Gait Authentication Using Time-Series Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gait-based authentication systems face challenges in accurately verifying individuals when the walking direction changes or when the video footage is short, leading to difficulties in extracting gait features and reducing recognition accuracy, especially as the number of registered individuals increases.
Innovation Solution
A person authentication apparatus that generates one or more second image sequences by applying a predetermined time-series operation to a first image sequence, extracts gait features from these sequences, and compares them with stored verification features to improve authentication accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a person walks for sufficient time in the same direction to extract gait features, then authentication accuracy is improved, but the system cannot handle cases where the video is short or the person changes walking direction
Solution Approach 1:
The patent segments the video sequence into multiple sub-sequences based on movement state changes (e.g., walking, standing, turning). Each sub-sequence is processed independently to extract gait features, allowing the system to handle videos with direction changes and varying lengths effectively.
Solution Approach 2:
The system performs preliminary analysis to detect movement state changes and segment the video before gait feature extraction. This preliminary action enables the system to prepare appropriate processing strategies for different video scenarios, improving both accuracy and adaptability.
2Loss of information
If gait features with time-series information are used, then expression capacity is improved, but extraction becomes difficult when video is short or direction changes occur
Solution Approach 1:
The patent employs dynamic gait feature extraction that adapts to different video conditions. The system dynamically adjusts the extraction strategy based on video length and movement patterns, maintaining high expression capacity while ensuring accurate extraction even in challenging scenarios.
Solution Approach 2:
The system changes extraction parameters based on detected movement states and video characteristics. By adjusting parameters dynamically according to the situation, the system maintains high gait feature quality regardless of video duration or direction changes.
3Quantity of substance
If the number of registered people increases, then system coverage is improved, but recognition accuracy decreases
Solution Approach 1:
The patent uses deep learning-based gait feature extraction that creates robust feature representations (copies of essential gait patterns) that remain distinctive even as the database grows. This copying approach maintains recognition accuracy by focusing on invariant gait characteristics rather than raw video data.
Data Source
AI summary
A person authentication apparatus (20) generates one or more second image sequences (50) by applying a predetermined time-series operation to a first image sequence (40) acquired from a camera (10). The person authentication apparatus (20) extracts, from each of two or more image sequences among the first image sequence (40) and the second image sequence (50), a gait feature of a person included in the image sequence. The person authentication apparatus (20) performs authentication of a person by comparing the plurality of extracted gait features with a verification feature stored in a verification database (30).