Modeling drivers visual attention
Visual attention We know that the vast majority of crashes are caused by human error Let’s boil that down a bit though shall we? Human error can be comprised of many different aspects Experience, attention, physiology Lets “focus” on attention for now Visual attention being a construct, is comprised of numerous aspects that when combined, account for a humans level of visual attention
seev Thankfully for us, a model for visual attention has already been established SEEV (Salience, Effort, Expectancy, Value) Salience: Attention grabbing properties(ie, bright lights, flashes etc) Effort: How much effort is required to attend (head and or neck movement) Expectancy: The likelihood of seeing an event in a particular location (notifications on a mobile phone) Value: How importance of the task (keeping a safe and regular speed limit is more important than AC settings of car
SEEV Seev Together: how people do allocate attention How people should optimally focus their attention
Seev By quantifying these variables we can actually sue SEEV into a computational model: P(A) = S – Ef + Ex + V
Implications of seev The SEEV model will predict Glance time and average duration in AOI(Area of Interest) Time gaze spent away from forward view (likelihood of missing hazards) The response time of events due to salience and where gaze was focused when event occurs The model actually accounts for about 90% of variance
Studies in seev Horrey, Wickens, Consalus, 2006 applied the SEEV model to the task of driving Wanted to see how well a computational model (SEEV) could be applied to predict visual scanning of the driver Focused on the impact of IVT’s (in vehicle technologies) on detection of road hazards Participants drove in a fixed base driving simulator while having their eye and head movements tracked
Studies in seev
Studies in seev
Studies in seev While driving, participants were to keep their vehicle in the center of their lane as closely as possible They also had a secondary task where they would need to periodically decide if a string of numbers displayed on the IVT added to an even or odd sum This IV was manipulated in two ways, complexity of the string and frequency
Studies in seev SEEV was able to significantly predict the amount of scanning by the drivers Why is this important?? By incorporating SEEV, designers will be able to more accurately predict visual scanning patterns of drivers This will obviously be an important piece of information to have as IVT’s become more common not only in cars but also in planes and trains
Questions?