Indoor-outdoor navigation without beacons: Compensating smartphone
AR positioning errors with 3D pedestrian network
Jinying XU1, Fan XUE2, Alain CHIARADIA3, Weisheng LU4 and JIN CAO5
This is the authors’ version of the publication:
Xu, J., Xue, F., Chiaradia, A., Lu, W., & Cao, J. (2020). Indoor-outdoor navigation
without beacons: Compensating smartphone AR positioning errors with 3D
pedestrian network. Proceedings of ASCE Construction Research Congress 2020,
ASCE, in press. Doi: LINK_TO_INSERT

5

6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37

1

PhD candidate, Department of Real Estate and Construction, The University of Hong
Kong, Pokfulam, Hong Kong SAR; PH (852) 6025-1696; FAX (852) 2559-9457;
email: jinyingxu@connect.hku.hk
2
Assistant Professor, Department of Real Estate and Construction, The University of
Hong Kong, Pokfulam, Hong Kong SAR; PH (852) 3917-4174; FAX (852) 25599457; email: xuef@hku.hk
3
Associate Professor, Department of Urban Planning and Design, The University of
Hong Kong, Pokfulam, Hong Kong SAR; PH (852) 3917-2721; FAX (852) 25590468; email: alainjfc@hku.hk
4
Associate Professor, Department of Real Estate and Construction, The University of
Hong Kong, Pokfulam, Hong Kong SAR; PH (852) 3917-7981; FAX (852) 25599457; email: wilsonlu@hku.hk
5
Research Assistant, Department of Real Estate and Construction, The University of
Hong Kong, Pokfulam, Hong Kong SAR; PH (852) 5225-5190; FAX (852) 25599457; email: caojin@hku.hk
ABSTRACT
Despite the extensive use of positioning and navigation in outdoor space, indoor
positioning and navigation systems, essential for intelligent building and smart city
services, are unsatisfactory in either performance or price, sometimes in both. This
paper analyzes and compares the performances and prices of existing indoor
positioning technologies that are categorized into a few classes according to their
spatial sensing and referencing methods. Based on the previous work in walkability,
this paper proposes a novel Walkability network-based Augmented Reality (WaNAR)
method using smartphones with AR positioning function for positioning and
navigation. In WaNAR, drifting of the AR positioning signals are corrected
continuously by the ground-truth 3D indoor/outdoor walkability network (e.g., nobody
is supposed to walk through a wall) in a 3D model. The error at the vertical axis of the
walking direction is corrected continuously and that of the walking direction is
compensated at every turn. WaNAR can be used in both indoor and outdoor navigation,
1

38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84

its performance and price are proved to be largely improved compared to existing
technologies. The only investment for a typical building is a 3D drawing of indoor
walkable space in a few staff-hours. WaNAR has broad application prospects at various
positioning and navigation scenarios.
Keywords
Augmented reality positioning, pedestrian network, error compensation, indoor
navigation, indoor-outdoor integration
INTRODUCTION
Location is becoming a prerequisite for various smart applications, be they
navigation or weather reporting services. Therefore, the sensing of locations has
attracted vast interests from both researchers and practitioners. Actually, the contactless
location-sensing technologies have developed rapidly within recent years. GPS (Global
Positioning System) is the most widely used technology in positioning and navigation.
However, it is suffering from its weakness in indoor positioning (Xu et al., 2019). To
overcome such limitations, radio signal based technologies including RFID, Ultrawideband (UWB), Bluetooth, and WiFi are meanwhile explored to be used for
positioning, mainly at indoor environments. Relying on specific radio devices and a
remote server makes such approaches expensive and not hard to use and therefore
hinders their widespread use. With the emerging of AR (augmented reality), a new ARbased positioning solution was proposed. AR, as a real-time interactive user interface
technology that augments the user’s real environment with computer generated virtual
entities in 3D (Xue et al., 2018), is widely embedded in smartphones. AR applications
can thus benefit from smartphone sensors, e.g., accelerometers and magnetometers, to
facilitate positioning. Most AR positioning solutions use object detection and
recognition techniques and consequently require reference databases of 3D virtual
objects or images, which is time-consuming and not accurate (Paucher & Turk, 2010).
3D walkable network is the 3D network of walkable roads, streets, tunnels,
footbridges, stairs, elevators, lifts, etc (Sun et al., 2019). It contains the connectivity,
Euclidean and geometric relationship between pedestrian path segments (e.g.,
sidewalk, crosswalk, and footpath), as well as other path characteristics such as, for
example, path width. It is believed to have the potential in a variety of applications such
as pedestrian navigation systems/services, urban planning and urban design. However,
2D map services usually fail to provide accurate and interactive 3D walkable network
for pedestrians. 3D maps also suffer from the lack of mature integrated indoor-outdoor
navigation technology. Pedestrians have to explore by themselves or ask for other
pedestrians in a complex and 3D high density city. A visible and interactive 3D
walkable network is in desperate demand.
This paper aims to develop an accurate and novel Walkability network-based
Augmented Reality (WaNAR) positioning method using calibration of ad hoc 3D
pedestrian network for seamless indoor-outdoor positioning and navigation at a very
low cost. In WaNAR, drifting of the AR positioning signals is corrected continuously
(every 5 seconds) by the ground-truth 3D indoor/outdoor walkability network (e.g.,
nobody is supposed to walk through a wall). The calibration eliminates the vertical
distance of the pedestrian ad hoc location and his/her nearest walkable line and directly
adjust his/her location to the nearest walkable line. By doing so, two hidden
2

