GestureRecognitionToolkit  Version: 0.2.5
The Gesture Recognition Toolkit (GRT) is a cross-platform, open-source, c++ machine learning library for real-time gesture recognition.
TimeSeriesClassificationSampleTrimmer.h File Reference

This class provides a useful tool to automatically trim timeseries data. More...

Go to the source code of this file.

Classes

class  TimeSeriesClassificationSampleTrimmer
 

Detailed Description

This class provides a useful tool to automatically trim timeseries data.

Author
Nicholas Gillian ngill.nosp@m.ian@.nosp@m.media.nosp@m..mit.nosp@m..edu
Version
1.0

Sometimes, when a user records a timeseries example, there is a significant pause at the start and end of the example where the recording has started but the user has not started to move yet. These areas of 'static' data (at the start and end of a recording) can generate poor classification models (as the classification model expects to see similar static data in the real time data).

The LabelledTimeSeriesClassificationSampleTrimmer attempts to detect and remove these static areas of data. This is done by computing the summed absolute energy of the timeseries data, normalizing the energy profile by the maximum energy value, and then searching for areas at the start and end of the timeseries that are below a specific trimthreshold (set by the user). Any data that is below the trimthreshold will be removed, up until the first value that exceeds the threshold. This search is run both from the start of the timeseries (searching forward) and the end of the timeseries (searching backwards). If the length of the new timeseries is below the maximumTrimPercentage, then the timeseries will be trimmed and the trimTimeSeries function will return true. If the length of the new is above the maximumTrimPercentage, then the timeseries will not be trimmed and the trimTimeSeries function will return false. Set the maximumTrimPercentage to 100 if you want the timeseries to always be trimmed.

Definition in file TimeSeriesClassificationSampleTrimmer.h.