely localize vehicles using sensor data, especially for tasks requiring precise lane-level positioning.
Innovation Solution
A system that estimates a point based on detected object points from an image and projects a map point for comparison, allowing accurate association and localization by comparing estimated points with map points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If HD maps use single point information to represent objects, then map simplicity is improved, but localization precision deteriorates
Solution Approach 1:
The patent segments the object representation from a single point into multiple candidate points with associated scores. Each candidate point represents a possible location of the object, and the system evaluates multiple hypotheses rather than relying on a single potentially inaccurate point, thereby improving localization precision while maintaining map simplicity.
Solution Approach 2:
The patent introduces a scoring dimension to evaluate candidate points, transforming the problem from selecting one point in 2D space to evaluating multiple points with associated confidence scores. This additional dimension allows the system to handle uncertainty in object location without increasing the fundamental complexity of the map structure.
2Loss of information
If HD maps have incomplete object information, then data collection burden is reduced, but object matching accuracy deteriorates
Solution Approach 1:
The patent applies partial action by using only the essential candidate point information from HD maps without requiring complete object details. The system generates multiple candidate points with scores based on available partial information, then uses sensor data to evaluate and select the most likely match, achieving accurate object matching despite incomplete map data.
Solution Approach 2:
The patent introduces candidate points with scores as an intermediary between the incomplete HD map data and the sensor observations. These candidate points serve as hypotheses that bridge the gap between partial map information and the need for accurate object matching, allowing the system to work with incomplete information while maintaining matching accuracy.
3Device complexity
If satellite-based navigation is used for location determination, then system simplicity is improved, but positioning precision deteriorates
Solution Approach 1:
The patent merges satellite-based navigation with sensor-based object detection and HD map matching. The system combines the broad location information from GPS with precise relative positioning from sensor data and candidate point matching, achieving high-precision lane-level localization while maintaining the simplicity of using satellite navigation as the base system.
Data Source
AI summary
Techniques and systems are provided for localization of an apparatus. For instance, a process can include: obtaining a map of an environment, the map including a map point representing an object in the environment; obtaining image data including the object in the environment, wherein the image data is associated with a camera pose; obtaining point information from the obtained image data, the obtained point information describing two or more points of the object in the obtained image data; determining whether to associate the point information with the map point based, at least in part, on a comparison of the map point and an estimated point, wherein the estimated point is estimated based on the camera pose and point information; and based on the determination to associate the point information with the map point, associating the point information with the map point to localize an apparatus.