The Perception of Walking Speed in a Virtual

EnvironmentBy T. Banton, J. Stefanucci, F. Durgin, A. Fass, and D. Proffitt

Presentation by Ben Cummings

For Animation (CS551-4), Fall 2003, with Dave Brogan

Perception of speed

● Important for simulation of motion in virtual environments (driving or flight sims, for example)

● Influenced by many cues, though we'll deal primarilly with visual and proprioceptual (physical sensation of body movement)

Radial Optical Flow

● Results from looking down the axis of movement

● Things expand from the vanishing point faster when they're closer to the viewer

Lamellar Optical Flow

● Results from looking closer to perpendicular to the axis of motion

● Objects' speed in image space stay constant when perpendicular

Optical Flow

● When looking forward and moving, radial optical flow occurs towards the center of the view and lamellar flow occurs towards the periphery

Perception of movement speed

● In reality, under normal conditions, people are usually good at judging their speed of movement

● In virtual environments, there is markedly different performance– Particularly, in situations on treadmills with

simulated correct optical flow, subjects perceive the speed of the visuals to be slower than their own speed

Perception of movement speed

● The claim: The difference in actual motion and perceived motion is due to decreased lamellar flow in virtual environments– Display technology often cuts off the peripheral

vision where lamellar flow is seen● Head mounted displays (HMDs) are usually about

one third of the natural field of vision● Wall-mounted screens also usually occupy a small

fraction of the natural field of vision

Experiment One

● Participants walk on a treadmill wearing a HMD with simulated optical flow and try to match the speed of walking to the speed of flow– Participants looked straight ahead– Treadmill ran at 3 mph ( a fast walk )

Experiment One: Results

● Subjects chose an optical flow corresponding to 4.6 mph to match the speed of walking

● Straight-ahead optical flow in the HMD is perceived to be too slow for the actual movement speed

Experiment Two

● This experiment is the same setup as experiment one, but instead of the subjects looking straight ahead, they look perpendicularly to the direction of motion– Both looking down at the ground– And looking over to a point on the horizon

● This maximizes lamellar flow

Experiment Two: Results

● When looking both down and over, participants chose the optical flow speed that corresponded with the walking speed– the error of their guesses increased with walking

speed, but without general over or underestimation

Experiment Three

● Does the walking simulation cause the misperception of speed?– Subjects walk at normal speed and in “baby

steps”● Although perception differs, subjects

perceived the optical flow to be closer to correct when using baby steps, so stride-length cannot be used to account for the error in estimation from experiment one

Experiment Four

● Is misperception caused by image latency or jitter?

● Experiment with perception when the eye point is tracked to head movement (which introduces latency) vs. when it is constant.

● Result: speed is similarly misperceived with or without jitter and latency.

Discussion

● Hypothesis is consistent with the effects of different fields of view in speed estimation– Cycling speed underestimated with FOV < 73

degrees, and overestimated with FOV > 103 (Van Veen, et al. 1998; Osaka 1988)

– What does this mean for perception of speed in games with variable FOV?

– Other functions such as estimation of time to contact and simulated flight accuracy get better with larger FOV (to a point)

More Discussion

● Some applications are not conducive to turning to the side (e.g. driving sims)– They suggest “adjusting the gain of optical flow”

when facing forward. How would this work while maintaining geometric consistency?

Questions?