| Summary: | 碩士 === 國立臺北大學 === 資訊工程學系 === 104 === This thesis aims to discuss partial event detection.In reality, the observation of events may be interrupted, such as occlusionsduring surveillance, or beginning to watch a facial expression of a person at the latter part.
In the proposed approach, thestructured output support vector machine is modified to tackle with partial event detection. then the training stage is reduced to a simpler version.Finally,two sets of constraints are used in the experiments.
In the experiments, four evaluation methodsare used, and a set of synthetic data and a set of sign language data are employed. For the synthetic signals, different amplitudes and frequencies are tested for different segments of whole signal sequences. For sign language data, two sentences are evaluated, namely, "I love you"and"happy boy crazy I draw", respectively. Two major resultsare as follows.First,the performance of equal importance for the whole time series is better than the performance of dramatic decreasing importance for the whole time series.Second,the performance of gentle decreasing importance for the whole time series is better than the performance of equal importance for the whole time series.
|
|---|
