Cross-View Gait-Based Gender Classification by Transfer Learning

The gender of a person is easily recognized by his/her gait when training data and test data are from the same view. However, when it comes to cross-view gender classification, traditional methods can not deal with large view variation without enough labeled data in the target view. In this paper, we solve this problem by introducing a transfer learning based framework. Firstly, Gait Energy Image (GEI) of each gait sequence for all views is generated, and Principal Component Analysis (PCA) is carried out to obtain efficient gait representations. Subsequently, an inductive transfer learning approach, TrAdaBoost, is adopted to transfer knowledge from the source view to the target view, which significantly improves the performance of gait-based gender classification. Abundant experiments are conducted and experimental results demonstrate the superiority of the proposed method over traditional gait analysis methods.