Speed-Invariant Gait Recognition Using Single-Support Gait Energy Image

Chi Xu, Yasushi Makihara, Xiang Li, Yasushi Yagi, Jianfeng Lu

Abstract

Gait is one of the most popular behavioral biometrics because it can be authenticated at a distance from a camera without subject cooperation. Speed differences between matching pairs, however, cause significant performance drops in gait recognition, and gait mode difference (i.e., walking versus running) makes gait recognition further challenging. We therefore propose a speed-invariant gait representation called single-support GEI (SSGEI), which realizes a good trade-off between speed invariance and stability by aggregating multiple frames around single-support phases. In addition, to mitigate the pose differences between walking and running modes at single-support phases, we morph walking and running SSGEIs into intermediate SSGEIs between walking and running mode, where we exploit a free-form deformation field from the walking or running modes to the intermediate mode obtained by training data. We finally apply Gabor filtering and spatial metric learning as postprocessing for further accuracy improvement. Experiments on two publicly available datasets, the OU-ISIR Treadmill Dataset A and the CASIA-C Dataset demonstrate that the proposed method yields the state-of-the-art accuracies in both identification and verification scenarios with a low computational cost.

Introduction

Gait recognition—the ability to identify individuals by their unique walking patterns—is a highly desirable biometric because it can be performed at a distance without the subject's active participation. However, gait recognition faces a significant hurdle: speed variation.

When a person changes their speed, their physical gait characteristics—such as stride length and arm swing—alter significantly. Traditional features like the Gait Energy Image (GEI), which averages a person's silhouette over a complete walking cycle, are highly sensitive to these changes. This sensitivity makes it difficult for recognition systems to match a person walking slowly in a database (the "gallery") with the same person running in a test video (the "probe"). This paper introduces a novel framework to overcome these speed and mode-based challenges.

The Single-Support Gait Energy Image (SSGEI)

The core innovation of this research is the Single-Support Gait Energy Image (SSGEI). The researchers observed that the "single-support phase"—the specific moment when only one foot is on the ground and the limbs are closest to the vertical axis of the body—is the most stable part of the gait cycle across different speeds.

Comparison of gait sequences and feature extraction (GEI vs. SSGEI) at different speeds. The SSGEI focuses on the single-support phase to maintain consistency.

While a single keyframe is often too noisy for reliable recognition, the SSGEI aggregates a small window of frames around this single-support phase. This approach provides a crucial balance: it maintains the speed invariance of the single-support pose while gaining the stability of temporal averaging, effectively filtering out segmentation noise.

Methodology: Morphing and Noise Suppression

To handle the transition between walking and running (cross-mode recognition), the authors developed a multi-stage pipeline:

  1. Gait Mode Classification: The system uses a Support Vector Machine (SVM) to automatically determine if the subject is walking or running based on the SSGEI and the gait period.
  2. Symmetric Morphing via Free-Form Deformation (FFD): Rather than attempting to transform a "walk" directly into a "run," the system morphs both into a shared "intermediate" mode. This symmetric approach reduces geometric distortion and preserves more identity-linked details.
  3. Attenuation Field: Not all body parts warp consistently. Using an outlier detection method based on Earth Mover’s Distance, the authors identify unreliable regions (like swinging arms) and apply an "attenuation field" to dim these pixels, focusing the algorithm on more stable anatomical features.
  4. Post-processing: The features are further refined using Gabor filtering for texture analysis and spatial metric learning (2DLDA) to maximize discriminative power.

Experimental Results

The proposed method was rigorously tested on the OU-ISIR Treadmill Dataset A (OUTD-A) and the CASIA-C Dataset.

Visual comparison of SSGEI features for four subjects at different speeds. The consistency between the 8 km/h probe and 7 km/h gallery is evident.

The results demonstrate a new benchmark for the field:

  • Mode Classification: The system achieved 100% accuracy in distinguishing walking from running.
  • Identification Accuracy: On the OUTD-A dataset, the method reached 99.3% accuracy for within-walking scenarios. Most notably, in the difficult cross-mode (walk vs. run) scenario, it achieved 80.4% accuracy—a massive leap from the 53.4% achieved by previous state-of-the-art methods.
  • Efficiency: The query time per sequence is approximately 0.01 to 0.03 seconds, making the system highly viable for real-time surveillance.

Conclusion

By focusing on the single-support phase and utilizing symmetric morphing, the SSGEI framework effectively mitigates the performance degradation caused by speed and gait mode changes. While cross-mode recognition remains a complex challenge, this method provides a robust and computationally efficient path forward for real-world biometric applications. Future research will focus on expanding these techniques to handle varying viewpoints and integrating deep learning as larger gait datasets become available.

Summary updated

Abstract

Gait is one of the most popular behavioral biometrics because it can be authenticated at a distance from a camera without subject cooperation. Speed differences between matching pairs, however, cause significant performance drops in gait recognition, and gait mode difference (i.e., walking versus running) makes gait recognition further challenging. We therefore propose a speed-invariant gait representation called single-support GEI (SSGEI), which realizes a good trade-off between speed invariance and stability by aggregating multiple frames around single-support phases. In addition, to mitigate the pose differences between walking and running modes at single-support phases, we morph walking and running SSGEIs into intermediate SSGEIs between walking and running mode, where we exploit a free-form deformation field from the walking or running modes to the intermediate mode obtained by training data. We finally apply Gabor filtering and spatial metric learning as postprocessing for further accuracy improvement. Experiments on two publicly available datasets, the OU-ISIR Treadmill Dataset A and the CASIA-C Dataset demonstrate that the proposed method yields the state-of-the-art accuracies in both identification and verification scenarios with a low computational cost.
BibTeX
@article{xu2019speed,
  author = {Chi Xu and Yasushi Makihara and Xiang Li and Yasushi Yagi and Jianfeng Lu},
  title = {{Speed-Invariant Gait Recognition Using Single-Support Gait Energy Image}},
  journal = {Multimedia Tools and Applications},
  year = {2019},
  url = {https://openreview.net/forum?id=Fh9EyCZp5f},
}
Figure