mobile sensing talk - kleomenis katevas · magnetometer no drift problem (depend on magnetic north)...
TRANSCRIPT
Introduction to Mobile Sensing TechnologyKleomenis Katevas
[email protected]://minoskt.github.io
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Image by CRCA / CNRS / University of Toulouse
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
• What is Mobile Sensing?
• Sensor data, but what can we actually sense?
• Sensor Fusion.
• Mobile Sensing on Android and iOS.
• Continuous Sensing Platforms.
• Mobile Sensing in Research.
In this talk
Camera(s)
Proximity Sensor
Ambient Light Sensor
Microphone(s)
GPS
Accelerometer
Gyroscope
Bluetooth Magnetometer
WiFi
NFC
Barometer
Water Sensor
Motion Coprocessor
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Satellite Navigation Systems• It calculates the location (latitude and longitude), altitude, and speed
of the device based on the distance from at least four satellites (trilateration).
• Average accuracy is ~3m.
• Not suitable for indoor positioning.
• GPS is the most popular one, but others are available or coming in the future (GLONASS, COMPASS, Galileo, IRNSS).
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
• Wireless networking technology.
• Can been used as an indoor positioning system (IPS) by triangulating the Received Signal Strength Indication (RSSI) of other Wi-Fi hot-spots.
• Some mobile devices combine GPS, cell tower triangulation and WiFi-based location to improve accuracy.
• Read “Survey of Wireless Indoor Positioning Techniques and Systems” by Hui Liu et al.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
• Wireless networking technology for exchanging data over short distances. (~10 meters for Class 2 radios)
• Can also be used for indoor positioning.
• Very useful for spatial-temporal data gathering (participant mobility and social interactions).
• Bluetooth LE (BLE), marketed as Bluetooth Smart, has very low battery consumption.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
iBeacon™• Apple’s implementation of Bluetooth
Smart, used as an indoor positioning system.
• Classifies the location of the device as Immediate, Near or Far.
• All devices can be iBeacon transmitters, receivers (or both), but the technology is also open to third-party hardware.
• Also available in Android using third-party libraries.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
iBeacon™ Data Format
Company ID (Apple: 0x004C)
Type(always 0x02)
UUID (16 byte hex)
Major(uint 0-36535)
Minor(uint 0-36535)
Measured Power(rssi in 1m distance)
Estimote Indoor Location
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Eddystone™• Google’s answer to iBeacon.
• Fully open-source specification (Apache v2.0) with source code available for Android and iOS.
• Three frames:
‣ Eddystone-UID (similar to iBeacon™)
‣ Eddystone-URL (broadcasts a URL)
‣ Eddystone-TLM (broadcasts telemetry information)
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Six degrees of freedom (3D Translation & 3D Rotation)
Some terminology
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Magnetometer
• Magnetic field sensor.
• Tells you the actual orientation of the device relative to the magnetic north (not the true north).
• Three axis magnetometer:x, y, z (East, North, Up)
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Magnetometer
• Magnetic field sensor.
• Tells you the actual orientation of the device relative to the magnetic north (not the true north).
• Three axis magnetometer:x, y, z (East, North, Up)
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Accelerometer• A sensor that measures the force of
acceleration of the device.
• 3 axis accelerometer:x, y, z (Surge, Heave, Sway)
• Can also calculate Pitch and Roll (but not Yaw!) using trigonometric calculations.
• Also used to understand the device orientation, by measuring the acceleration caused by the gravity.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Accelerometer• A sensor that measures the force of
acceleration of the device.
• 3 axis accelerometer:x, y, z (Surge, Heave, Sway)
• Can also calculate Pitch and Roll (but not Yaw!) using trigonometric calculations.
• Also used to understand the device orientation, by measuring the acceleration caused by the gravity.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Gyroscope
• 3 axis gyroscope:x, y, z (Pitch, Roll, Yaw)
• Tells you how much your device is being rotated over time.
• Less computationally expensive Pitch and Roll.
• Also provides Yaw.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Demo time…
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Comparing the three sensors+ -
Magnetometer No drift problem(depend on magnetic north)
Noisy. Hard to calibrate.Not accurate for rotation detection.
Tilt compensation.
Accelerometer No drift problem(depend on earths gravity)
Noisy, especially in high frequencies.Mixes linear acceleration and gravity.
GyroscopeSmooth reading even in high
frequencies.Great dynamic response.
Large errors due to the drift problem.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
What is sensor fusion?
• “The use of multiple sensors so that they compensate each other’s weaknesses.”
• Users (developers) do not want to know about sensors. What they really want is accurate and responsive 3D translation, 3D rotation and orientation.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Accelerometer and Gyro Fusion
(Sachs, 2010)
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Accelerometer and Gyro Fusion
(Sachs, 2010)Integration
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Sensor Fusion (Sachs, 2010)
Accelerometer
Gyroscope
Gravity
Orientation 3D Rotation
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Sensor Fusion (Sachs, 2010)
Accelerometer
Gyroscope
Gravity
Orientation 3D Rotation
Integration
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Magnetometer and Gyro Fusion
(Sachs, 2010)
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Sensor Fusion (Sachs, 2010)
Accelerometer
Gyroscope
Gravity
Orientation 3D Rotation
Magnetometer Direction
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Gravity and Linear Acceleration
• Linear Acceleration = Accelerometer Data - Gravity
• In most cases, it is very hard to separate Gravity from Linear Acceleration as gravity changes really fast.
