Research on Landmark Cognition for Pedestrian
Navigation Services
1 School of Geography, Nanjing Normal University, Nanjing 210023, China
2 Research Group Cartography, Vienna University of Technology, 1040 Vienna, Austria
Litao Zhu1,2, Jie Shen1
CONTENTS
Study 1
Zhu, L., Shen, J.*, Zhou, J., Stachoň, Z., Hong, S., & Wang, X. (2021).
Personalized landmark adaptive visualization method for pedestrian
navigation maps: Considering user familiarity. Transactions in GIS
Outdoor landmarkPART
01
1. Introduction: Landmarks
Any sufficiently prominent object can be considered a landmark. The definition of indoor
landmarks and outdoor landmarks is unified.
Spatial cognition research indicated that landmarks, as key elements of wayfinding, can reduce
navigation time (Golledge, 2003), decrease error rates (Goodman et al., 2004), improve route
learning (Tlauka & Wilson, 1994), reduce user cognitive load and increase confidence in
navigation decisions (Millonig & Schechtner, 2007).
Indoor Landmarks (Ohm, 2015).
1 Introduction: Landmark salience
Landmark salience is the nature of the landmark itself, the strong contrast with the
surrounding environment resulting in the attraction to people (Presson & Montello, 1988).
Landmark salience is divided into Visual, Semantic, and Structural (Raubal & Winter, 2002)
StructuralVisual Semantic
Landmark salience
2 Motivation
Empirical studies indicated people with different levels of familiarity have different
preferences for landmark selection and representation. Spatial familiarity is an important
variable related to personalized navigation but is often ignored.
Real World Space
Cognitive Map Space
More familiar with cube：
The cuboid is next to the cylinder.
More familiar with cylinder：
The cylinder is next to the cube.
A
B
The motivation is to investigate landmark selection and visualization preferences of people
and apply the results to the landmark-based pedestrian navigation system.
3 Research questions
1. How to select personalized landmarks for target users.
2. How to realize the adaptive visualization of landmarks in pedestrian
navigation maps.
In this study, we propose a personalized landmark adaptive visualization
method considering user familiarity. We focus on two research questions:
4 Methodology
Step 1: The influence of spatial familiarity on landmark salience and symbols based on cognitive
experiments is explored.
Step 2: Association rules between landmark salience and symbols are mined.
Step 3: A personalized landmark adaptive visualization method based on these rules is proposed.
The framework of methodology
4.1 Step1: Cognitive experiments
The aim is to explore the influence of spatial familiarity on landmark salience and symbols.
Study area.
(2) Materials and Procedure
• Self-assessment task: Santa Barbara Sense-ofDirection
Scale (SBSOD) and two additional
questions (about visits and mapping ability)
• Sketch-mapping task
• Questionnaire for landmark symbols
evaluation
Landmark symbols
(1) Participants
Familiar group (27) vs. Unfamiliar group (24)
The results of landmark extractionExamples of sketches drawn by the familiar group and the
unfamiliar group
Result 1: Analysis of sketch mapping task.
The t-test results indicated a significant difference (t = 3.70, p = 0.001 < 0.05) between the familiar group (M
= 16.93, SD = 5.14) and the unfamiliar group (M = 11.58, SD = 5.15). It showed that the level of
environmental detail provided by familiar and unfamiliar people varied substantially.
4.1 Step1: Cognitive experiments
Result 2: Mathematical model of landmark salience.
: : = 5:2:3vis sem strW W W
: : = 3:5:2vis sem strW W W
L vis vis sem sem str strS W S W S W S+ +=
For the unfamiliar individual,
For the familiar individual,
Result 3: the preference for landmark symbols.
The results showed significantly different proportions of landmark symbol selection between the familiar group
and unfamiliar group ( ). Table 5 presents the descriptive statistics of the preference
rates of the familiar group and the unfamiliar group for different landmarks..
4.1 Step1: Cognitive experiments
4.2 Step2&3: Personalized landmark adaptive visualization method
(1) Data collection, including user data, landmark data, and the range of the map.
(2) Features extraction, including user features and landmark features.
(3) Rule execution, it aims to identify the personalized mode of user interactions with pedestrian
navigation maps based on user familiarity using association rule mining.
5 Verification experiments
(1) Prototype system
(2) User experiments
• Participants: Familiar group (14) vs. Unfamiliar group (14)
• Materials: the prototype vs. Baidu Map for Mobile (BMM).
