Indoor-outdoor navigation without beacons:
Compensating smartphone AR positioning errors with
3D pedestrian network
9 March 2020
CRC 2020, Tempe, Arizona, USA

Jinying Xu, Fan Xue, Alain Chiaradia, Weisheng Lu, and Jin Cao
FoA, HKU, HK SAR

Outline
1

Introduction

2

The WaNAR Method

3

Discussion

Xu et al.: WaNAR positioning @ CRC2020, Tempe, Arizona

2

1. Motion, position, and indoor positioning
 Motion & position
 Core concepts in Newton’s laws
 With respect to a reference frame
o E.g., a passenger on a flying maglev train

 Indoor positioning

 “Where am I?” in a building / underground space
 Important for the AECO industry
o  but no satellite signals here …

Newton’s Laws (Source: phy-astr.gsu.edu)

 Possible reference frames
o Building structure
o Space

o Beacons

Xu et al.: WaNAR positioning @ CRC2020, Tempe, Arizona

Maglev motion

3

1. Related works
 Beacon-based methods are more expensive ($$$)
 Vision & Radio Frequency (RFID, Ultra-wideband (UWB), Bluetooth (BLE), & WiFi)

 Beacon-free methods are more prone to errors
 Sonic & magnetic field: Natural “beacons”
 PDR stepper & Augmented reality (AR): Integration of motion data over time

 Question: How to position indoor accurately and inexpensively?
Technique classes
Sonic
Magnetic
Vision
Radio Frequency (RF)
Pedestrian Dead
Reckoning (PDR)
Augmented reality (AR)

Performance Price
Examples
★★★
★
★★★
★★
★★
★★★★★ Marker, floor pattern, image-to-location reasoning
★★★★
★★★
Infrared, light, WiFi, BLE, GSM, UWB, etc.
★
★★★★★ Step counter + motion sensors
★★

Xu et al.: WaNAR positioning @ CRC2020, Tempe, Arizona

★★★★★ iPhone 11, Google Tango / Pixel, Huawei Mate 30P

4

1. Opportunity
 Indoor-outdoor motions are constrained
 For walking, driving, etc.
o E.g., No walking through walls

o Restricted in some areas (e.g., locked)

 3D walkablility network

Indoor motion example

 A digital reference for PDR stepper and AR

HKU Main Campus
(Source: Google Maps)

 Cheap: From BIM + GIS

 This paper
 A Walkable Network-based AR (WaNAR) positioning
o Beacon-free

o Accurate, by compensating AR’s errors

Xu et al.: WaNAR positioning @ CRC2020, Tempe, Arizona

HKU Main Campus 3D walkablility
Network (partial indoor + outdoor)
(Sun et al. 2019)

5

2. The WaNAR method
 Given a true position p0 sensed as pAR

procedure WaNAR_error_compensation:

 The “drifting” can be compensated as:

input translationAR, network3D, drifting_vec

 Green: on the path; Red: in perpendicular

way := nearest_path (network3D, translationAR)

 WaNAR is a loop of three steps

if distance (translationAR, tails_of (way)) > TURNING_BUFFER then
foot := perpendicular_foot (translationAR, way)

 1. To “snap” the pAR to a walkable path w

drifting_vec := drifting_vec + (translationAR – foot)

 2. To compensate the red if w not changed

if || drifting_vec || ≤ OFF_TRACK then

 3. To compensate the green if w changed
y

translationAR := translationAR – drifting_vec
end if
end if

pAR

w

return translationAR

p0

end procedure
Figure 2. Pseudo codes of the WaNAR error compensation algorithm

2D example
Xu et al.: WaNAR positioning @ CRC2020, Tempe, Arizona

x
6

2. Pilot test
 10 minutes walk at HKU Main Campus
 Outdoor: Knowles Bld.  outdoor  CYM Amenities Center
 Indoor: …  Knowles Bld. G/F

 Lobby  1/F  2/F

 Android app
 On Android Studio (version 3.1)
 3D walkablility network (Sun et al. 2019)
 Standard AR API (Google Tango, 2017)
 Flight mode: (as circled)

(a) Android APP

(b) Error compensation

(c) On slopes

 Active messages: (in boxes)
o Errors compensated
o On slopes/steps

Xu et al.: WaNAR positioning @ CRC2020, Tempe, Arizona

7

2. Demo video 1/2 (outdoor)
 Knowles Bld.  outdoor
 CYM Amenities Center
 Flight mode: On
 No RF signals

 AR drafting was
compensated
 Continuously
 Small

 Accurate, real-time
 E.g., from cross to cross

Video available
YouTubeTempe,
(link in
CRC Proceedings)
Xu et al.: WaNAR
positioningon
@ CRC2020,
Arizona

8

2. Demo video 2/2 (indoor)
 Knowles Bld.
 G/F Lobby  1/F  2/F

 Flight mode: On
 Worked well on steps,
corridors, and doors
 Failed in lifts
 Too high acceleration on z

 Limited in an open area
 Must follow a line

Video available
YouTubeTempe,
(link in
CRC Proceedings)
Xu et al.: WaNAR
positioningon
@ CRC2020,
Arizona

9

3. Discussion
 Pros
 WaNAR is accurate, cheap (a few modeler-hours), AR-ready
 Taking advantage of the walkability in BIM/GIS models

 Cons
 A location synchronization before use
 Limited to linear network in this paper
 Cannot handle lifts’ acceleration yet

 Future work
 Automatic location synchronization (e.g., a few BLE, RFID)
 More types of walkability spaces
 Integration of rear camera and POIs (e.g., voice navigation for visually impaired)
Xu et al.: WaNAR positioning @ CRC2020, Tempe, Arizona

10

Let BIM/GIS contribute
to smart cities!
For questions, pls Email
Frank: xuef@hku.hk
11