Cognitive Load: Assessment with Visual Behavior

0 downloads 0 Views 867KB Size Report
Oct 24, 2016 - Gaze Concentration: Visual-Manual. Baseline. Visual Manual Task. (Victor, Harbluk, Engström , 2005). 22. Fixation Density Plots ...
Cognitive Load: Assessment with Visual Behavior Cognitive Load and In-Vehicle HumanMachine Interaction Workshop Joanne L. Harbluk Transport Canada Auto UI 2016, Ann Arbor 24 October 2016

Overview 1. A few words about Cognitive Load 2. What does the impact of Cognitive Load look like? 3. Approaches to Cognitive Load using glances 4. Standard - ISO 15007 (revision & new development) 5. Challenges, future directions

2

Cognitive Load: Mental resources required to perform a task: planning, decision making, error detection, inhibiting habitual actions, utilizing information in working memory and resolving novel and complex situations

Cognitive Distraction: Diversion of mental

resources from driving in dual-task conditions

3

Visual: eyes off the road Manual: hands off wheel Cognitive: mind off driving V V

M

Texting C

M

Hands free phone C

V

M C

Reaching

Cognitive Demand Is Part of Most Tasks 4

[Lee, 2009; Distracted Driving Summit]

Tasks Can Be Very Cognitively Demanding: Email reader

(From Harbluk & Lalande, 2005) 5

Tasks Can Be Very Cognitively Demanding: Email reader

• Slower detection of events in the environment • Poor speech quality resulted in missed events (From Harbluk & Lalande, 2005) 6

Demand on the driver is more than just the mode of presentation “Driving is a visual manual task” – present a task in a non-visual format (e.g., speech)

Motivation: – presentation format vs processing of that information – Might account for conflicting results in literature (Trbovich, 2007) 7

Logic of study: – Driving has strong visual demand – Different formats for presentation (auditory, visual) – Tasks require different processing resources (phonological, visual, spatial)

8

Task Processing Requirements: – Phonological task: 2 words “compute pewter” • Subvocal speech: Common phonetic sound? “pute”

– Visual task: 3 words “LUCKY, FURRY, DICE” • Visualize the words; “yes” if at least 2 words contain a closed letter; else no

– Spatial Task: 2 times to compare (8:30 & 9:00) • Imagine analog clock faces: which of the two times forms the greater acute angle (90deg) 9

10

11

Type of Processing Required for the task (Trbovich, 2007)

12

Visual Presentation Mode: Reduced driving performance (looking away from road)

Auditory Presentation Mode: Reduced driving performance for tasks requiring visual spatial processing

(Trbovich, 2007)

13

2. The impact of cognitive load looks like?

Effects of cognitive load are difficult to observe Video example:

(Harbluk, Noy, Trbovich & Eizenman, 2007)

14

Visual Scanning: No Additional Task

15

Visual Scanning: Cognitive Task

16

3. Approaches to Cognitive Load 1. Changes in where drivers look inside & outside the vehicle 2. Concentration of gaze in forward view

Areas of Interest (AOI)

Measures of Gaze concentration

17

Areas of Interest (AOI) Visual Behavior: • location, duration & frequency of glances Area of Interest (AOI): predetermined area within the visual scene • # of times a driver looks to that area • Driving time spent looking at that area • % of task time spent looking at that area • Fail to look 18

19

Changes seen in AOIs With increased cognitive load: • Look less to relevant places: intersections, traffic lights, instruments etc • Look at the device itself (e.g., NHTSA, 2003) Is visual confirmation needed? Necessary to explore the data; you may not know what those changes are..... A lot to be learned from actually watching what the drivers do; video & audio of the session very helpful 20

Gaze Concentration • Tendency not to look around – consistent with AOI neglect • Looking at the road, but this behaviour needs to be considered in the context of the tasks and the driving situation

21

Gaze Concentration: Visual-Manual

Baseline

Visual Manual Task Fixation Density Plots 22 (Victor, Harbluk, Engström , 2005)

Gaze Concentration: Cognitive Task

Baseline

Auditory Task Fixation Density Plots 23 (Victor, Harbluk, Engström , 2005)

Percent Road Centre Percent Road Centre (PRC): The % of fixations directed towards the road centre during a task (Victor et al., 2005) Intuitive measure Interpretation: Greater cognitive load results in more concentrated visual behavior Variations in the procedure • Definition of centre: circle 16 degree diameter centered around road center point; also 8, 12, 16 degrees (Wang et al., 2014) • May be influenced by driving environment and task (road type, simulator) • Rectangle 20 degree horizontal by 15 degree vertical rectangle • % of time spent looking in that area •

type of data used; ease of analysis, type of analysis packages used; variation in how “center” is defined, area as circle, rectangle, size varies 24

Changes in Gaze Behavior % change in time spent as cognitive demand increases

(Harbluk, Noy, Trbovich & Eizenman, 2007)

25

Standard Deviation of Gaze Standard Deviation of Gaze: is the standard deviation of the combined horizontal & vertical angles Interpretation: The smaller Standard Deviation of Gaze, the more concentrated the visual behavior as a result of increased cognitive workload Variations: • Often only horizontal gaze is used • gaze angle or gaze position (what is available by system)

(e.g., Recartes & Nunes, 2003; Sodhi et al., 2002, Wang et al., 2014)

Standard Deviation of Gaze

27

4. Measurement of Driver Visual Behaviour – Revision & Elaboration • ISO 15007-1 & 2 • Key terms & parameters for analysis of driver visual behaviourglances & glancerelated measures • Common reference, consistency in approach, empirically based methods, best practices 28

• Equipment & procedures for analysing driver visual behaviour • To include:

Technical Specification

• Planning of evaluation trials • data capture equipment • Analysis, interpretation and reporting • add video reduction etc. 29

5. Challenges and Future Directions • • • • •

Glances: looking but attention? Processing? Significance? Decision making with results. Often only have relative data; no redline. What metrics to use & how many do we need? Safety impact, crash relevance? Technical aspects of dealing with the data

• Glance data is very compelling from a safety perspective. • Choice of metric: who is your audience -Policy/science • Assessment of optimum workload, measures of engagement? • Turn distraction around- is enough attention being paid? How do we know that drivers are sufficiently engaged?

Thank you for your attention! [email protected]

31