�ݺ�ߣ

Assistive technologies: experiences from
AAL for the blind and visually impaired
within the ALICE project
Andrei BURSUC, Prof. Titus ZAHARIA
Institut Mines-Télécom; Télécom SudParis
firstname.lastname@telecom-sudparis.eu
Invited talk by DemaCare FP7 Project

• Context and objectives
• The ALICE project and AAL
• State-of-the-art
• User requirements
• System prototype
• Obstacle detection
• Navigation assistant
• Human-Machine interface
• Conclusion and perspectives
2
Outline
Experiences from the ALICE project

• VI persons face many problems every day:
– overall contextual understanding of space semantics
– interaction with surrounding objects
– planning, orientation, communication, navigation
• 285M registered visually impaired people: 39M blind, 246M
with low vision (WHO report)
• The degree of visual impairment is increasing with an
ageing population
3
Context and objectives
Nowadays

• Provide navigational assistive device for elderly blind
with cognitive capabilities:
– Positioning
– Obstacle detection/alerting
– Landmark/object recognition
• Offer VI users a cognitive description based on a fusion of
of perceptions gathered from multiple sensors
• Personal benefits:
– Enable independency of blind and partially sighted people
– Save stress and time of the end-users
– Improve the individual self-esteem
4
Context and objectives
Objectives

• 7 partners (academics, SMEs, VI persons associations)
• 4 European countries (ES, FR, SI, UK)
• Duration: June 2012 – November 2014
• Final product: device consisting of smartphone with
additional sensors, wirelessly connected with local processing
unit
The project
ZVEZA
SLEPIH
5Experiences from the ALICE project

• Ambient Assisted Living - funding activity that aims:
– to create better condition of life for the older adults
– to strengthen the industrial opportunities in Europe through the
use of ICT
• Funding across-national projects involving SMEs, research
bodies and user’s organizations
• Time-to-market perspective of max 2-3 years after the end
of the project
• Project total budget: 1-7 M€ (funding 3 M€ at most)
AAL Joint Programme

Experiences from the ALICE project 7
What’s possible?
State of the art

How VI orient themselves?
• With the help of the guide (other person)
• Using a white cane, guide dog
• Using electronic devices, GPS
• By listening familiar sounds
• By looking for something familiar (edge of pavements,
curves, crossroads, very large inscriptions)
• Underfoot textures, different surfaces
• Sun, wind directions, smell
• Road signs
8
State of the Art

• Current techniques are still not very advanced
9
State of the Art

• Cane and dogs are still kings!
10
State of the Art

How VI (could) orient themselves?
• Navigation systems:
– GPS + computer vision (clear path, landmark recognition)
• Object recognition systems:
– Grocery shopping assistant
– RFID tags on objects
– OCR (Optical Character Recognition)
– Detectors: crosswalk , walk lights, staircase, street signs, pedestrians
• Obstacle avoidance systems:
– Integrating depth information
– Step and curb detection
11
State of the Art

• Conclusions:
– Few systems work in real time
– Many approaches require the use of heavy equipment
– Some systems need tags
– The research field should get a new boost with the advent of the
Google Glass
How VI (could) orient themselves?
12
State of the Art
[Lee, 2012]
[Marduchi, 2012] [Pradeep, 2010]

• Limited computational resources: light and low powerful
wearable devices
• Real-time responsiveness
• Reliability and no false positives
• Adequate and non-overwhelming communication with the
user (alerts, indications)
13
State of the Art
Challenges

24 July 2013 14
Setting up the path
User feedback and requirements

• Participants’ profile:
– Age: 55-75
– Countries: Slovenia, UK
– Degree of visual impairness: blind and partially sighted
– Total: 40 participants (20 from each country)
Questionnaire for end-users
15
User requirements

Questionnaire conclusions
• 50 % of participants are using only familiar routes
• Most participants need someone to guide them to certain
places.
• Some of them need the guide every time – often they
depend on the time and will of others.
• It is important to know where they are positioned, how far
the destination is and the vicinity of the route
16
User requirements

Questionnaire conclusions - Device
• Not very much confidence placed in the electronic
navigation system (only after several successful tests)
• Necessity of training and information about electronic
devices.
• Half of users use speech synthesis
• Willingness to use headphones, but hearing shouldn‘t
be obstructed.
• “Turn by turn” functionality should not give too much
info
17
User requirements

Questionnaire conclusions - Indoor
• 85 % of respondents have difficulties with orientation
through indoor public institutions.
• Difficulties the users are facing in indoor environments:
– the size of the room
– glittering surfaces
– room darkness
– no orientating points to navigate with white cane
– difficulties to recognize the landmarks
– background music.
18
User requirements

Questionnaire conclusions - Obstacles
• Obstacles that users want to be warned about:
– pillars
– curves
– overhanging branches
– edge of pavements
– street furniture
– steps
– down slopes
– ramps
– holes
– bumps
19
User requirements