• High pass filters do not work very well.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Sensor Fusion (Sachs, 2010)
Accelerometer
Gyroscope
Gravity
Orientation 3D Rotation
Magnetometer Direction
Linear Acceleration 3D Translation
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Sensor Fusion (Sachs, 2010)
Accelerometer
Gyroscope
Gravity
Orientation 3D Rotation
Magnetometer Direction
Linear Acceleration 3D Translation
Double Integration
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Motion Coprocessors• Apple’s M7 / M8 / M9 chip.
• Continues measures motion data from Magnetometer, Accelerometer, Gyroscope and Barometer sensors.
• Automatically performs activity detection: stationary, walking, running, automotive, unknown. M8 also provides distance and elevation using the Barometer sensor. M9 can also track Pace and Cadence.
• Apps can query data of the past 7 days when loaded.
• Does not provide the raw data.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Sensing Motion in iOS and Android
• Raw and Calibrated data of:
- Accelerometer
- Gyroscope
- Magnetometer
• Fused data of:
- User Acceleration
- Gravity
- Attitude
- Rotation rate
- Magnetic field
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Mobile Sensing on Android• Sampling Rate in motion sensors is not consistent.
• Timestamp is not consistent.
• Lack of a motion coprocessor, only Pedometers.(this will change very soon…)
• Android fragmentation problem.
• Bluetooth Smart (BLE) is only fully supported in Lollipop.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Sampling rate on Android
Samsung Galaxy S2 Google Nexus 4
FASTEST 100 Hz 200 Hz
GAME 50 Hz 50 Hz
UI 15 Hz 30 Hz
NORMAL 5 Hz 15 Hz
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Mobile Sensing on iOS
• Running a process in the background is (almost) not possible.
• Access to some sensors is limited (e.g. WiFi, NFC).
• Restrictions in the App Store.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
SensingKit Server
Web
Ser
vice
s
Com
mun
icat
ion
Mod
ule
Plug-inSystemModel
Manager
SensorModules Data
Data DataExtraction
Internet
SensingKit: A Multi-Platform Mobile Sensing Framework
• Works in Android and iOS mobile systems.
• Captures Motion, Location, Proximity, Environmental data.
• Power efficient using Bluetooth Smart (4.0).
• Easily extensible using a modular design.
• Automated time sync and data processing on the server.
• Available in open-source under the GNU LGPL v3.0.
Platform Characteristics
For more info, check www.sensingkit.orgCognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
SensingKit: Battery Consumption
Battery consumption of SensingKit running on an iPhone 5S.
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
00 03 06 09 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54Time (hours)
0
10
20
30
40
50
60
70
80
90
100
Pow
er
(%)
iPhone 5S
IdleAccelerometer (100Hz)Gyroscope (100Hz)Magnetometer (100Hz)Device Motion (100Hz)Location (Best Accuracy)iBeacon Broadcast (1Hz)iBeacon Scan (1Hz)iBeacon Scan + Broadcast (1Hz)Microphone (44100.0 Hz)
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Detecting Group Formations
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
141
142
143
144
145
146
147
148
149
150
time (sec)
P1
P2
P3
P4
P5
P6
P1
P2
P3
P4
P5
P6
P1
P2
P3
P4
P5
P6
P1
P2
P3
P4
P5
P6
P1 P2 P3 P4 P5 P6
P1 5.72 21.78 29.07 39.95 30.60
P2 5.72 15.45 29.06 19.67 25.41
P3 21.78 15.45 4.60 3.94 1.22
P4 29.07 29.06 4.60 10.49 2.43
P5 39.95 19.67 3.94 10.49 3.04
P6 30.60 25.41 1.22 2.43 3.04
P1 P2 P3 P4 P5 P6
P1 0 0 0 0 0
P2 0 0 0 0 0
P3 0 0 1 1 1
P4 0 0 1 0 1
P5 0 0 1 0 1
P6 0 0 1 1 1
(a)Weighed Adjacency Matrices
(b)Binary Adjacency Matrix
(c)Graph
}
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
Results
Confusion Matrix
Thank you!
Cognitive Science Research Grouphttp://cogsci.eecs.qmul.ac.uk
David Sachs, “Sensor Fusion on Android Devices: A Revolution in Motion Processing”, Google Tech Talk, February 2010. Retrieved from https://www.youtube.com/watch?v=C7JQ7Rpwn2k.
Kleomenis Katevas, Hamed Haddadi, Laurissa Tokarchuk, Richard G. Clegg, "Detecting Group Formations using iBeacon Technology", 4th International Workshop on Human Activity Sensing Corpus and Application (HASCA2016) in conjunction with UbiComp 2016, September 2016, Heidelberg, Germany.
Kleomenis Katevas, Hamed Haddadi and Laurissa Tokarchuk, “SensingKit: Evaluating the Sensor Power Consumption in iOS devices”, 12th International Conference on Intelligent Environments (IE'16), September 2016, London, UK.
Kleomenis Katevas, Hamed Haddadi and Laurissa Tokarchuk, “Poster : SensingKit -- A Multi-Platform Mobile Sensing Framework for Large-Scale Experiments”, Extended abstract, ACM MobiCom 2014, Maui, Hawaii, September 2014.