Person Identity Tracking Using Pose Detection in Occluded Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis techniques fail to accurately determine identical persons in a series of images, particularly when individuals are partially hidden behind objects or in overlapping situations.
Innovation Solution
An image analysis system that includes an image acquisition unit, a detection unit for identifying persons and their poses, and a decision unit using pose feature values to determine identity across images, employing conditions such as similarity degree, temporal, and spatial overlap to accurately identify individuals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature value matching is used to determine identical persons, then identification can be performed based on visual characteristics, but accuracy deteriorates when persons are partially hidden or overlapping in images
Solution Approach 1:
The patent segments the person detection problem into two independent components: pose detection (skeleton extraction) and feature value detection. By separating these functions, the system can rely on pose information which remains detectable even when parts of the person are occluded, while feature values provide additional confirmation when visible. This segmentation resolves the contradiction by making identification reliability less dependent on complete visual visibility.
Solution Approach 2:
The patent introduces pose information as an intermediary element that bridges the gap between visual feature matching and reliable identification in occlusion scenarios. Pose data serves as a mediator that can be detected even when feature values are partially hidden, enabling the system to maintain identification accuracy under varying visibility conditions.
2Device complexity
If only feature values are used for person identification, then the system is simpler to implement, but the ability to track persons through occlusion is lost
Solution Approach 1:
The system segments the identification function into pose detection and feature value detection components. This modular approach allows the system to maintain reasonable complexity while significantly improving tracking reliability, as each component can be implemented independently and combined to solve the overall problem.
Solution Approach 2:
The patent makes the identification system multi-functional by enabling it to operate effectively in both visible and occluded scenarios. The pose-based identification works universally across different visibility conditions, while feature value matching provides additional capability when persons are fully visible, creating a universal system that adapts to various situations.
3Measurement precision
If pose detection is added to feature value matching, then identification accuracy improves in occlusion scenarios, but system complexity increases
Solution Approach 1:
By segmenting the detection system into separate pose detection and feature value detection modules, the patent manages complexity through functional separation. Each module can be optimized independently, and their results are combined to achieve high identification accuracy without requiring a completely complex integrated system.
Solution Approach 2:
The pose detection system serves itself by providing structural information that is inherently more reliable for tracking than feature values alone. The skeleton-based pose representation is self-sufficient for maintaining person identity through occlusion, reducing the burden on the feature value matching component and allowing each subsystem to operate more efficiently.
Data Source
AI summary
To accurately determine an identical person in a plurality of images continuous in time series, an image analysis apparatus 100 includes an image acquisition unit 102 that acquires a plurality of images continuous in time series, a detection unit 103 that detects a person and a pose of the person in each of the plurality of images, and a decision unit 104 that decides identity of persons detected in images different from each other by using the pose of the person detected by the detection unit 103.