Scribd
Upload
35 views

Short-Term Motion Tracking Using Inexpensive Sensors: Filip Matzner, Roman Barták

Short-term motion tracking is possible using inexpensive sensors in smartphones like accelerometers, gyroscopes, and magnetometers. The document outlines sensor fusion techniques to integrate sensor data from these devices into estimates of position over time in 3D space. Experiments show tracking works for short motions but sensor inaccuracies lead to drift over longer periods. Stabilization using gravity and magnetic field fixes can help but advanced models like Kalman filters may further improve reliability.

Uploaded by

derre
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
35 views

Short-Term Motion Tracking Using Inexpensive Sensors: Filip Matzner, Roman Barták

Short-term motion tracking is possible using inexpensive sensors in smartphones like accelerometers, gyroscopes, and magnetometers. The document outlines sensor fusion techniques to integrate sensor data from these devices into estimates of position over time in 3D space. Experiments show tracking works for short motions but sensor inaccuracies lead to drift over longer periods. Stabilization using gravity and magnetic field fixes can help but advanced models like Kalman filters may further improve reliability.

Uploaded by

derre
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 19

Short-term Motion Tracking Using

Inexpensive Sensors
Filip Matzner, Roman Barták
Introduction
● Sensors became extremely small, lightweight, low-
power and cheap.
● In theory, having perfect inertial sensors, it is possible
to track the full 3D motion.
● Is it possible in practice with cheap low-end sensors
present in a smartphone?
Outline
● Sensor Description
o What sensors do we need and what data they produce.

● Sensor Fusion
o Fusion of the sensor data into the position.

● Experiment with a Smartphone


o Does it work? Try yourself!
Sensors
● MEMS (micro-electro-mechanical-systems)
● Usual combination present in smartphones is:
o accelerometer
o gyroscope
o magnetometer
Sensors - accelerometer
● Measures acceleration of the device in three
coordinate axes [ms-2]. raw 3-axis accelerometer data

2D accelerometer concept upward acceleration


Sensors - accelerometer (gravity)
● Every accelerometer is affected by a permanent
gravity force.
● The gravity force cannot be distinguished from upward
acceleration.
Sensors - gyroscope
● Historical gyroscopes were able to produce absolute
orientation in space.
● MEMS gyroscopes only produce angular speed [rad/s]
for each coordinate axis. raw 3-axis MEMS gyroscope data

historical gyroscope concept


Sensors - magnetometer
● Measures magnetic field vector, i.e., magnetic field
strength for each coordinate axis.
● Sensitive to an interference caused by WiFi, electrical
wires, etc.
raw 3-axis magnetometer data
Sensor Fusion - coordinate systems
● We will be working with two coordinate systems.
o global-coordinates: the system the device was in at the beginning
o device-coordinates: the system fixed to the device
device-coordinates

global-coordinates
Sensor Fusion - problem specification
● We have the following data in device-coordinates:
o acc(t) = accelerometer vector in time t
o gyro(t) = gyroscope vector in time t
o mag(t) = magnetic field vector in time t

● We want to fuse the data to get the position of the


device in time t in global-coordinates.
o pos(t) = position vector in time t
Sensor Fusion - theoretical model 1
The device does not rotate during the entire motion.
● First some basic definitions from physics:

o But we only have the accelerometer data!


 We will use the definition the other way around:
Sensor Fusion - theoretical model 1
The device does not rotate during the entire motion.
● The accelerometer vector also measures gravity, which
has to be subtracted:

● Now we have the formula for the position of the device


as long as it does not rotate.
Sensor Fusion - theoretical model 2
The device rotates during the motion.
● If we allow rotation, the sensor data from different time
can be relative to different coordinate systems.
● To convert the data from device-coordinates to global-
coordinates, we need to know the orientation of the
device.
● The orientation will be represented by a 3x3 rotation
matrix.
Sensor Fusion - Orientation
● The orientation will be a rotation matrix representing
the rotation of the device-coordinate system relative to
the global-coordinate system.
● The matrix will be created by sampling the data from
the gyroscope into rotation matrices and multiplying
the matrices together.
Sensor Fusion - the final formula
● Putting the orientation formula and the simple position
formula together gives us the final formula for the
position including the rotation of the device.
Sensor Fusion - algorithm
Automated Stabilization
● The gyroscope bias introduces a drift, which can be
compensated by gravity and magnetic field.

gravity fix

magnetic field fix


Experimental Evaluation
● We have developed a software and performed a series
of experiments.
Conclusion
● The cheap sensors are not accurate enough to track
long motions.
● If the motion can be separated into multiple short
intervals, the automated stabilization can prolong
reliability.
● A model method, such as Kalman filter, might improve
the results and reduce noise.

You might also like