• Procedure: Pedestrian navigation task; and System Usability Scale Questionnaire
Prototype system The prototype vs. Baidu Map
5 Verification experiments
(3) Results
• Time efficiency
• The number of map views
• Analysis of the System Usability Scale (SUS)
The average completion time The average number of map views
6 Discussion
Using landmarks in maps helps users identify their location (Hile et al., 2008).
Previous studies proposed the user-centered visualization method for outdoor landmarks
(Elias & Paelke, 2008).
However, few studies have explored the visualization of indoor landmark symbols.
Landmark symbols (Elias & Paelke, 2008).
Indoor landmarkPART
02
Study 2
Zhu, L., Švedová, H., Shen, J.*, Stachoň, Z., Shi, J., Snopková, D., & Li, X.
(2019). An instance-based scoring system for indoor landmark salience
evaluation, Geografie, 2019/2.
1 Research questions
1. How to evaluate the landmark salience in the indoor environment.
2. How to verify the usability of indoor landmark salience evaluation results.
This study proposed an instance-based indoor landmark salience
evaluation method to address the lack of indoor landmark salience evaluation
methods. We focus on two research questions:
2 Methodology
• Propose indoor landmark indicators and scoring system
• Calculate landmark salience weight using AHP
3 Experiment and Result
To verify the usability of the proposed method, we applied it to a shopping mall (Nanjing,
China) using questionnaire and a headquarter (Brno, Czech Republic) using eye-tracking.
(Nanjing, China) (Brno, Czech Republic)
4 Discussion
In this paper, the landmark indicators are scored by users and the weights are
scored by experts. The proposed method is tedious and complicated. The process need
to be repeated for each scenario.
Typical human and computer vision pipelines (Wäldchen & Mäder, 2018).
In future work, we will
consider using machine
learning methods to
automatically identify
indoor landmarks.
Indoor landmarkPART
02
Study 3
Zhu, L., Shen, J., Gartner, G., & Hou, Y. (2021). Personalized Landmark
Sequence Recommendation Method using LSTM-based Network for
Navigating in Large Hospitals. Abstracts of the ICA, vol. 3.
1 Introduction: The market for hospital navigation is considerable
1. A large number of Chinese hospitals
China has a total of 35,394 hospitals.
2. A large number of total hospital visits
In 2020, 3.32 billion visits
In 2019, 3.84 billion visits
In 2018, 3.58 billion visits
In 2017, 3.44 billion visits
In 2016, 3.27 billion visits
3. In 2017, the National Health Commission of China requires that the informatization
construction of hospitals includes indoor navigation.
The hospital navigation market is at least €2.1 billion
1 Introduction: The hospital navigation
• Indoor space of the hospital is unique
in that most of the facilities are related
to the task of medical visits.
• The users (patients/visitors) who use
hospital navigation are unique. They
usually accomplish many tasks under
time constraints and discomfort.
• Hospital guidance information
contains a great deal of medical
terminology and knowledge.
In summary, the specificity of hospital navigation lies in its close connection with user
behavior, medical processes, and hospital space (departments).
1 Introduction: Hospital navigation APP
Hospital navigation apps are medical wayfinding tools for specific visitors or
patients who visit the hospital for any purpose.
To the best of our knowledge, few landmark-based pedestrian navigation systems
have been developed for hospitals.
Gozio Health
https://www.goziohealth.com/
MediNav
https://medinav.eu/home/
JoySuch (China)
https://www.joysuch.com/
IPSMAP (China)
https://www.ipsmap.com/
1 Introduction: Behavioral analysis of medical visits
Registered
Check in treatment
department
Pick up
reports
Physical
examination
Payment
Payment
Pick up
medicine
Leave the
hospital
Further
consultation
Waiting
Medical
consultation
Further
consultation
Location Sequence, Time Sequence, Location Hierarchy, Location Distance,
and Medical Treatment Sequence
User preference which related to the disease and task.
Entrance
Entrance
L1 L2 L3 L4
L5L6L7L8
L13 L14L12L11L10
L9
L15
Semantic Trajectories (i.e., Landmark Sequence)
1 Introduction: Landmark recommendation method
• 93% of user behavior is predictable (Song et al., 2010). Hospital landmark sequence
recommendation is also closely related to user behavior.
• Existing studies on personalized landmark recommendations are mainly used for
outdoor travel recommendations, but few studies for navigation and wayfinding.
• The POI recommendation methods include traditional machine learning and deep
learning (e.g., RNN, LSTM, GRU).
We adopted RNN to model landmark sequences for recommendation due to their
superiority in capability of processing the sequential data.
Song, C., Qu, Z., Blumm, N., & Barabási, A. L. (2010). Limits of predictability in human mobility. Science, 327(5968), 1018-1021.
