Skeleton–Silhouette Complementary Perception: Toward Robust Gait Recognition

Abstract

Gait, the unique pattern of how a person walks, has emerged as one of the most promising biometric features in modern intelligent sensing. Unlike fingerprints or facial characteristics, gait can be captured unobtrusively and at a distance, without requiring the subject’s awareness or cooperation. This makes it highly suitable for long-range surveillance, forensic investigation, and smart environments where contactless recognition is crucial. Traditional gait-recognition systems rely either on silhouettes, which capture the outer appearance of a person, or on skeletons, which describe the internal structure of human motion. Each modality provides only a partial understanding of gait. Silhouettes emphasize shape and contour but are easily distorted by clothing or carried objects; skeletons describe motion dynamics and limb coordination but lose discriminative details about body shape. This article presents the concept of Complementary Semantic Embedding (CSE), a unified framework that merges silhouette and skeleton information into a comprehensive semantic representation of human walking. By modeling the complementary nature of appearance and structure, the approach achieves more robust and accurate gait recognition even under challenging conditions.