Map Verification via Object Occlusion for Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving systems face challenges in verifying the accuracy of encoded maps, particularly in high-object-density environments where errors in HD maps, such as missing or mislocated traffic signals, can be difficult to detect.
Innovation Solution
A computer-implemented method that presents a traveling scene of a vehicle's route, determines known roadway objects based on a previously generated map, occludes these objects in the scene, and receives input from a verifier regarding visible objects to verify the map's accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If verifiers review the entire traveling scene including all roadway objects, then they can detect missing or mislocated objects, but it becomes difficult to identify discrepancies in high-object-density environments
Solution Approach 1:
The patent extracts and removes known roadway objects from the traveling scene by overlaying occluding elements (such as colored boxes or blurred regions) at the precise locations of objects identified in the previously generated map. This allows verifiers to focus exclusively on detecting missing or mislocated objects without being distracted by the full complexity of the scene, thereby improving discrepancy detection efficiency while maintaining verification accuracy
Solution Approach 2:
Instead of having verifiers directly search for missing objects in the complete scene, the patent inverts the approach by proactively hiding known objects and asking verifiers to identify what should be there but is obscured. This inversion transforms the verification task from active detection to passive confirmation, making it easier to spot discrepancies in high-density environments
2Measurement precision
If the map verification process includes detailed review of all roadway objects, then verification accuracy improves, but the verification time increases
Solution Approach 1:
The patent segments the verification process into two distinct phases: (1) automated identification of known roadway objects from the previously generated map with precise location marking, and (2) human verification focused only on detecting occluded or missing objects. This segmentation reduces the cognitive load on verifiers and accelerates the process while maintaining high accuracy through the combination of automated object identification and targeted human inspection
3Reliability
If the previously generated map contains accurate roadway object locations, then autonomous driving safety improves, but errors in high-object-density environments remain difficult to detect
Solution Approach 1:
The patent introduces an intermediary verification layer between map generation and autonomous driving deployment. The occlusion-based verification system acts as a mediator that systematically checks map accuracy by comparing expected object locations against actual scene observations, providing a reliable quality assurance mechanism that catches errors before they affect autonomous driving safety
Data Source
AI summary
Systems and methods are provided for verifying the accuracy of a map. Information regarding objects, such as roadway or roadway-related elements may be encoded into a map to be used for navigation and/or control of a vehicle, such as an autonomous vehicle. In order to verify that the map accurately reflects the roadway or roadway-related elements making up a section of roadway, the vehicle may be driven/ridden along the same section of roadway. A camera feed can be captured of this subsequent traversal of the section of roadway to check if the map includes the requisite roadway or roadway-related elements. The known roadway or roadway-related elements can be occluded from view in the camera feed. Accordingly, if any roadway or roadway-related elements do appear in the camera feed, they can be more easily detected by a verifier, and the processor verification is simplified.