3D Shape Estimation for Cross-Camera Subject Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in accurately identifying the same person across different cameras due to variations in installation position and viewing angle, leading to differences in feature quantity vectors.
Innovation Solution
An image processing apparatus that extracts subject images from multiple cameras, estimates the three-dimensional shape and state of subjects, generates predictive images based on these estimates, and determines subject identity by comparing predictive images with actual images from other cameras.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature quantity vectors are extracted from images captured by different cameras, then person identification can be performed, but identification accuracy deteriorates due to variations in installation position and viewing angle
Solution Approach 1:
The invention transforms the identification approach from direct 2D image comparison to 3D shape-based comparison. By estimating three-dimensional shapes of persons from 2D images captured at different angles and comparing these 3D representations, the system achieves viewpoint-invariant identification. This dimensional transformation allows the same person to be identified regardless of camera position or viewing angle.
2Area of stationary object
If multiple cameras are installed to cover large-scale areas, then monitoring coverage is improved, but difficulty in detecting and measuring identical subjects increases due to viewpoint differences
Solution Approach 1:
The invention resolves the difficulty of identifying subjects across multiple cameras by transforming 2D viewpoint-dependent images into 3D viewpoint-independent shape representations. The 3D shape estimation process creates a unified representation that can be compared across different camera perspectives, enabling reliable identification throughout the entire monitored area regardless of how many cameras are deployed.
3Measurement precision
If direct image comparison is used for subject identification, then processing complexity is reduced, but identification accuracy deteriorates due to appearance variations
Solution Approach 1:
The invention introduces 3D shape estimation as an intermediate processing step that transforms appearance-dependent 2D images into appearance-independent 3D representations. While this adds processing complexity, it enables accurate identification by comparing fundamental geometric properties rather than superficial visual characteristics that vary with lighting, angle, and pose.
Data Source
AI summary
An image processing apparatus includes a first extraction means for extracting a first subject image from a first image captured by a first imaging unit, a second extraction means for extracting a second subject image from a second image captured by a second imaging unit, a first estimation means for estimating a three-dimensional shape of the first subject based on the first subject image, a second estimation means for estimating a second subject state based on the second subject image, an acquisition means for acquiring a predictive image predicted to be acquired when the first subject is captured by the second imaging unit based on the estimation results of the first subject's three-dimensional shape and the second subject's state, and a determination means for determining whether the first subject and the second subject are identical by comparing the predictive image with the second subject image.