Multiple Feature Fusion Based on Co-Training Approach and Time Regularization for Place Classification in Wearable Video

Figure 5

Effect of the DAS late fusion approach on the final performance, for various supervision levels. Plot of the accuracy as a function of the parameter 𝛼 that balances the fusion between SPH features (3 levels) if 𝛼=0 and CRFH if 𝛼=1 (IDOL2 dataset, random setup).