User expectations
• The device should be accurate:
– Exact info about the obstacles
– Find safe corridors for walking
– Warn the user when is safe to cross the road, the green light is on,
if traffic is coming (especially bikes, electric cars)
• The device should be small, portable, phone sized.
20
User requirements

User expectations
• Other features:
– Give the distance to the building
– Find the right bus stop, post box.
– Text-to-speech for: letters, journey‘s instructions , street
inscriptions, shop names
– Tell the weather, temperature, local taxi availability.
– Recognize faces and the person‘s name.
21
User requirements

First tests and experiments
System prototype

Sensor evaluation
• Evaluation of multiple sensors: camera (ToF, stereo, web),
compass, gyroscope, ultra-sonic ranger, GPS, pedometer)
• Samsung Galaxy S3 used as baseline
23
System prototype
Image
Comunication
Sound commands
Tactile comunication
Orientation
Positioning
Light sensor
Inclination

Sensor evaluation
• Sensors have different sampling speeds
24
System prototype

Sensor evaluation - Conclusions
• All sensors in Samsung S3 are superior than the external
ones tested (except GPS).
• External GPS has better reception due to antena – but in
areas with strong multipath effect, the advantage is reduced
• Accuracy of GPS: 10 – 40 meters in urban areas
• Ultrasonic ranger would be useful for obstacles in front of
the user
25
System prototype

Possible camera positions
26
System prototype

27
System prototype

• Setting used for video recording
28
System prototype

Headphones
• Bone conduction headphones:
– Effective even in very loud enviroment (city traffic)
– Does not obscure sounds from enviroment
– Very High frequencies not as good as in normal headphones
29
System prototype

30
Platform configuration
System prototype

Conclusion
24 July 2013 31
Conclusion and perspectives
Parsing the visual domain
Obstacle detector

32
Input video stream
Method overview
Obstacle detection

33
Input video stream
Interest points extraction
Grid of points regularly spread in a frame
Method overview
Obstacle detection

34
Input video stream
Grid of points regularly spread in a frame
Interests points matching and
tracking
Multiscale Lucas-Kanade algorithm
Method overview
Obstacle detection

35
Input video stream
tracking
Multiscale Lucas-Kanade algorithm
Background / Camera motion
estimation
Global geometric transform – RANSAC
algorithm
Method overview
Obstacle detection

36
Input video stream
tracking
estimation
Global geometric transform – RANSAC
algorithm
Static / Dynamic obstacle
motion estimation
Agglomerative clustering based on
proximity computation
Method overview
Obstacle detection

37
Input video stream
tracking
estimation
motion estimation
Agglomerative clustering based on
proximity computation
Interest points refinement
K-NN algorithm and small clusters removal
Method overview
Obstacle detection

38
Input video stream
tracking
estimation
motion estimation
Obstacles classification
K-NN algorithm and small clusters removal
Method overview
Obstacle detection

Input video stream
tracking
estimation
motion estimation
Obstacles classification
Obstacle classification based on position
and direction relative to the video camera
Experimental results
Method overview
Obstacle detection

40
Experimental results
Obstacle detection

41
The algorithms were run on an Intel Xeon Machine 3.6 GHz, RAM 16 GB RAM and on a NVIDIA Quadro 4000 video board (256 cores CUDA, 256 bits of external memory
interface and 9945 MB graphical memory), under a Windows 7 platform (desktop).
Preprocessing steps
Time - without GPU
(msec)
Time - with GPU (msec)
Interest points detection (image grid) 0.05 – 0.5
Interests points matching and tracking
(unidirectional Lucas – Kanade optical flow)
22 - 23 10 - 11
Background / camera motion estimation (unidirectional
homographic motion model (RANSAC)
6.5 - 8.0
Object / obstacle motion estimation
(agglomerative clustering)
0.05 – 0.15
Interest points refinement (K-NN algorithm) 0.05 – 0.1
Obstacle classification
(approaching / departing and urgent / normal)
0.05 - 0.1
Saving results (video) 1.5 – 2.05
TOTAL TIME / FRAME (average) 31 ms 20 ms
Computational time
Obstacle detection

Objectives
Human-Machine interface
Taking the path
Navigation assistant

Accessible Maps
• Crow-sourced application for maps annotation
• Routes are entered, edited and shared with Google Maps
• OpenStreetMaps used as repository and online access to
information about points of interest.
43

Accessible Maps
• Waypoints annotations:
– WHAT: presence of crosswalk, traffic lights in an intersection, type
of intersection, walk buttons, Stop signs, median strips.
– WHERE: information in form of absolute geographic form (Lat, Long)
44

