1-2hit |
Kenji MATSUI Toru TAMAKI Bisser RAYTCHEV Kazufumi KANEDA
We propose a feature for action recognition called Trajectory-Set (TS), on top of the improved Dense Trajectory (iDT). The TS feature encodes only trajectories around densely sampled interest points, without any appearance features. Experimental results on the UCF50 action dataset demonstrates that TS is comparable to state-of-the-arts, and outperforms iDT; the accuracy of 95.0%, compared to 91.7% by iDT.
Two new segmentation methods for Japanese syllable recognition were proposed and were experimentally compared. In one of these methods, the adaptive power envelope template was compared with the logarithmic power envelope of input speech, and the most suitable position was searched. The second difference of the clipped power envelope was calculated and the point where is gave the maximum value was detected as the segment boundary in another method. A segmentation rate improvement of 2.1%, and the accurate syllable recognition rate of 90.5% in average for 10 speakers were obtained by the later method.