核心概念
GaitContourは、新しいContour-Pose表現を活用して効率的な歩行認識を実現します。
要約
Abstract:
Gait recognition focuses on walking patterns for subject identification.
Proposed Contour-Pose representation combines body shape and parts information efficiently.
Introduction:
Challenges in biometric identification outdoors lead to gait analysis as an alternative.
Recent advancements in deep learning methods for gait recognition are discussed.
Method:
Contour-Pose representation is introduced, combining silhouette edges and pose keypoints.
GaitContour model leverages this representation with a local-to-global architecture for efficient computation.
Experiments:
Evaluation on large-scale datasets shows significant performance improvements compared to previous methods.
Conclusion:
Contour-Pose and GaitContour offer enhanced efficiency and performance in gait recognition.
統計
歩行ベース[15]と比較して、GPGait [17]は10倍のFLOPsを要求します。
Silhouetteベースの方法は、通常、シルエットシーケンスの各ピクセルで畳み込み操作を実行するため、高い計算コストがかかります。