quartz/content/notes/11-orientation sensors-2.md

title	tags
11-orientation sensors-2	lecture info305

[!INFO] we need sensors to that computing can be ubiquitous. from this we can think of completey new interfaces

[!INFO] gyroscope say a phone is rotating by 5 degree per seconds. usualy want to know absolute angle. this is the integral over time, we know what the angular rate is and the time. problem: cannot take perfectly continuous integral. they do more than 100 measurements per second. gryoscope data changes faster than samplling frequency. this is called drift/bias, as the error accumulates. when the sensor is stationary, it still measures a small value, which is inherent in accuracy. so the sensor thinks it is moving when it is not. the bell curve of inaccurate measurements is not centred.

Advanced sensing

• Current situation:
  • Many different sensors with different characteristics
    • Precise but slow, noisy but fast, drift biased,…
    • We cannot measure everything…
• Can we combine sensors for model more complex states?
• Can we filter sensor measurements to improve the data quality?

[!INFO] this is what is happening, there is a lot of processing on top of the raw data. we combine sensors to get better results

Sensor Fusion

• The idea:
  • Combine (fuse) sensor to achieve better results
  • Better Location (When signal ist lost, not enough satellites, ..)
  • Better Orientation (Less drift (gyroscope), less error prone (compass), ..)

[!INFO] e.g., fuse wifi signal with GPS to make AGPS.

Sensor Fusion

• Example for Orientation
  • Combine strength of several integrated sensors
  • Avoids gyro drift and noisy orientation
    • http://www.thousand-thoughts.com/2012/03/android-sensor-fusion-tutorial/
  • Gyroscope output is applied only for orientation changes in short time intervals (high-pass filtering)
  • Accelerometer/Magnetometer is used as support information over long periods of time for absolute orientation (low pass filtering)

[!INFO] when phone is still, use the magnetometer, when moving faster use the gyro low pass filter, lets the low frequencies through high pass filter, lets high frequencies through

[!INFO] you can create more sophisicated algorithms, but these usually introduce latency

Dead Reckoning

• Dead Reckoning
  • Estimates current position using
    • Previously known position
    • Speed
    • Orientation/direction
    • Elapsed time
• Used in all kinds of navigation (Marine, Aerial, Car, Robot, Pedestrian, .. )

[!INFO] guessing where you are based on you previous movements

• Starting with a reference point (known position)
• Detect relative movements
• Smartphone sensors can be used
  • Accelerometer – step detection (distance)
  • Magnetometer and gyroscope (orientation)
  • Barometer (level changes)

[!INFO] GPS in cars sometimes just put you on the nearest road robots use a lot of dead reckoning

• Starting with a reference point (known position)

• Detect relative movements

• Smartphone sensors can be used

  • Accelerometer – step detection (distance)
  • Magnetometer and gyroscope (orientation)
  • Barometer (level changes)

• Issues:

• Even small errors lead to incorrect positions

• Subsequent errors increase incorrectness

• Orientation is biggest issues

• Periodic recalibration with new reference point required

[!INFO] detect when you take a lift as a calibration point