85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129

assumptions are made: pedestrians always walk along with the walkability network,
and the deviation of AR positioning along a walkable line is within 5%. WaNAR is
very accurate, easy to implement, and inexpensive, where the only investment for a
typical building is a 3D drawing of walkable indoor space in a few staff-hours. In
contrast, the popular radio frequency (RF) beacon-based methods such as Bluetooth
Low-Energy (BLE) and UWB are expensive, heavy in carbon footprint, and hard to
manage. It can be used for seamless indoor-outdoor navigation, facility management,
and intelligent business.
The rest of the paper is organized as follows. Followed by the Introduction,
related works on AR positioning methods, indoor positioning and indoor-outdoor
integration methods, and 3D model-based error calibration methods are reviewed.
Afterward, the algorithm of compensating smartphone AR positioning errors with 3D
pedestrian networks are introduced. A method is then presented thoroughly based on
AR smartphone hardware and software architecture, with a pilot study carried out at a
university campus. Finally, a conclusion and future work are drawn.
RELATED WORK
AR Positioning Methods
AR, augmenting the user’s real environment with computer-generated 3D
virtual entities, is often jointly used with location sensors for mobile applications,
especially for facility maintenance (Koch et al., 2014) and emergency management
(Bellini et al., 2014). Those mobile applications are mainly designed for smartphones,
which are ubiquitously embedded with various advanced sensors and technologies. The
ever-enhancing hardware and computing capacities make smartphones perfect
platforms for high-tech applications. Therefore, mobile AR applications can take
advantage of smartphone sensors such as gyroscopes, digital compasses,
accelerometers, and magnetometers to improve its performance and facilitate user
tracking (Paucher & Turk, 2010).
In AR community, there are several methods used to do positioning, including
marking, pose estimation, and SLAM (simultaneous localization and mapping). The
most used one is marking the objects in the environment with a unified code. This
method is robust and requires low computing capacity, but meanwhile requires
additional marking works and thus leads to the increase of investment and difficulties
of promotion (Paucher & Turk, 2010). The pose estimation method is also very
commonly used with the pervasive adoption of 6 DoFs (degree of freedoms) position
sensor and the increasing computing capacity of smartphones. It also requires an image
database of the environment and thus is constrained in large and unknown
environments. The SLAM method tracks the user’s location by constructing a map onthe-fly using several different sensors, mainly optical sensors such as 2D camera or 3D
laser scanner. It is widely used in unknown environments for robots, UAV (unmanned
aerial vehicles), and self-driving cars, etc. Since the sensing and processing of the
mapping and positioning data require powerful computers, such an approach is now
more available for small scenarios. To conclude, most current AR-based positioning
and navigation methods rely on either high-quality tracking of a small, constrained

