Investigating the Combination of Location Sensing Technologies

1
Investigating the Combination of Location Sensing
Technologies
Nathan Lemieux Supervisor Hanan Lutfiyya MSc
Thesis Proposal
2
Definitions

• Context
• describes the state of the environment where the
  application is used. e.g. computing, user,
  physical, time and location
• Location
• describes a space or area in which an object may
  be found. Location information can be either
  absolute (GPS) or relative (landmark)
• 43.0071857N, 81.2889752W (UWO)
• Clock Tower on Campus (Middlesex College)

3
Motivation

• Ubiquitous Applications
• Collect context or environmental information,
  process it and use this derived data to deliver
  valuable services
• Location is a significant context which is
  already used in outdoor applications
• Location-Aware Applications
• Navigation and Guide Services
• Follow-me services
• Asset (objects or people) tracking and monitoring
  (presence or availability) services
• Advertising

4
Motivation

• Current outdoor localization (GPS, Cellular)
  technologies fail to work indoors
• Line of sight requirements and location accuracy
• Current accurate and precise indoor localization
  technologies require installation of a dedicated
  infrastructure or the deployment of tags
• Expensive (user and provider)
• Scalability

5
Current Indoor Location Technologies and
Techniques

• Signals
• Techniques
• Trilateration, Multilateration, Proximity and
  Fingerprinting
• Technologies
• Wi-Fi, Bluetooth, Ultrasound, Infrared, RFID
• Image Analysis
• Techniques
• Machine learning, Direct Image Comparison
• Technologies
• Camera phones, webcams, CCTV, stereo cameras
• Sensors
• Techniques
• Dead Reckoning, Activity Recognition, Proximity
• Technologies
• Accelerometers, compasses, altimeters, pressure
  and light sensors

6
Summary From Reading Course

• Every technology or technique has disadvantages
• Next evolution in indoor location sensing is the
  creation of hybrid systems that combine two or
  more technologies and/or techniques
• Hybrid systems should be designed such that the
  advantages of one technology or technique are
  offset by advantages of another

7
Related Work Indoor Hybrid Systems

• Easy Living (Microsoft)
• Proposed using different technologies (GPS,
  infrared/ultrasound, stereo vision) to provide
  location estimations at different levels
  (building, room, object)
• Place Lab (Intel)
• Relies on the signal fingerprinting technique of
  Wi-Fi access points, cell towers, Bluetooth
  devices
• GETA Sandals
• Combination of Dead-reckoning and RFID.

8
Drawback of Existing Work

• Prototypes were developed in controlled
  environments
• They require specialized hardware or proprietary
  software
• Installation and deployment costs
• User costs
• Reliance on only signaling techniques and
  technologies

9
More Definitions

• Location Accuracy
• How much, in terms of meters, is the estimated
  users position is deviate from the users true
  position.
• Location Precision
• The percentage of the time the location system
  provides the given accuracy
• Waypoint/Checkpoint
• A distinctive reference point that can be used
  for navigation. A specific location or easily
  identifiable area.

10
Proposed Hybrid System

• Combination of Indoor location Sensing
  Technologies and Techniques
• 802.11 Wi-Fi Access Points
• Fingerprint and Proximity signaling techniques
• Image Analysis
• Discriminative Classifier generated by Machine
  learning
• Sensors
• Altimeter, Proximity and Accelerometers

11
Proposed System

• Why Wi-Fi?
• Already Existing 802.11 Infrastructure
• Users do not have to purchase any specialized
  hardware
• Research has demonstrated that Wi-Fi signals can
  be effective for indoor localization
• Herecast, Place Lab, Ekahau

12
Proposed System

• Wi-Fi
• Look at two well known signaling techniques for
  comparison results
• Proximity with location accuracy between 25 and
  50 meters
• Received Signal Strength Indicator (RSSI)
  Fingerprinting (1,2 and 3 Access Points) with
  location accuracy of 5 to 10 meters

13
Proposed System

• Why image analysis?
• To complement the Wi-Fi system by reducing
  location error caused by fluctuating RSSI values
• Has been demonstrated in research to be fairly
  effective
• Indoor localization using camera phones
• Room level accuracy 80 of the time
• Probabilistic Location Recognition using reduced
  Feature set
• Classified different hallway locations 95 of
  the time

14
Proposed System

• Image Analysis
• User wears a webcam to capture images
  periodically
• Images will then be classified as a particular
  scene depending on environment using Pixit
  Software
• Classroom, hallway, office, computing lab,
  conference room, group office, stairwell, outside
• Kitchen, Living room, bedroom, bedroom2, hallway,
  dining room, stairs, outside
• Location accuracy will be room level
• If there are several scene of the same type in
  close proximity and second Multiple Same Scene
  Classifier could be used

15
Scene Analysis

• Pixit Software
• Out-of-the-box software for image classification
• Uses random sub-window extraction and machine
  learning decision tree ensemble technique
  proposed by Raphaël Marée
• Able to distinguish between multiple classes not
  just Boolean classification
• Robust to uncontrolled conditions
• Illumination, scale, orientation, occlusions
• Performs successfully on numerous already defined
  image datasets
• Provides a confidence level for each scene used
  in creating the classifier
• Ability to adjust and try many different
  parameters

16

• An example of a classified image using a
  classifier created by the Pixit software

17
Proposed System

• Why Sensors?
• Usually low-cost
• Research has shown that Altimeters can be used to
  locate a user to a specific floor
• Research has shown that Accelerometers can be
  used for simple user activities (walking,
  running, jumping, stationary)
• Honeywells DRM-5 Dead reckoning module estimates
  users distanced traveled and direction of
  movement. Accuracy is about 2 percent of
  distanced traveled since last position fix.
  Position fixes can be generated by GPS or another
  external source.

18
Example Scenario

• Wi-Fi

Scene Analysis
Sensors
Provide altitude, dead-reckoning position
estimation with probability.
19
Example Scenario
Gathers information from all other components to
infer the best possible location

• Decision
• Component

A second classifier could be used to provide
even more additional information about location
probability
20
Why these Technologies?

• Cellular phones will eventually have all theses
  technologies built-in (Wi-Fi, Camera,
  Accelerometers).
• Disadvantages of Wi-Fi, fluctuating signals
  values, can be offset by scene analysis to
  increase location precision.
• Secondly combination of Wi-Fi and Scene analysis
  could be used to create checkpoints/waypoints to
  reset drift error in dead-reckoning sensors.
• Also, additional technologies allow for the
  possible creation of more interesting
  applications.

21
Drawbacks of the Selected Technologies

• Wi-Fi
• No Control of Access Points in some environments
• Suffers from multi-path propagation, signal
  reflection and fading

22
Drawbacks of the Selected Technologies

• Image/Scene Analysis
• Image capturing issues
• lighting, orientation, motion and occlusions
• Each environment requires its own scene
  classifier
• Have to wear a camera

23
Drawbacks of the Selected Technologies

• Sensors
• Suffers from constant growing positioning error
  known as drift
• Drift-error occurs due to slight errors
  introduced in the manufacturing process of
  sensors
• The error is small but the error is accumulative

24
Potential Achievements

• Show that the location system has the ability to
  work effectively in multi-floor buildings which
  have minimal to no control over the environment
• illustrate that the location system has the
  ability to work in different environments
• To compare results of the singleton technology
  location systems to the results of different
  combination of hybrid systems
• Demonstrate that location accuracy increases as
  technologies are combined
• Show that disadvantages of technologies can be
  offset by other technologies
• Create location waypoints/checkpoints based on
  location probability to reduce/reset drift error
  in dead reckoning sensors

25
Questions ?