Abstract
Perceived motion is usually consistent with image shift on the retina when self motion (eye/head etc) is eliminated. However, when estimating the actual speed of an object in the real world, observers may use other information in addition to motion signals. We investigated the effect of image size on driving speed estimation based on real-world videos. In our study we manipulated motion signals through selection of 84 driving video clips (taken from the driver’s perspective), with driving speeds ranging from 5 to 60 mph at 5-mph intervals. Each clip was 5-second long, and shown in one of 4 image sizes (44º, 33º, 21º and 16º wide, aspect ratio 0.38 to 1). Nine subjects each viewed the sequence of video clips, including all combinations of size and driving speed, in a random order, and then reported estimated driving speed. For each subject and each image size, the dependence of speed estimation on actual driving speed was described through linear regression. The slopes (mean 0.84) of the regression lines were not significantly different across image size (F(3,8)=2.69, p=0.074), and neither were the offsets (mean 5.25) of the regression lines significantly different (F(3,8)=2.24, p=0.11). The middle speed range (25–40 mph) was more accurately estimated than the lower (5-20 mph) and higher (45-60 mph) speed ranges. The mean errors in each speed range were 4.07±6.47(SD) mph for the lower range (overestimate), 0.23±4.99 mph for the middle range, and -2.57±6.32 mph for the higher range (underestimate). As we did not find any effect of image size on driving speed estimation, our results suggest that observers can calibrate their estimation of driving speed for the image scale, possibly by using reference and experience.
Keywords: Speed perception; Image scale; Optic flow
Meeting abstract presented at VSS 2012