3

130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156

157
158
159
160
161
162
163

environment with given tracking devices or low accuracy outdoors environment only
with GPS that delivers positional information worldwide.
Indoor Positioning and Indoor-outdoor Integration Methods
Despite the extensive use of positioning and navigation in outdoor space, indoor
positioning and navigation systems, vital for 3D cities, are unsatisfactory in either
performance or price, sometimes in both. Existing indoor positioning technologies can
be categorized into a few classes according to their spatial sensing and referencing
methods, see Table 1. We compared the performance (accuracy and easiness to use)
and price (from the user side) of different techniques. Sonic signals can be divided into
audible sound and ultrasound, though they are cheap in use but can only reach a
decimeter level accuracy. Magnetic signals are accessible for users with smartphone
magnetometers but easy to be affected by magnetic field anomalies (Li et al., 2012).
Vision analysis can also be used for indoor positioning, but its performance is
questionable, and the initial investments on cameras can be quite high (Kawaji et al.,
2010). There are lots of radio frequency based indoor positioning and navigation, such
as Infrared, light, GSM, WiFi, BLE, RFID, and UWB (Deng et al., 2019). These
techniques are capable of reaching centimeter accuracy, but initial investments on RF
devices, system development and maintenance can be high (Xu et al., 2019). Pedestrian
Dead Reckoning (PDR) is also applied in smartphones and smartwatches; however, it
suffers from very low-performance accuracy (Kang & Han, 2014). Current AR
methods, as discussed above, also suffer from low performance problems and requires
high initial investments of databases and high-profile AR smartphones. The WaNAR
we proposed can function very accurately and easily with any low-profile AR
smartphones.
Table 1. Comparison between different indoor positioning techniques
Technique classes Performance
Price
Examples
Sonic
Magnetic
Vision

★★★
★★★
★★

★
★★
★★★★★

Radio Frequency
(RF)
Pedestrian Dead
Reckoning (PDR)
Augmented reality
(AR)
Our WaNAR

★★★★

★★★

★

★★★★★

★★

★★★★★

★★★★★

★★★★

Marker, floor pattern, imageto-location reasoning
Infrared, light, WiFi, BLE,
GSM, UWB, etc.
Step counter + motion sensors
iPhone 11, Google Tango /
Pixel, Huawei Mate 30P
Ditto.

Apart from indoor positioning, another heated research topic is the integration
of indoor and outdoor positioning information and systems. With people moving
seamlessly between buildings and surrounding areas, positioning and navigation tools
should support seamlessly integrated indoor-outdoor scenarios instead of merely
outdoor or indoor guidance (Vanclooster et al., 2016). A plethora of research works on
the integration of GPS, the preferred outdoor positioning and navigation technology,
4

164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210

and other indoor positioning technologies. For example, Cheng et al. (2014) proposed
a seamless outdoor/indoor pedestrian navigation system where GPS serves for stable
and continuous outdoor navigation and WiFi as a reliable and stable indoor navigation
technique. Other emerging technologies such as BIM (building information modeling),
stereo-vision are also applied for indoor-outdoor integrated positioning. Stereo-vision
based navigation system for unknown indoor and outdoor environments was designed
and introduced for both flying robots and pedestrians. Again, these solutions are
dependent on specific equipment and constrained by initial investment and precision
problems.
Related 3D Model-based Error Calibration Methods
The 3D model-based error calibration method used in this paper is not a newlycreated one. Eisert (2002) used synthesis analysis to calibrate extrinsic and intrinsic
camera parameters based on a 3D computer graphics model. Sochor et al. (2017)
calibrated traffic surveillance camera by 3D model bounding box alignment for
accurate vehicle speed measurement. The model-based error calibration method is also
widely applied in robot calibration, in which it can be divided into four parts, i.e., robot
kinematic definition, robot position measurement, robot kinematic model
identification, and compensation of position errors (Bai, 2007). When adopted in other
scenarios, the steps are more or less the same. The compensation is based on the
deviation between the real-time value and the model. If it is assured that the kinematic
features are based on the predefined model, then the deviation can be eliminated within
a preset tolerance scope.
THE PROPOSED METHOD
The research methods of this study is shown in Figure 1. There are two inputs
prior to the study. One is literature review which helps conclude and compare different
error compensation methods and AR positioning methods. The other input is a 3D
walkability network. Under the ground-truth assumption that people only walk along
horizontal lines such as roads, paths, corridors and slopes including stairs of qualified
walking conditions and go through doors that can be opened, the 3D walkability
network can be drawn based on surveying. Based on the methods reviewed and the
linear features in the walkability network, a WaNAR error compensation algorithm is
designed for consistent mitigation of drifting errors in AR positioning. Accordingly,
coding work builds the prototype application in an Android APP. After debugging and
testing, the APP is ready for pilot studies. By gathering and revising with the validation
data and feedbacks from pilot studies, the WaNAR application can be upgraded and
deployed.
The WaNAR error compensation algorithm, as shown as pseudo codes in
Figure 2, is the core part of the propose method. It receives two inputs at any time: one
is the translations on x, y, z axes (translationAR), and the other is a set of 3D lines of the
walkability network (network3D). There is a drifting vector drifting_vec recording the
accumulated errors from comparing the AR motions to the walkability network. Before
the compensation, the vector of drifting error is assumed as zero. First, the nearest line
of walkable way is selected and the perpendicular foot from the AR-sensed translation
to the line is computed. The distance from the foot to the AR-sensed translations is the
estimated drifting error. However, if the translation is too close to the end of the line
5

