Introduction
Event Capture
Defining Example Events
Capturing the Events
Event Correlation
Succesful Event Detection
The Event Correlator PlugIn has two processes - a Capture process, and a Correlation process.
The capture process samples trajectory data around an event which you have already defined. The capture process is the hardest part to master for good event detection results, since you must carefully identify trajectory based data which has a characteristic pattern to it. However, a small time spent in understanding the techniques involved will give a good return in the automatic identification of events in the future.
Once you have captured the events, the correlation process is very simple to apply, just by running the pipeline. It compares the captured event data with any other trial, to detect and generate new events.
Captured events are held in files. You can capture several different files representing the same events for different subjects, or you can capture completely different events in different files. This allows you to build a "library" of captured events which could be used to correlate with different trials or even new subjects that have a similar motion to your captured subject.
Event capture actually consists of sampling the data from trajectory data in your trial for a few frames either side of when the event occurs.
The first stage in the process is to identify the events which would like to capture and correlate, and identify one or two marker trajectories, or other trajectories containing calculated data which is relavent to your event. You should try to identify a characteristic shape in the time curve of your trajectories - a sharp change in direction of a marker for example, or perhaps a short peak in a calculated variable.
A good tip when doing this is to view the trajectories in the Workstation 3D Workspace, with the trajectory path extended a few frames either side of the current time cursor, and step back and forth a few frames around your chosen event. This should also help you decide how many frames of the trajectories to capture either side of the event.
Normally between 15 and 30 frames is recommended, depending on the event and the frame rate of your system. You should not capture too many frames, or you may capture more than one similar event. For greater accuracy where events in different trials occur at different speeds, it may also be better to select the same number of events either side of the event.
You should try to select your example events with care, since it is these events which are going to be correlated with other trials to determine the existance of matching or similar events.
It's also important that you choose an event where the trajectory has no gaps in the data. You should avoid events near the beginings and ends of trials, to make sure you capture a full length of data.
If you have more than one event you should try to find a good series of these events, so that they can be defined in the right order.
Once you have determined your example events, you should specify them using the tools on the Workstation Timebar. Move the time cursor to the correct frame, make sure you have the correct context selected (press the L, R or G button), and press one of the diamond, arrow or vertical line buttons to define an event (Foot Strike, Foot Off, or General). It doesn't matter that the actual events you are defining are unrelated to gait, since EventCorrelator allows you to change the event names when it finds a correlation. The symbol which is used will be replicated however, so it's best to choose a mix of different event types.
You should make sure that you note the different types of event you define, the order which you define them in, and in which context. When you set up the events to be captured, you must define them to be captured in the correct order.
Once you have defined your example events, you are ready to set up the events for capture. In the Workstation pipeline, select the "Capture matching events for correlation" process, and press the Options button. See the Event Capture Settings help for details of how to set up your events ready for capture.
Once you've defined the example events, and set up the Event Capture Settings, just make sure that the "Capture matching events for correlation" process is selected in the pipeline, and press the Process Now button.
Since the Capture process does not make any changes to the trial data, you should not see any changes in Workstation (unless you have other processes enabled). To review any problems in the capture process, refer to the Workstation Processing Log (under the View menu). This indicates which events were found, and which trajectories sampled.
Once you have captured the data, you should make sure that the capture process is no longer selected in the pipeline, or you may cause other events to be captured, overwriting your existing data. One way to automate this is to define a new "trial type" in Workstation, for event capture processing (see your Workstation manual for details on how to do this).
The event correlation process proceeds almost automatically. The only thing you must do is ensure that you correlate your trial with the correct captured data. Select "Correlate captured events" from the pipeline, and press the Options button to get the Correlation Settings dialogue, and make sure the correct event data file is selected.
Once that's done, you only have to select the "Correlate captured events" process, and run the pipeline. This should result in new events appearing in the time bar.
You can (of course) correlate your captured events with the trial from which you captured the events. If you delete the events before correlation, new events should appear in exactly the same place the old events were - other new events may also be found, depending on your trial. If you leave the original captured events in place, they will not be replaced. If you view the processing log, you should see that it detects events which match the exisiting events, however.
Succesful event detection depends on the event that you want to detect, and especially on the configuration of the trajectories at the time of the event.
There should be some kind of charateristic movement that the trajectory makes. For example, the heel marker of a toe walker shows a charateristic sharp, vertical dip as the weight of the subject is borne by the foot. Note that it is the shape of the time based curve which is important.
Naturally, the events need to be quite similar in order for a good correlation to be achieved. This may not be the case where you try to correlate events from one subject with trials of another, or in cases where the same subject is moving in a characteristically different manner in the captured and correlated trials. In these cases you may need to capture a new set of events.
In many cases (depending on your application) you will be able to detect events quite adequately by correlating trials from new subjects against captured data from existing subjects. It is envisaged that you can build up a collection of events for typical movement patterns, saved in different, named capture files. These can be selected for new subjects, depending on a subjective assessment of their motion.
If you do require greater accuracy for you event timings, you can always capture a new set of events for each new subject. Only one such capture should be required.