Amodal Landmark Localization Under Dynamic Object Occlusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in accurately determining the position and route of moving objects based on semantic road information, especially in environments with many dynamic objects.
Innovation Solution
A processor-implemented method that determines amodal regions in image frames, calculates occluded region confidence information, and uses this information to calculate transformation and localization data for electronic devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If semantic road information is used to estimate position and route of moving objects, then autonomous driving stability is improved, but measurement precision deteriorates in environments with many dynamic objects causing occlusions
Solution Approach 1:
The patent segments the road surface into multiple regions based on semantic information (e.g., drivable areas, sidewalks, obstacles). By dividing the complex environment into manageable segments, the system can process each region independently, improving position estimation accuracy even when some regions are occluded by dynamic objects
Solution Approach 2:
The patent performs preliminary classification of road surfaces into different semantic regions before position estimation. This advance segmentation and labeling of the environment allows the system to pre-identify reliable reference areas, enabling more accurate position calculation even when dynamic objects occlude parts of the scene
2Measurement precision
If amodal region analysis is performed to handle occlusions, then position determination accuracy is improved, but device complexity increases
Solution Approach 1:
The patent extracts and focuses specifically on the amodal region (the region behind occluding objects) from the full image. By isolating and analyzing only the relevant occluded areas rather than processing the entire scene, the system achieves accurate position determination while keeping computational complexity manageable
Solution Approach 2:
The patent introduces amodal completion as an intermediary processing step between raw image capture and final position estimation. This intermediate analysis of occluded regions provides bridging information that improves accuracy without requiring direct observation of all areas, effectively mediating between limited visual input and comprehensive position determination
Data Source
AI summary
A processor-implemented method including determining, from a first image frame, a first amodal region including a visible region in which a static landmark is visible and an occluded region in which the static landmark is occluded, calculating an occluded region confidence information for the occluded region in the first amodal region based on the first amodal region, determining a second amodal region corresponding to the static landmark from a second image frame temporally subsequent to the first image frame, calculating transformation information between the first image frame and the second image frame based on the first amodal region, the second amodal region, and the occluded region confidence information, and calculating localization information of an electronic device comprising the processor based on the transformation information.