211
212
213
214

(i.e., within the constant value of TURNING_BUFFER) or too far away from the
nearest line (i.e., meeting the constant value of OFF_TRACK), the compensation will
be dropped due to arrival at a possible turning point or unfollowing the guided
pedestrian network.

215

216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232

Figure 1. Research methods of this study
For realizing the WaNAR algorithm in a beacon-free AR positioning method
and demonstrating the replicability, we used a Google Tango (model: Lenovo PB2690Y) smartphone with Android 4.4. In the prototype Android APP, the 3D translation
(positions on x, y, z axes) and the rotation (on x, y, z axes) of the AR phone pose are
consistently read from the Tango API TangoSupport.getPoseAtTime(). The vector of
translation is an input to the WaNAR algorithm (translationAR in Figure 2) subjecting
to correction, while the rotation of camera pose is used for the graphic display of the
APP via updateRenderCameraPose() to synchronize the AR experience. It should be
noted that although Google Tango phone is very powerful in “area learning” through
infrared distance sensor, the “area learning” options are disabled to simulate a low-end
AR phone. That is, only the position and rotation vectors data collected and integrated
from inertial motion sensors (including the accelerator and gyroscope) are used in the
realization of the method.
procedure WaNAR_error_compensation:
input translationAR, network3D, drifting_vec
way := nearest_path (network3D, translationAR)
if distance (translationAR, tails_of (way)) > TURNING_BUFFER then
foot := perpendicular_foot (translationAR, way)
drifting_vec := drifting_vec + (translationAR – foot)
if || drifting_vec || ≤ OFF_TRACK then
6

233
234
235
236
237
238
239
240
241
242

translationAR := translationAR – drifting_vec
end if
end if
return translationAR
end procedure
Figure 2. Pseudo codes of the WaNAR error compensation algorithm
BEACON-FREE AR POSITIONING WITH A PILOT STUDY
Experimental Setup
A pilot study, as shown in Figure 3, was conducted at an area around the Main
Campus, the University of Hong Kong (HKU). The study area is a hilly area with
compact campus buildings and complex vertical and horizontal connections between
the buildings. Therefore, a 3D walkability network for guidance and navigation is much
desired. We employed a 3D walkability network developed in Sun et al. (2019). The
network covers the whole outdoor pedestrian paths and some indoor areas.

243

244
245
246
247
248
249
250
251
252
253
254
255

(a) The study area around HKU Main (b) The 3D walkability network, warmer
Campus
color indicates higher altitude
Figure 3. 3D walkability network in the study area
Android APP Demonstration
Figure 4 shows the user interface of the prototype APP which was called “HKU
Walk.” The development environment was Android Studio (version 3.1). The 3D
walkability network-based error calibration is made to compensate for the error of AR
positioning. The results of error compensation were showed regularly as messages, as
shown in Figure 4.b. And the APP can sense slopes and stairs (0.05 ≤ Δz / ||Δlocation||
≤ 0.7) and elevators (Δz / ||Δlocation|| > 0.7), as shown in Figure 4.c. After some trials
and errors, the two constant values were also set for a fluent navigation experience,
where TURNING_BUFFER = 0.5m and OFF_TRACK = 0.5m. By using such passive
methods, no RF beacon signals are required.

