Gait-Based Human Identification Across Occluded Frames
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional human identification and tracking methods face challenges in accurately identifying individuals across temporally-distant frames or frames captured by different cameras, especially when individuals are partially hidden, under varying lighting conditions, or when positional changes occur, leading to difficulties in maintaining tracking.
Innovation Solution
A human identification apparatus that extracts and compares spatiotemporal period, phase, and position information from walking sequences to verify whether human images represent the same person, using a gait information verifying unit to judge identity based on periodic motion characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional methods use circumscribed rectangles and motion vectors to identify persons in neighboring frames, then identification is possible when persons are visible and moving continuously, but identification fails when detection fails in certain frames or when persons are hidden behind objects
Solution Approach 1:
The system extracts gait information from walking sequences in advance and stores it for later verification. By preparing gait templates beforehand from continuous walking sequences, the system can perform reliable identification even when frame detection fails or persons are temporarily occluded, as the pre-extracted gait information serves as a reference for matching against partial or intermittent observations.
2Measurement precision
If conventional methods use colors and image patterns to identify persons across frames, then identification may be possible in some cases, but positional changes due to person movement and variations in lighting conditions, posing, and camera direction make it difficult to correspond human images
Solution Approach 1:
The system transforms the identification problem from comparing raw image pixels (which are sensitive to lighting and position) to comparing extracted gait parameters (period, phase, position information). By changing the parameter space from visual appearance to biomechanical motion characteristics, the system achieves robust identification that is invariant to lighting conditions, camera angles, and temporary occlusions.
3Quantity of substance
If the system uses temporally-distant frames for identification, then more frames are available for comparison, but positional changes increase making correspondence more difficult
Solution Approach 1:
The system extracts essential gait characteristics (period, phase, and position information) from walking sequences, separating the identification-critical motion patterns from the distracting visual appearance changes. This extraction allows the system to compare gait signatures across temporally-distant frames without being affected by accumulated positional drift, as the gait parameters remain consistent despite changes in absolute position.
Data Source
AI summary
A human identification apparatus, which can judge for identification of human images even in temporally-distant frames or frames shot with different cameras, judges whether or not persons represented by human images respectively included in different image sequences are the same person, and includes: a walking posture detecting unit which detects first and second walking sequences, each sequence being an image sequence indicating a walking state of respective first and second persons respectively included in the different image sequences; and a walking state estimating unit which estimates a transition state of a walking posture in the periodic walking movement of the first person at a time or in a position different from a time or a position of the walking sequence of the first person; and a judging unit which verifies whether or not the estimated transition state of the walking posture of the first person matches the transition state of the walking posture of the second person, and judges that the first person and the second person are the same person in the case where the transition states match with each other.