Gait Recognition Under Various Viewing Angles Based on Correlated Motion Regression

摘要: It is well recognized that gait an important biometric feature to identify a person at distance, e.g., in video surveillance application. However, reality, change of viewing angle causes significant challenge for recognition. A novel approach using regression-based view transformation model (VTM) proposed address this challenge. Gait features from across views can be normalized into common learned VTM(s). In principle, VTM used transform one (source) another (target). consists multiple regression processes explore correlated walking motions, which are encoded features, between source and target views. the learning processes, sparse based on elastic net adopted as function, free problem overfitting results more stable models construction. Based widely database, experimental show method significantly improves upon existing VTM-based methods outperforms most other baseline reported literature. Several practical scenarios applying recognition under various also discussed paper.