Wearable Virtual Guide for Indoor Navigation
Jan 29, 2016
Wearable Virtual Guide forIndoor Navigation
Introduction• Assistance for indoor navigation using a
wearable vision system• Novel cognitive model for representing visual
concept as a hierarchical structure
Introduction• Indoor localization and navigation help in
context-aware services• Challenges of sensor-based localization– Infrastructure cost– Accuracy of localization– Reliability of signals
• Of-late recent advances in wearable vision devices (google glass)
• First person view (FPV)
Introduction
• Only camera to receive visual information, but it’s alone is not sufficient, human intelligence needs to be incorporated– Representation of navigation knowledge– Designing interactions between user and the
system• Building the cognitive model is necessary and
this is absent in existing sensor-based systems
Introduction• Mimic human behaviour of wayfinding
• Go through the glass door and turn left– Easier to follow, reduces stress from user– But requires cognitive knowledge of the building
• Contribution– Model to represent cognitive knowledge for indoor
navigation– Interaction protocol for context-aware navigation
Methodology
• Cognitive Knowledge Representation• Interaction Design with Context-awareness
Area : (1)Shopping area, (2) Transition area ( and (3) Office area.The bottom level contains sub-classes of locations in eacharea. For example, the conceptual locations in the Shoppingarea are Lobby, MainEntrance, Shop, GateToLiftLobby, etc.;the locations in Transition area are LiftLobby, InLift, Entrance, etc.; the Office area has MeetingRoom, Junction, Corridor, Entrance/Exit etc. The nodes within each level are connected if there is a direct pathbetween them
Cognitive Knowledge Representation
• Hierarchical context-model
Cognitive model
• Scenes are mapped to the location/area nodes – Image classification algo
• Generate cognitive route model • Given source-destination– Chain of nodes --- area and location – Area nodes connects child location nodes
• Define Trip segments
Working principle
• In an actual navigation task with a given origin and destination– scenes (i.e., image sequences) are captured continuously
using the wearable camera. • The scenes are categorized into area nodes and
location nodes, which are compared with the nodes in the cognitive route model.
• Once a match is found between a recognized node and that in the cognitive route model– specific navigation instructions are retrieved according to
predefined rules.
Interaction Design with Context-awareness
• 3 types of contexts– Recognize egocentric scenes: localization of the
user in the environment – Temporal context: right time when the
instructions to be provided– User context: user’s cognitive status
Interaction Design with Context-awareness
• Data Collection– 6 participants in sequence– 3 destinations in sequence – M0 -> M1 -> M2 -> M3
• Different floor• Different tower
Main entrance
Training Setup
• Self reported ability of spatial cognition – Santa Barbara Sense of Direction (SBSOD)– Score is used to adjust system behavior
• Vision system-webcam-tablet PC• Resolution of 640*480 at 8 frames/sec• Send scenes to PC• Human assistant=>Explicit help/confusion
SBSOD
Localization using cognitive visual scenes
• A task-specific route model is constructed– Visual scenes are captured continuous and used to
build the task specific route model• Once route model is built, determine location• Two cues– Image categorization • Image-driven localization can achieve accuracy upto
84%
– Time
Image categorization
Improve localization using temporal inference
• Localization using cognitive visual scenes• Tackle the dynamics of wayfinding – Various walking time
Improve localization using temporal inference
• Localization using cognitive visual scenes– Li
– ti0
– Rand(p) random Number between 0 and p
First Compute
Probability that a scene is associated with that segment
Interaction Design with Context-awareness
• Context-aware navigation instructions– Provide effective navigation aids– Determine a decision point Dj and associate a
probability value Pj related with number of subjects n who requested help
– What if the user do not comply to the aids?
Interaction Design with Context-awareness
• Context-aware navigation instructions– TTj-1 narration at decision point Dj
– TTj-2 rephrased narration
Interaction Design with Context-awareness
Interaction Design with Context-awareness
• Context-aware navigation instructions– Utility of the instructions are measured using
SBSOD score– Cp denotes the spatial cognitive level
– At time tk the navigation instruction is provides as per the rules below
Experimental Evaluation• Participants– 12 participants (6 M, 6 F)
• Experiment Design– SBSOD score as input for each participant– Task1 : M0->M1, Task2: M1->M2; Task3: M2-> M3– At different order for different participant
• Measures• Hypotheses
Experimental Evaluation• Measures– Objective
Experimental Evaluation• Measures– Subjective
Experimental Evaluation• Hypotheses
Results• Task Performance– One-way ANOVA– Posthoc analysis with paired t-test
Results• Subjective Evaluation– Easy-to-use– Reduced cognitive load– More intelligent– CNG has poor performance in terms of enjoyment,
stress and trust