Using your AX3 device

Mounting Conventions

The AX3 device is suitable for monitoring human movement data. However, for successful data capture the AX3 must be securely fastened to the target with minimal room for vibration, slip or twist; this helps preserve only the motions of the target are captured. In addition, an attachment convention for device orientation will assist in consistent and comparable datasets being gathered. The below is a suggested orientation convention for popular body mounting sites. With the exception of the left wrist, the USB port is configured to point towards the ground.

Mounting conventions

To fix the sensor to other body positions where a strap or clip is not feasible, Axivity recommends the use of Hypafix or Flexifix dressings.

Wrist mounting considerations

A popular site for mounting AX3 devices is the wrist. For mounting the device on the wrist, the Axivity Wrist Band is recommended. This accessory has a convention where the arrow on the Puck aligns with the arrow on the band.

Align arrows on Wrist Band and Puck

When in this location the user must consider the dexterity of the wearer. Depending on what sort of data is of interest to the user, the dominant, non-dominant or both hands may be chosen.

Typical use cases for gathering data from both hands include performing bi-manual skill assessments, fine-grained activity recognition and motor skill performance.

For single handed monitoring the choice between dominant and non-dominant hand must be based on if the data set will be used for activity classification and physical activity monitoring type applications (for which the non-dominant hand is a popular choice) or finer grained skill assessment of activities (for which the dominant hand is often used). Of course there are exceptions to these generalisations and the each experiment must be individually considered.

Data markers

When capturing experimental data it is often useful to place markers in the data. Such markers can retrospectively be used to identify certain event start and stops. A good marker should have the following characteristics;

  • Be orientation tolerant
  • Be easy to automaticaly identify in the data set
  • Occur over a short period of time
  • Be repeatable

A popular choice for creating a data marker is to subject the device to a short impulse force. Such forces can be generated through clapping with the device attached or held in the hand. The graph below is a graphical representation of a data marker generated by 5 short successive claps.

Data marker made up of 5 hand claps

When multiple devices are to be used in the same experiment and it is important that the captured data is retrospectively time aligned, subjecting multiple devices to the same data marking step will help this process.

Video annotation techniques

In some experiments it is desirable to video the target while wearing the AX3 sensors. This video stream can retrospectively be used to validate any inferences made about the captured data. To assist with this process it a useful technique is to place a data marker in the data stream and video the marker being placed. Once the experiment has finished being conducted, the video and the data stream can then be aligned based on the time sequence of the video and the timestamp of the data stream. A useful third-party software tool for performing this procedure is ELAN which is multi-platform, Open Source and available at https://tla.mpi.nl/tools/tla-tools/elan/.

Using Elan as an video/data annotation tool

Binary and CSV file types

The AX3 logs data internally in a binary packed format. This format is named Continuous Wave Accelerometer (CWA) format. This format is very efficient for storing large amounts of data but is not natively supported by many applications. In order to use the data in the CWA file there are two options:

  1. Import. Importing data involves reading it in within the program environment. While memory efficient (no intermediate files are needed), a certain amount of programming experience is needed. For many of the mor common programming environments (Matlab, Java, Python, PHP) scripts have been developed and are hosted on the Open Movement project site at www.openmovement.co.uk

  2. Convert. The majority of analysis or post processing environments support either Comma Separated Value (CSV) or Multi track audio (WAV) format files. The OMGUI software supplied with the sensors provides provision to convert the raw CWA files into either of these formats with a variety data interpolation and timestamp formatting options.