Distillation-guided Representation Learning for Unconstrained Gait Recognition

TL;DR

This work proposes a framework, termed GAit DEtection and Recognition (GADER), for human authentication in challenging outdoor scenarios that leverages a Double Helical Signature to detect segments that contain human movement and builds discriminative features through a novel gait recognition method, where only frames containing gait information are used.

Abstract

Gait recognition holds the promise of robustly identifying subjects based on walking patterns instead of appearance information. While previous approaches have performed well for curated indoor data, they tend to underperform in unconstrained situations, e.g. in outdoor, long distance scenes, etc. We propose a framework, termed GAit DEtection and Recognition (GADER), for human authentication in challenging outdoor scenarios. Specifically, GADER leverages a Double Helical Signature to detect segments that contain human movement and builds discriminative features through a novel gait recognition method, where only frames containing gait information are used. To further enhance robustness, GADER encodes viewpoint information in its architecture, and distills representation from an auxiliary RGB recognition model, which enables GADER to learn from silhouette and RGB data at training time. At test time, GADER only infers from the silhouette modality. We evaluate our method on multiple State-of-The-Arts(SoTA) gait baselines and demonstrate consistent improvements on indoor and outdoor datasets, especially with a significant 25.2% improvement on unconstrained, remote gait data.

Figures (6)

• Figure 1: For each baseline, GADER raises the Rank-1 accuracy on BRIAR significantly, especially over 25% at 500m. Close Range(CR), Unmanned Aerial Vehicle(UAV).
• Figure 2: Overview of end-to-end pipeline. GADER consists of two parts: gait detection and GAR. The gait detector utilizes gait representation to filter out segments without gait information or incomplete body which will be processed by a body recognition algorithm, and only frames with human movement are fed to GAR. GAR leverages ratio attention and RGB feature space to extract a more robust silhouette feature for recognition.
• Figure 3: Four cases of DHS(a-d) are shown using two variables: full/part - indicates whether the body is complete, and stand/gait - shows whether gait information is present.
• Figure 4: The gait detection process. The gait detector processes split DHS segments to obtain prediction followed by Non-Maximum Suppression (NMS) and concatenation to pinpoint frames where gait information is present in a sequence. The yellow bounding boxes indicate the frames that contain gait.
• Figure 5: Some examples showing the absence of human movement in videos appearing in the curated dataset. We record the dataset name and corresponding subject id of the sequence.