Object Localization in Discontinuous Observation Scenes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for dynamic object reconstruction and localization in computer vision rely on continuous observations and known CAD models, which are not feasible in real-world scenarios due to limited viewing angles and occlusions, leading to inaccurate object observation, reconstruction, and localization during observation interruptions.
Innovation Solution
A method for object localization in discontinuous observation scenes that acquires an object model based on a benchmark image after observation resumption and associates it with a pre-existing object reconstruction model to establish correspondence between objects before and after the interruption, enabling accurate localization without continuous observations or CAD models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If continuous observation is used for object reconstruction and localization, then observation completeness is improved, but system complexity and requirement for uninterrupted sensing increase
Solution Approach 1:
The system performs preliminary actions by capturing and storing key object features (edges, corners, contours) during continuous observation phases before interruption occurs. These pre-captured features are then used to reconstruct and localize objects during discontinuous observation periods, eliminating the need for uninterrupted sensing while maintaining reliability.
Solution Approach 2:
The patent creates simplified copies of object models by extracting essential geometric features (edges, corners, contours) from complete observations. These feature-based copies enable object reconstruction and localization during interruption periods without requiring the original continuous observation data, reducing system complexity while maintaining observation completeness.
2Measurement precision
If CAD models are used for object reconstruction, then reconstruction accuracy is improved, but applicability to unknown objects deteriorates
Solution Approach 1:
The system changes the parameter representation from complete CAD models to extracted geometric features (edges, corners, contours). This parameter transformation enables accurate reconstruction of both known and unknown objects by focusing on invariant geometric properties that can be captured and matched without requiring pre-existing CAD libraries.
Solution Approach 2:
The patent extracts essential geometric features (edges, corners, contours) from objects, separating these invariant properties from the complete CAD model representation. This extraction enables reconstruction accuracy comparable to CAD-based methods while extending applicability to unknown objects that lack pre-existing models.
3Adaptability or versatility
If object features are extracted for model-free reconstruction, then versatility for unknown objects is improved, but reconstruction precision deteriorates
Solution Approach 1:
The system applies local quality by focusing computational resources on extracting and processing specific local geometric features (edges, corners, contours) rather than attempting to capture complete object surfaces. This selective feature extraction maintains reconstruction precision by concentrating on the most informative geometric elements while enabling versatility for unknown objects.
Data Source
AI summary
The present disclosure relates to method and apparatus of object localization in a discontinuous observation scene, and a storage medium. Provided is a method of object localization in a discontinuous observation scene, comprising: acquiring an object model based on a benchmark image that is obtained when the observation is resumed from interruption, and based on the acquired object model and an object reconstruction model, achieving association between objects before and after the observation interruption.