2 Research questions
Inspired by research on POI sequence recommendation methods,
we propose a landmark sequence recommendation method using
LSTM-based network for hospital navigation. We focus on two
research questions:
1. How to model the complex sequential users behavior in hospital navigation.
2. How to develop an indoor landmark sequence recommendation algorithm
for hospital navigation.
3 Methodology: The research framework
We propose an indoor landmark sequence recommendation method for
hospital navigation based on LSTM with an attention mechanism.
The research framework can be divided into three modules: input,
attention-based encoder–decoder LSTM model, and output.
4 Experiments: Indoor trajectories
• Let represent indoor trajectory consisting of n points data
and the information on its location (latitude and longitude coordinate) and
timestamp. Let represent semantic trajectory consisting of i
landmarks data and the information on its location and timestamp.
• A user behavior sequence is a list of three-tuples.
 1 2 1
, ,..., , nn
P P P P P−
=
 1 2 1
, ,..., , ii
L L L L L−
=
4 Experiments
The process of experiment
5 Discussion
• we proposed a novel hospital landmark sequence recommendation
framework;
• we incorporated an attention mechanism into the LSTM, which helps to
capture the correlation between different landmarks.
Contributions
Outlook
In the future, we will do further research work in the following aspects:
• Refining the experiment
• Applying the proposed model to more complex hospital scenarios to verify
the performance of navigation.
Indoor landmarkPART
02
Study 4
Zhu, L., Shen, J., & Gartner, G. (2021). Ontology-driven context-aware
recommendation method for indoor navigation in large hospitals, LBS
2021: Proceedings of the 16th International Conference on Location Based
Services (pp. 23–26).
1 Introduction: Analysis of user behavior
Registered
Check in treatment
department
Pick up
reports
Physical
examination
Payment
Payment
Pick up
medicine
Leave the
hospital
Further
consultation
Waiting
Medical
consultation
Further
consultation
There are some contextual information in the hospital navigation: individual,
location, time, department, facility, medical process, medical services, medical
knowledge, schedule, navigation services.
Q1: what is the context model for hospital navigation?
Context model for outdoor
navigation (Richter et al. 2010).
Context model for indoor
navigation (Afyouni et al. 2012).
Healthcare context model
(Kim et al. 2014).
The lack of the context model for hospital navigation.
Richter, K. F., Dara-Abrams, D., & Raubal, M. (2010, September). Navigating and learning with location based services: A user-centric design. In Proceedings of the 7th International Symposium on LBS and
Telecartography (pp. 261-276).
Afyouni, I., Ray, C., & Christophe, C. (2012). Spatial models for context-aware indoor navigation systems: A survey. Journal of Spatial Information Science, 1(4), 85-123.
Kim, J., & Chung, K. Y. (2014). Ontology-based healthcare context information model to implement ubiquitous environment. Multimedia Tools and Applications, 71(2), 873-888.
Q2: what services are recommended by systems?
Lack of the recommendation method that combines medical services
and navigation services.
？？？
2 Research questions
In this study, we propose an ontology-driven context-aware
recommendation method for hospital navigation that adapts to
dynamically changing needs, tasks, and processes for various users.
We focus on two research questions:
1. How to develop a context model for hospital navigation?
2. How to realize personalized service recommendations?
3 Methodology: Research Framework
We designed the framework
of an ontology-driven
context-aware recommender
system as shown in the
Figure.
4 Discussion
• Developing a context model for hospital navigation using ontology to
complete the complex medical processes and provide personalized
navigation services;
• Developing a personalized recommendation mechanism using SWRL
rules to infer contextual information.
Contributions
Outlook
In the future, we will do further research work in the following aspects:
• Developing a prototype system
• Applying the proposed method to hospital scenarios.
Future workPART
03
Future work
In the future research on landmark-based pedestrian navigation service should
involves:
• Design user cognitive experiments to evaluate indoor landmark selection and symbols
• Design a user-center indoor landmark visualization method
• Understand the needs of different types of users for indoor navigation
• Design a more user-friendly the interface for hospital navigation
• Dynamic recommendation of landmarks based on context-aware.
Thanks！
Any question？
Litao Zhu
School of Geography
Nanjing Normal University
181301028@njnu.edu.cn
shenjie@njnu.edu.cn
http://schools.njnu.edu.cn/geog/person/jie-shen
Funding:
• National Key R&D Program of China (2021YFE0112300);
• National Key R&D Program of China (2016YFE0131600);
• National Natural Science Foundation of China (NSFC) (No. 41871371);
• The State Scholarship Fund from the China Scholarship Council (CSC)
(No. 201906860035).