13.07.2015 Views

Thesis - Instituto de Telecomunicações

Thesis - Instituto de Telecomunicações

Thesis - Instituto de Telecomunicações

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

100 CHAPTER 5. FEATURE SELECTION AND CLASSIFICATIONThe classifier spends time applying the mo<strong>de</strong>l of each feature selected for each user’s datafor the same user, giving a linear <strong>de</strong>pen<strong>de</strong>nce on both n f and n c . Nevertheless, experimentalresults show that, the classification of one user’s data sample is always faster than the dataacquisition time, behaving well for real time operation.Feature selection constitutes the heaviest processing block since it implements a searchalgorithm that needs to be run for each subject. This block imposes the upper bound onthe time complexity of the system. We used a user tuned feature selection method thatruns separately for each user.For each user, the feature selection step apply the classifier for the n i features with iin a linear progression from i = 1 to i = n f (see line 7 of algorithm 2). Since the classifierhas O(n i ) complexity for each sample, the feature selection block has a time complexity of∑ nfi=1 O(n i)=O(n 2 f ). Therefore, for all users we obtain an overall complexity of O(n3 f n c).This time constraint, imposed by the feature selection step, means that this block canonly be used in offline mo<strong>de</strong>. The offline mo<strong>de</strong> would be composed by the complete set ofblocks, that would be run in a task intensive mo<strong>de</strong> prior to the <strong>de</strong>ployment of the real timeclassification system, that, in this case, only have the feature extraction and classificationblocks that have a linear behavior. In figure 5.4 we <strong>de</strong>pict the separation between the off-lineand the on-line parts of the proposed system.Figure 5.4: Blocks of the classification system divi<strong>de</strong>d in off-line (enrolment phase) andon-line (real-time) parts of system due to time complexity constrains.To reduce the time spent in this offline step, we proposed and implemented boosting

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!