Augmented Reality Tactile Maps
for Pre-Journey Learning
Ryosuke Ichikari and Takeshi Kurata
National Institute of Advanced Industrial Science
and Technology (AIST), Japan
Walking Assists for the
visually-impaired people
Assisting technologies when they are walking
• White cane
• Guide dog
• Assisting technologies with computer
– Remote Sensing by RGB(-D) cameras or LRF
– Voice navigation system based on localization
technology such as GNSS
Pre-Journey learning and re-viewing before/after
walking
Tactile Map
• Maps with tactile presentation for the visually-impaired
people
• UV offset printing
:
Printing transparent plastic on the
color maps
• PIAF(Picture in a flash)
:
Using heat sensitive papers
which can swell black-printed area (lines/images)
Problems about tactile map
• Difficulties for constructing mental image
of the contents with haptic sense
– Needs experience and training
• Limitation of the amount of contents
• Production issues
– Cost: UV offset printing is expensive
– Design issue
Proposed system
:
Utilizing AR for tactile map
• Augmented Reality; AR
– Seamless integration of real
world and virtual world
– Visual augmentation is main
stream, but AR concept
can be applied for audio sense
• AR Tactile Map
(
Proposed System
)
:
– Provides audio-visual feedbacks in response to
interactive operations
– Enables dynamic and multimodal contents
Requirements for AR tactile map
• Defined requirements by hearing to the
visually-impaired people
– Augmentation for the tactile map by audio representation (for completely blind people)
– Enlarging/enhancing display for people with week sight – Providing POR(Point of reference) information
– Helping comprehension of the tactile map with interactive training
– Intuitive interaction with the system
– Simple and inexpensive configuration with off-the-shelf products – Hands free arrangement
Augmentation for the tactile map by
audio representation
• Adding audio representation makes the
information more
reachable
• The amount of the contents by the
dynamic audio
representation
is not limited by physical tactile
map
• Dynamically Providing information according to
Enlarging/enhancing display for people with
week sight
• The people with week sight might perceive map
if the contents enlarged/enhanced.
• Image modifications in responding to the users
potential modifications: optical zoom, edge
enhancement, contrast adjustment, color
adjustment
Providing POR(Point of reference) information
• Target contents of the AR tactile map:
POI
(
Point of Interest
)
and
POR
(Point of Reference)
• POR
:
Specific point location where it is easily recognized
for confirming routes such as characteristic shape and
material of ground (steps, stairs, sloop, door), sound/noise,
and scent/odor)
[Okuno et. al, CSUN2015] [Denoncin,CSUN2014] [Theberge, CSUN2015]• Integration with POR/POI data base
– Voice navigation system with POR/POI data base [Okuno et. al, CSUN2015]
– Utilizing as the communication tool between visually-impaired people and sighted people at “Mapping Party”
Intuitive interaction with the AR tactile map
• Interactivity is important for
learning
tactile maps
and
dynamic representation
of the contents
– Trial-and-error with interactive system is short cut for
learning tactile map
• Inputting interface by
actually touching
tactile
maps
– Most
intuitive
and the
simplest
– Perceiving and inputting at the same time
Requirements of system configuration
• Adopting off-the-shelf products for availability
perspective
– For widely promoting the system
– Consists of only simple and inexpensive products
• Enabling hands-free arrangement
– From the visually-impaired people opinions
– For touching as many as possible points of the tactile
map
System configuration
System configuration of the prototype system
• Tablet PC
⇒
For Visual-Audio feedbacks
• RGB-D camera
:
DepthSense DS325
(
ToF, Range
:
15
㎝
~100
㎝)
⇒For finger gesture recognition and robust extraction of hands area, measurement of the distance between the map and finger
• PC for recognition
• Arm for fixing tablet
• Tablet mount parts
Related Works
• Image to Sound Converter [Kuroda]
– Image is converted to sound by changing key and difference of L/R sound volume according to the pixel’s coordinate
– Transparent tactile sheet can be overlaid
– Display size and UI are limited by touch display
• Kiko-AR by cybernet systems
– Mobile AR system based on Metaio junaio with audio presentation
– Image recognition and text-to-sound
– Collaborating with a printing company to produce printed material with tactile presentation and AR augmentation. – Need to shoot with smartphone’s camera
Characteristics of AR Tactile Map
:
- Various sizes and types of tactile maps can be used
- Intuitive interface for the system are focused
Implementations
Server Parts
Recognition
Client side AR tactile app
Gesture command, Finger points, lon/lat
Dababase POR data POR/POI management parts POR/POI data Finger points tracking Gesture recognition RGB-D camera Tablet PC Camera Image Fixed on backside ID of Map Lat/Lon POR/POI data collection app. POR/POI, Lat/Lon data
Visual Feedback(AR,
Enlargement/Enhancement)
Accepting finger interaction
POR data
Recognition of Tactile Map Tracking Position Of the Tactile Map
Recognition of gesture
• Recognizing “Tapping”,
“Double Tapping”,
“Searching” gestures as
the interaction commands
• Algorithm for detecting “Tapping”
– Extracting hand area by RGB-D camera with background differencing of depth image and distance threshold
– Detecting state transition within time window (①→②→ ① )
Distance between map and
finger ②
Variety of finger gestures
• “Tapping”
– For getting user’s focused points in the map
at the intended timing
– Equivalent to “Clicking” of mouse
• “Double Tapping”
– For another command
– Equivalent to “Double Clicking” of mouse
Converting coordinate from the map coordinate to
latitude/longitude
• Tapping points are detected by counting pixels with the
patterns and calculating center of gravity
• Estimated points are on the coordinate of RGB-D image,
they are required to be converted to latitude/longitude
– RGB-D images are rectified by perspective transform with known points of corners.
