1. Indoor Localization and Navigation of Wheelchair Users with Smartphones
Ruolin Fan, Silas Lam, Emanuel Lin, Oleksandr Artemenkoⱡ, Mario Gerla
University of California, Los Angeles (UCLA)
{ruolinfan, silaslam, emanuel,
ⱡIlmenau University of Technology

2. Outline Introduction Background System Design Implementation
Evaluation
Conclusion

3. Introduction GPS does not work indoors
Lack of satellite signals
Need an alternative way to position ourselves indoors
Try to utilize unique features pertaining to wheelchairs
Transform measured wheel rotations into both distance and angular displacement
Crowd sourcing popular wheelchair access paths
Useful for blind/impaired wheelchair riders
More motivations: crowd sourcing popular wheelchair access paths
Blind/impaired wheelchair riders

4. Background: Indoor Localization
Triangulation methods from cellular, WiFi, or acoustic (Signal strength or signature)
Require landmark placement knowledge, previous mapping of the site; affected by obstacles
Dead reckoning
Compute the current position based on a previously known position and incremental displacement
Can complement and rescue GPS and triangulation methods (eg Autogait[Percom 10])

5. Wheelchair Dead Reckoning - Overview
Get initial position of the wheelchair via GPS coordinates or other means
Mark the wheels on the wheelchair at each spoke
Track the wheelchair’s movements by counting rotations of the wheels using the marks (a “tick”)
Simple model (perfect traction, no sliding):
If wheels rotate at the same rate => straight movement
If wheels rotate at different speeds => turns

6. Inferring Movements Straight forward movement: Sharp turns:
Both wheels move at the same rate
cwheel: the wheel’s circumference
n: the number of marks on each wheel
Sharp turns:
One wheel is moving while the other stays still
wchair: the width of the wheelchair
cchairTurn: The circumference when the chair turns a full circle
dtravelled: The distance travelled by the turning wheel

7. Inferring Movements (Cont’d)
General Turns
One wheel moves faster than the other
Derive equation using radians ,
And therefore
In degrees,

8. Implementation Wheelchair Specifications 8 magnets per wheel
1 reed switch per wheel
Reed switches connected to Bluetooth mouse
When magnet moves close to reed switch, it trigger a mouse click event

9. Implementation (Cont’d)
Translate left/right mouse clicks to distance/direction traveled
Base calculations on physical wheelchair measurements
Implemented straight movement and sharp turns
Clicks detected by JavaScript in web browser
Events are sent via AJAX to PHP server and MySQL database
Visualize wheelchair movement on a map

10. Implementation Challenges
Wheels are not always synchronized together
Magnets are far apart from one another
Result: coarse-grained data
Wheels may “slip” due to physical imperfections

11. Our Solution (can you explain better please??)
Find ways to do “approximately equals”
Made our own low-pass filter in counting the clicks
Single values that look like (1,0) would behave like (1,1), and pairs like (1,0),(0,1) would also behave like (1,1)
Count small turns as straight movements until confirmed to be a turn
When a turn is confirmed, backtrack the last forward movement and aggregate the turn
Basically this takes care of the problem when you have left and right wheels not having the same number of ticks even when they are moving forward, due to either missed clicks or the wheels being unsynchronized. You can take a look at the “Implementation” section of the paper, with the paragraph beginning with “Unfortunately, this simple design does not work as intended by itself…”, it should explain quite clearly what’s going on here and in the next two slides.

12. Example Forward
| time | state | magnitude | | ... | ... | ... | | | F | | | | F/R | | | | F | | | | F | 0 | | | F/L | | | | F | | | | F/L | | | | F | |

13. Example Turn
| time | state | magnitude | | ... | ... | ... | | | F/L | | | | F | | | | L | | | | L | | | | L | | | | F | | | | F/L | | Total turn = degrees

14. Evaluation - General Movements
Move the wheelchair around Boelter Hall 3rd floor, the main engineering building at UCLA
Straight forward movement is accurate
Turns are off
Only 8 magnets on a wheel: can only measure degrees in increments of 24.5
The closest to a 90 degree turn is 98 degrees

15. Evaluation – Straight Movements
Error Rate vs. Travelling Speed
Increasing the travelling speed can cause a decrease in accuracy, but the approximate equality filter greatly increases accuracy
Error rate is the absolute value of the difference between actual distance travelled and the measured distance, divided by the actual distance travelled

16. Evaluation – Straight Movements (Cont’d)
Error Rate vs. Update Period (Fast Travelling Speeds)
Increasing the update frequency can lead to more accurate results for straight forward movements, both with and without filters

17. Improving Turn Accuracy assuming blue print is known
Right angle correction
Assume 90 degree turns when the turning angle is close to it
Correction via boundary detection
Detect building boundary and make corrections accordingly
Projected results
Projected results

18. Conclusions
Indoor localization with a wheelchair can be accomplished by translating wheel rotation measurements into distance and direction
Accuracy is high for slow to medium speeds, but decreases as speed goes up
Improvements can be made by simply adding magnets
Successful proof of concept project

19. Future Work
Improve the accuracy by exploiting existing smartphone sensors:
Compass, altimeter (in a multilevel building), gyroscope, accelerometer
Synergize wheelchair dead reckoning with WiFi signature methods
The wheelchair is used as surveyor, to calibrate the signatures

20. Thank You!