OpenGait: A Comprehensive Benchmark Study for Gait Recognition Toward Better Practicality

Area of Science:

• Computer Vision
• Biometrics
• Artificial Intelligence

Background:

• Gait recognition shows promise for remote person identification but struggles with real-world and outdoor datasets.
• Existing indoor-focused gait recognition methods often lack generalizability and practical robustness.

Purpose of the Study:

• To conduct a comprehensive benchmark study focused on improving the practicality of gait recognition systems.
• To address the limitations of current methods in real-world, outdoor scenarios.

Main Methods:

• Development of OpenGait, a flexible and efficient gait recognition platform.
• In-depth ablation experiments using OpenGait to analyze recent gait recognition advancements.
• Development of three novel baseline models: DeepGaitV2 (appearance-based), SkeletonGait (model-based), and SkeletonGait++ (multi-modal).

Main Results:

• Identification of previously overlooked insights and imperfections in existing gait recognition methods.
• Development of DeepGaitV2, SkeletonGait, and SkeletonGait++ achieving state-of-the-art performance.
• Demonstration of practical robustness and empirical power in the new baseline models.

Conclusions:

• The study provides new perspectives on deep gait modeling and the representational capacity of gait modalities.
• The developed models offer significant improvements in practical gait recognition, especially for outdoor environments.
• Future research should focus on key trends and challenges to enhance the practicality of gait recognition.