256

7

257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290

(a) Android APP
(b) Error compensation
(c) On slopes
Figure 4. HKU Walk demonstration snapshots
User can rotate and pan the screen to attach the green ball, which represents
one’s location, to the 3D walkability network as the start location; so it is the same with
the re-position on demand. The accumulated vector drifting_vec will be reset to [0, 0,
0] at the mean time. After starting from a specific point, the APP can guide the
navigation by keep compensating the error.
Results and validation
We conducted an indoor-outdoor walking test for about 10 minutes. A video
about the test results was recorded during the test, of which the full version is available
at: https://www.youtube.com/watch?v=jFy_MFYSgBY. In the test, the user walked
from the Knowles Building, G/F to the entrance of Chong Yuet Ming Cultural Centre,
back to Knowles Building, G/F, then walking upstairs to 1/F and 2/F. The “Flight
mode” was on through the test, so the conventional radiofrequency signals including
GSM, WiFi, Bluetooth, GPS, and RFID, were disabled.
The results showed that the indoor/outdoor positioning by AR and WaNAR was
accurate all through the test path in the 10-minute period. With the error compensation
algorithm, the proposed WaNAR method can mitigate the sensor drifting and maintain
AR sensing accuracy at a subcentimeter level, which is much better than most of the
other positioning techniques. Therefore, the pilot study validated the technological
feasibility of the WaNAR method. The application can even notice pedestrians the
slopes and stairs and elevators.
We also had a few findings regarding the implementation. Firstly, during the
test, users are asked to walk along the center line of the roads which will lead to smaller
error. Secondly, the better the drifting error is compensated, the more precise the APP
works. Thirdly, most walkable roads indoor is not very wide. Therefore, the 0.5 meter
is an acceptable value for the parameter OFF_TRACK in the pilot case.
Limitations and Discussion
For the scalability, any phone with simple motion sensors and basic AR
function can run and utilize the application very well even under the flight mode. And
the error compensation requires no external beacon signals, which means the WaNAR
solution has no on-cost in maintenance. The only prerequisite is a 3D indoor-outdoor
8

291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335

map which can be released by the building’s owner and promoted to users via WiFi
access points or so.
However, due to the linear 3D pedestrian network, the WaNAR method
performs the best in linear indoor-outdoor spaces, and cannot well cover large open
areas such as a podium. Another drawback is the presented method may lead to a small
error if the user is not walking on the center line. In addition, the functionality of the
demonstration APP requires user’s manual re-positioning.
There are future development directions to resolve the above problems. First,
the linear 3D pedestrian network can be extended with areas. Secondly, the infrared
depth sensor can offer the depth image (as 3D point cloud) of the path, which can be
processed automatically for the symmetry, center line, and iconic 3D objects (Xue et
al., 2019a; 2019b) for additional error compensations, e.g., perpendicular to the guide
walkability line and iconic environment object-based re-positioning. Thirdly, for
automatic repositioning and the new comers who is not acquainted with the
environment, we expect to integrate passive beacons, such as QR code or near-field
communication (NFC) tags, for the start point and re-positioning in complex indooroutdoor navigation.
CONCLUSION AND FUTURE WORK
This paper developed a straightforward, beacon-free error compensation
method for precise smartphone AR positioning based on previous works of AR
positioning, indoor positioning, indoor-outdoor positioning integration, and 3D modelbased error calibration. It was proven to be more effective, accurate, and cheaper than
other positioning methods in the pilot study in both indoor and outdoor positioning as
long as the 3D walkability network covers. Different from other AR positioning
methods which require high-profile AR phones with more sensors and big databases to
store the images or models of the environment, our method just needs low-end AR
phones with basic motion sensors and several manual hours to draw a 3D walkability
network for a building.
However, it also has some obvious limitations. First of all, it may not work in
small and complex environments such as an equipment room because walkability
network in such an environment is hard to draw. Besides, for very wide roads and areas,
several parallel lines should be drawn to ensure diversity of route choice.
Even though, it has great potential in areas including seamless indoor-outdoor
navigation, facility management, and any other location-based services. When
integrating with map services, it can contribute to the precise and seamless indooroutdoor integration for positioning and navigation. It enables facility managers to
provide better indoor navigation services for users which will furtherly enhance the
convenience for users and efficiency of businesses. It also has excellent potential in
location-based services such as shop searching in shopping malls, UAV navigation in
unmanned warehouses, and office navigation in hospitals.
Future research directions include walkability areas in 3D map, infrared depth
sensor and on-the-fly 3D path recognition, and integration of inexpensive beacons like
NFC tags. Further development work includes rich navigation functions, map services
integration, and automated walkability network generation.