– Pairs of points between the RGB-D image coordinate and coordinates on latitude /longitude for the final conversion
Automatic identification and tracking of
tactile map
• Originally defined manually.
• ORB Feature point detector/local feature descriptor is
used for identifying tactile map by RGB image
– Comparing mean distance of feature descriptors
• Corners of the tactile map can be tracked by estimating
homography matrix between rectified image templates
and input image
– At lees 4 pairs are required for calculating homography matrix – RANSAC is used for robustly calculating homography matrix
Enlarging the map with text-to-speech reading
as the reaction of “Tapping” gesture
Mapping party
fo
People are gathered for
making map (Mapping Party)
Virtual Mapping Party with AR Tactile Map
Panoramic image and 3D-Sound
Workshop on Virtual Mapping Party
• Held at Miraikan, Tokyo, Japan
• March 18
th-19
th, 2016. 1h x 6 times
• Collecting POR/POI by experiencing panoramic images
and 3D sounds with HMD(Oculus, GearVR. Etc.), PC
and headphone
Watching panoramic image with 3D sound
for collecting POR/POI
Usage of the AR tactile map for virtual mapping
party
• Allowing the visually-impaired people to join
the mapping party by gesture
– Search: Confirming POR/POI on the tactile map with
sound for telling existence of POR/POI where user
touches
– Tap: Confirming POR/POI
with Text-to-Sound when
user taps the specific point
– Double Tap: Requesting
the POR/POI for the
Summery of a questionnaire for the AR tactile map
Comments about gesture interface
• It is fun to search POR/POI by actually touching.
• It took time to get used to “double-tap”, but finally I mastered. • It’s good concept, but needs more accuracy of detection.
• It lacks feedbacks to tell whether the system detected gesture or not. • It was difficult to use.
• Other techniques can be used for more robust detection of gesture.
• Touch panel with tactile presentation(sheets?) can be used for robustly detecting gesture.
• It is hard to distinguish between detection error and nothing there to be displayed.
Summery of a questionnaire for the AR tactile map
(cont.)
Comments about requesting function
• Requesting function is good because the it can reflect visually-impaired people’s demand.
• Filtering technique might be required for extracting important requests • It’s better to add function to send request by voice recognition.
• Currently only points of the request are defined, contents of the request might be needed (Just want to get POR/POI or feel scared)
Comments about the contents
• It seems helpful for the visually-impaired people to check obstacles such as trees.
Summery of a questionnaire for the AR tactile map
(cont.)
General Suggestions and Comments
• Scale size of the map is not fitted. Considering to accuracy of the gesture recognition, the tactile maps should be bigger in this case. • It seems interesting if it can realize dynamic tactile presentation in
some day.
• In addition to requesting, other opportunities for the visually-impaired people to join the events seem to be needed.
• Functions to increase the number of participants such as position based SNS seem to be required.
Conclusions
• AR tactile map: Audio-visual augmentation for tactile map
by RGB-D image recognition and text-to-speech.
• Audio-Visual feedbacks by AR for dynamic presentation
– Intuitive gesture interface by using RGB-D camera
• Future works
– Increasing accuracy of the gesture recognition – Covering individual difference of gesture
– Noise and unclear situation (sunlight, close to paper)
– Virtual Mapping Party with the AR tactile Map
• Acknowledgement
– This research is conducted as a part of the project sponsored by JST RISTEX.
プロトタイプでの実装の詳細
• Client
側
AR
触地図アプリ
– Android
アプリとして実装
–
カメラ映像を用いた触地図の
ID
とカメラ・触地図間の位置
関係の認識
–
ユーザの注目箇所をジェスチャ認識部と
UDP
通信
–
音声提示,視覚提示(拡大)等で情報提示
•
指ジャスチャ認識プログラム
– Windows
アプリとして実装
– DepthSenseSDK
を用いて
RGB-D
カメラ映像を処理
サイトワールド
2015
でのデモ
デモ展示等で判明した課題
•
カメラ・触地図間の位置関係がずれる
– アームに意図せず接触 – 自動的なキャリブレーション方法の必要性•
タップ動作の個人差の問題
– 個人のタップ動作をシステムへ登録する必要性関連事例
•
画像聴覚化装置
*(
画像聴覚化研究所
)
–
きっかけ:視覚障害者に美術館で絵画を体感できるように
–
音程,音が聞こえる方向により画像を聴覚化
–
触図シートを重ねることで,触覚も併用可能
AR
触地図:触地図の自由度,簡素な機器構成,
直観的インタフェースを重視,
画像聴覚化装置の音自体に位置情報を
AR
触地図とのインタラクション
• RGB-D
カメラを用いて触地図を触る動作を認識
•
前提:触地図と
RGB-D
カメラの位置関係は真上から
の撮影位置で固定
⇒背景差分が適用可能
•
距離チャネル(
D:depth
)を用いることで
…
–
触地図の色や手の肌の色に影響されない認識
–