Average gait differential image based human recognition.

Area of Science:

• Computer Vision
• Biometrics
• Pattern Recognition

Background:

• Human gait identification relies on analyzing walking patterns.
• Previous methods like Gait Energy Image (GEI) capture static information but may miss dynamic variations.
• The difference between consecutive frames in human walking offers valuable, yet underexplored, biometric data.

Purpose of the Study:

• To propose a novel human gait recognition method using silhouette differences between adjacent frames.
• To introduce the Average Gait Differential Image (AGDI) as a feature representation that captures both kinetic and static gait information.
• To evaluate the effectiveness of AGDI against GEI and assess the efficiency of 2D Principal Component Analysis (2DPCA) for feature extraction.

Main Methods:

• Generation of Average Gait Differential Image (AGDI) by accumulating silhouette differences between adjacent frames.
• Utilizing 2D Principal Component Analysis (2DPCA) for feature extraction from AGDI.
• Experimental validation on the CASIA gait dataset.

Main Results:

• AGDI effectively preserves both kinetic and static information, offering a better representation of silhouette variations during walking compared to GEI.
• Experiments demonstrated superior identification and verification performance of AGDI over GEI.
• 2DPCA proved to be a more efficient method with lower memory consumption for feature extraction in gait recognition compared to traditional PCA.

Conclusions:

• The proposed AGDI method enhances human gait recognition accuracy by effectively capturing dynamic and static walking characteristics.
• AGDI represents a significant improvement over existing methods like GEI for gait-based identification.
• 2DPCA is an efficient feature extraction technique suitable for AGDI-based gait recognition systems, reducing computational overhead.