336

9

337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382

REFERENCES
Bai, Y. (2007). “On the comparison of model-based and modeless robotic calibration
based on a fuzzy interpolation method.” The International Journal of Advanced
Manufacturing Technology, 31(11-12), 1243-1250.
Bellini, P., Nesi, P., Simoncini, M., and Tibo, A. (2014). “Maintenance and emergency
management with an integrated indoor/outdoor navigation support.” Journal of
Visual Languages & Computing, 25(6), 637-649.
Cheng, J., Yang, L., Li, Y., and Zhang, W. (2014). “Seamless outdoor/indoor
navigation with WIFI/GPS aided low cost Inertial Navigation
System.” Physical Communication, 13, 31-43.
Deng, Z., Fu, X., Cheng, Q., Shi, L., and Liu, W. (2019). CC-DTW: An Accurate
Indoor Fingerprinting Localization Using Calibrated Channel State Information
and Modified Dynamic Time Warping. Sensors, 19(9), 1984.
Eisert, P. (2002). “Model-Based Camera Calibration Using Analysis by Synthesis
Techniques.” In Vision, Modeling, and Visualization 2002, Erlangen, Germany,
307-314.
Kang, W., and Han, Y. (2014). “SmartPDR: Smartphone-based pedestrian dead
reckoning for indoor localization.” IEEE Sensors Journal, 15(5), 2906-2916.
Kawaji, H., Hatada, K., Yamasaki, T., and Aizawa, K. (2010). “Image-based indoor
positioning system: fast image matching using omnidirectional panoramic
images.” Proceedings of the 1st ACM International Workshop on Multimodal
Pervasive Video Analysis, Firenze, Italy, pp. 1-4.
Koch, C., Neges, M., König, M., and Abramovici, M. (2014). “Natural markers for
augmented
reality-based
indoor
navigation
and
facility
maintenance.” Automation in Construction, 48, 18-30.
Li, B., Gallagher, T., Dempster, A. G., and Rizos, C. (2012). “How feasible is the use
of magnetic field alone for indoor positioning?.” 2012 International
Conference on Indoor Positioning and Indoor Navigation (IPIN), 1-9.
Paucher, R., and Turk, M. (2010). “Location-based augmented reality on mobile
phones.” 2010 IEEE Computer Society Conference on Computer Vision and
Pattern Recognition-Workshops. IEEE. 9-16
Sochor, J., Juránek, R., and Herout, A. (2017). “Traffic surveillance camera calibration
by 3d model bounding box alignment for accurate vehicle speed
measurement.” Computer Vision and Image Understanding, 161, 87-98.
Sun, G., Webster, C., and Zhang, X. (2019). “Connecting the city: A three-dimensional
pedestrian network of Hong Kong.” Environment and Planning B: Urban
Analytics and City Science, in press.
Vanclooster, A., Van de Weghe, N., and De Maeyer, P. (2016). “Integrating indoor and
outdoor spaces for pedestrian navigation guidance: A review.” Transactions in
GIS, 20(4), 491-525.
Xu, J., Lu, W., and Wang, J. (2019). “The gateway to integrating user behavior data in
‘cognitive facility management’.” 4th International Conference on Civil and
Building Engineering Informatics (ICCBEI). Sendai, Japan.
Xue, F., Chiaradia, A. J. F., Webster, C. J., Liu, D., Xu, J., and Lu, W. (2018).
“Personalized walkability assessment for pedestrian paths: An as-built BIM
10

383
384
385
386
387
388
389
390
391
392
393

approach using ubiquitous augmented reality (AR) smartphone and deep
transfer learning.” In Proceedings of the 23rd International Symposium on the
Advancement of Construction Management and Real Estate, in press, Springer.
Xue, F., Lu, W., Chen, K., and Zetkulic, A. (2019a). “From semantic segmentation to
semantic registration: Derivative-Free Optimization–based approach for
automatic generation of semantically rich as-built Building Information Models
from 3D point clouds.” Journal of Computing in Civil Engineering, 33(4),
04019024.
Xue, F., Lu, W., Webster, C. J., and Chen, K. (2019b). “A derivative-free optimizationbased approach for detecting architectural symmetries from 3D point clouds.”
ISPRS Journal of Photogrammetry and Remote Sensing, 148, 32-40.

11