Assistance
• Crossing ahead:
• Turn left and then cross:
45

Assistance
• Demo:
46

Objectives
Making the connection

Objectives
• Create a communication/presentation system:
– Highly adapted to user needs
– Enable the VI to perceive and interact with the surrounding
environment
• Instructions for navigation will have to acknowledge that
user perception is similar to moving blindfolded in a maze:
– Verbalization: for description of surrounding objects
– Enactive methods: for presenting orientation, distance, motion
and position of moving objects

Methods
• 2 separate groups of users according to:
– Level of visual impairment
– Other criteria (age, education, etc.)
• Interface modalities:
– Audio semantics using sound, music and synthesized voice
– Text-to-speech synthesis using headphones
• Input modalities: screen, tapping, gestures, voice
• Output modalities: audio, haptic, tactile

Enactive methods
• Communication with the user: what, when, how
– Not just how to transfer information between the system and the
user, but what information and when.
– The timely delivery of the right information avoids information
overload.
– Translate the sensory impressions about the surroundings into
tactile or sound information ( faster and easier to comprehend
than verbalization).

User warning
• Directional warnings: earcons
• Positional warning:
– alerting a user must give user enough time to prepare (2-3 sec for
a voice message)
– acoustic signal (sequence of beeps) with varying frequencies
– vibrations in the bone conduction headphones
51

Menu
• Hierarchical menu
52

Georgie prototype
• Sample user-interface
53

24 July 2013 54
Next steps
Conclusion and Perspectives

Conclusion
• Encouraging first achievements within the ALICE project
• Human-Machine interfacing is a difficult challenge
• User feedback is essential
• Still plenty of things left to improve
55

Perspectives
• Learning and recognizing user-defined landmarks and
objects of interest
• Obstacle classification according to degree of risk to the
user and generation of adequate alerts
• Improve navigation and recognition at key points of trip
(start and finish)
• Navigation and obstacle recognition modules integrated
into a single application
56

ALICE benefits in day-to-day life?
• Jean:
– is partially sighted
– works at UBPS
– travels the same route to his office every day
57

• Jean:
– knows the route
– with his white cane he manages to travel safely from the bus stop
to the building.
58

• Paul:
– is blind
– goes at the UBPS once a week
– uses different route (he doesn’t feel safe enough)
59

• Paul:
– Paul’s route
60

• Paul and some other blind people usually need to take
longer routes (more then 400m)
61
Paul’s routeJean’s route

How can ALICE bring benefits?
24 July 2013 62
Find out more at
www.alice-project.euThank you!

• �ݺ�ߣ 2: http://www.flickr.com/photos/gullevek/3240421172/
• �ݺ�ߣ 7: http://www.flickr.com/photos/pointshoot/3590816656/
• �ݺ�ߣ 10: http://blog.grdodge.org/wp-content/uploads/2011/08/Morris-and-Buddy-1.jpg
http://www.iowablindhistory.org/sites/default/files/image/History%20Site%20Images%20and%20Audio%20/Pic%20o
f%20Jernigan.jpg
http://www.flickr.com/photos/library_of_congress/8190452507/
http://www.globalride-sf.org/images/0608/images/2_PedInfra_TactileWarnings.jpg
http://images.ookaboo.com/photo/m/Geleidehond_testparcours_m.jpg
http://www.robertschroeder.com/wordpress/wp-content/uploads/2011/01/GuidedWalkSchroeder.jpg
http://abramsonscorner.files.wordpress.com/2011/06/img_9072-13-of-54-version-2-1-of-1.jpg
• �ݺ�ߣ 14: http://farm4.staticflickr.com/3459/3188288778_3d44b943b4_b.jpg
• �ݺ�ߣ 15: http://blockingfortheblind.org/wp-content/uploads/2013/02/peoplewithcanes.jpg
• �ݺ�ߣ 20: http://i.huffpost.com/gen/819993/thumbs/r-BLIND-MAN-TASERED-large570.jpg
• �ݺ�ߣ 31: http://www.flickr.com/photos/swiiffer/4593608484/
• �ݺ�ߣ 42:
http://upload.wikimedia.org/wikipedia/commons/thumb/a/af/Blind_Leading_the_Blind_by_Lee_Mclaughlin.jpg/1024px-
Blind_Leading_the_Blind_by_Lee_Mclaughlin.jpg
• �ݺ�ߣ 47: http://i.imgur.com/f3fqnEY.jpg
• �ݺ�ߣ 54: http://www.flickr.com/photos/84681882@N00/5467879589
• �ݺ�ߣ 62: http://www.austindowntownlions.org/Resources/Pictures/Gucci%20looking%20forward%20and%20canes.jpg
63
Photo credits

�ݺ�ߣ

Assistive technologies: experiences from AAL for the blind and visually impaired within the ALICE project

More Related Content

Assistive technologies: experiences from AAL for the blind and visually impaired within the ALICE project