Object Region Extraction via Back Projection Angle Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object region extraction systems in three-dimensional space face issues with non-extraction and incorrect extraction due to mis-extraction of moving object regions from camera images, particularly when objects move to positions relative to cameras that face each other or are parallel, leading to incorrect region extraction.
Innovation Solution
An object region extraction system that evaluates the effectiveness of back projections by calculating an effectiveness evaluation quantity based on the angle between them, determining the region as occupied by an object when the evaluation quantity meets a predetermined threshold, and using object likelihood to enhance robustness in object position estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the restrictive condition is changed to extract a three-dimensional space whose back projections are obtained from a large number of cameras, then non-extraction of object region is eliminated, but incorrect region extraction occurs when objects move to positions where two cameras face each other or are in parallel
Solution Approach 1:
The patent changes the evaluation parameter from simple overlap detection to angle-based effectiveness evaluation. By calculating the angle between back projection rays and using this as an evaluation criterion, the system can distinguish between valid object regions and incorrect regions caused by adverse camera configurations, thus resolving the contradiction between extraction reliability and position precision
Solution Approach 2:
The patent replaces the mechanical/geometric approach of simple back projection overlap with an evaluation system based on angular measurements. This substitution allows for nuanced assessment of back projection effectiveness, enabling the system to maintain high extraction reliability while avoiding incorrect position estimation in challenging camera configurations
2Measurement precision
If all back projections must overlap to determine object region, then incorrect extraction is reduced, but non-extraction occurs when mis-extraction happens in any one camera
Solution Approach 1:
The patent applies partial action by not requiring complete overlap of all back projections. Instead, it evaluates each back projection's effectiveness individually based on angular criteria and determines object regions based on aggregated evaluation results. This allows the system to tolerate mis-extraction in individual cameras while maintaining overall extraction reliability
Solution Approach 2:
The patent introduces feedback through the effectiveness evaluation mechanism. Each back projection is evaluated based on its angle, and this evaluation feeds into the overall object region determination process. This feedback system allows the algorithm to compensate for individual camera mis-extractions by relying on evaluations from other cameras with better viewing angles
Data Source
AI summary
An object region extraction system, an object region extraction method, and an object region extraction program,including an effectiveness evaluation quantity acquisition unit to acquire an evaluation quantity which increases or decreases when an angle formed by two back projections selected from back projections of one point of an object approach a right angle, wherein the back projections cross each other at a position in three-dimensional space and obtained from respective images, which include the one point and captured by respective m-number of image acquisition units, where m is greater than one, and an object region determination unit to, when the evaluation quantity is equal to or larger than or equal to or smaller than a predetermined value, determine the position to be a region occupied by the object, and output the determination result.