Human Action Recognition with Weighted Feature Fusion Based on STIP and Dense Trajectory Feature

Aiming at the low performance of feature fusion for human action recognition, we proposed a new weighted fusion method based on feature fusion. This method, by fusing weighted features based on Dense Trajectory with feature based on Spatial Temporal Interest Point, and by utilizing SVM classifier, realized recognizing human actions. Our method can add more weight to feature with higher classification power, which can benefit final recognition result. We test the proposed method on used-widely KTH dataset and UCF sports dataset. Comparison of our experiment results with those of the newest methods shows that our method is effective and has strong robustness.