Assignment Activity Recognition

To do this assignment you will need access to a phone running Android, i.e. each group should have at least one person with an Android phone.

Download the (free) application Sensor Fusion from Google Play. It has been developed at LiU. More information is available here: http://sensorfusion.se/sfapp/ Links to an external site. Links to an external site.

Links to an external site.Make sure you have read about sensor imperfection modelling in the self-study material. You might also like to check this video out Links to an external site., illustrating the problem with long-time accuracy when using the IMU only (accelerometer + gyro. Adding a compass/magnetometer helps only slightly)

Your task is to implement an "app" (it is ok if it runs offline on saved data) that does activity recognition based on the data from the phone's sensors.

Requirements: The algorithm should be able to discriminate between, standing still, walking and running. You don't have to implement the algorithm on the phone, you can work offline on collected data. Handin a presentation where your solution is presented as well as code+data. 

You will peer-review this so thing about the other group that will get your handin.

Submit a zip file that contains

• Presentation
• Code, or link to accessible repository
• Data, or link to accessible repository
• README file that describes how to run 

A start could be to:

• Check what sensor/signals are available? Most likely GPS, Accelerometer, WiFi RSSI. Some might have gyro also.
• Log data when the phone is lying still and observe variations in sensor outputs.
• What sensors will be useful for the activity recognition

Make up task: If you submit the task after the deadline the zip-file should also contain an additional report that shows your work on Sensor error modeling. Log data when the phone is lying still. Write a small report  describing how to model observed variations in sensor output. The report should contain:

• • GPS (with good coverage). Plot histogram of position variations. Calculate  standard deviations in x,y and z. Are position errors normally distributed, are there many outliers? Any other interesting observations?
  • Accelerometer (try at least two different orientations of the phone). Calculate the norm of the accelerometer vector, does it correspond to the expected value of constant of gravity at your latitude? Plot histograms of the accelerometer signals. Are the variations normally distributed? Can discretization effects of the ADC converters be seen? Any other interesting observations?
  • Gyro (if available) Plot histograms and calculate standard deviations. Estimate the bias in the gyros. Are the errors normally distributed? Any other interesting observations?
  • Wifi RSSI (Try at least three different distances to the access point). Plot histograms and calculate standard deviations. Describe the variations. How accurate would you say the RSSI signal is as an estimator for the distance to the access point. Try both direct signal paths and situations with obstruction (e.g. walls).