1. Change detection and occlusion modes in road-traffic scenarios Professor: Liu Student: Ruby

2. Purpose & Motivation Purpose –Finding what amount different occlusions may cause to the failure of change detection in road traffic. Motivation –Change blindness are known in cognitive science, but the authors think the relation to driving is not clear.

3. References YearAuthors Results 1997Simons & Levin The observers are actively searching for the changes or directly looking at places where they occur. 1.1996 2.2000 3.1996 4.1997 5.1995 6.1997 7.1998 1.Grimes 2.O’Regan, deubel, Clark, & Rensink 3.Simons 4.Rensink, O’Regan, & Clark 5.Blackmore, Brelstaff, Nelson, & Troscianko 6.Levin & Simons 7.Simons & Levin This ‘change blindness’ is typically related to natural global disruptions of the retinal image, such as: 1.saccades 2.eye blinks 3.blink screen 4.flicker 5.picutre shifts 6.film cuts in motion pix 7.physical occluders

4. References YearAuthorsResults 1991 Macknik, Fisher, & Bridgeman The flicker has a less natural character and may interfere with different aspects of perceptual processing. 1997Rensink et al. Changes to objects of central interest are detected more quickly than changes to objects of marginal interest. 1984 Stelmach, Bourassa, andDiLollo Adding an item to the visual array was more likely to be detected than deleting one. 2000 Werner and Thies In expert-novice comparison of change detection in football images, domain-specific expertise decreases change blindness.

5. References YearAuthorsResults 2001 Reingold, Charness, Pomplun, & Stampe Experts’ advantages in detection of changes to meaningful structures. 2000Barbato et al. People blinks 12–15 times per minute lasting 150 ms or more. 1978Nagayama More than 50% of all collisions in road traffic can be lead back to a missing or delayed hazard perception.

6. Experiment 1: Change detection in static road traffic scenario 2.1 Method 24 participants (11 male and 13 female) viewing distance 60 cm. SR-Research EyelinkTM recorder the eye movement. Three different occlusion modes—blinks, saccades and artificial blanks were showed randomly. Press a button as soon as they noticed a change.

7. Experiment 1: Change detection in static road traffic scenario Fig. 1. Sample image-pairs used in the experiment: (a) relevant change: a bicyclist on the right side, (b) irrelevant change: an advertisement object on the right side.

8. Experiment 1: Change detection in static road traffic scenario The detection rate of occluded relevant changes was much higher than that of irrelevant ones: 80% vs. 34%, respectively (F1,18 =360.2; p <.001). The relevant changes were detected 150 ms faster than the irrelevant ones (F1,18 =6.32; p <.013). 2.2 Results

9. Experiment 1: Change detection in static road traffic scenario In all occlusion modes, insertions were detected better than deletions, with the detection rate for insertions equal or above 90% (for all comparisons, p <.001) Insertions were detected in all occlusion modes faster than deletions (for all comparisons, p <.001)

10. Experiment 1: Change detection in static road traffic scenario 2.3 Discussion –When the duration is about 110 s, blanks can simulate blinks and saccades in static environments. –Relevant changes were detected faster than irrelevant ones. –In road traffic the more hazardous events that are related to the emergence of objects.  the more hazardous changes are detected better and faster than the less hazardous ones.

11. Experiment 2: Change detection in a dynamic road traffic scenario 3.1 Method 12 participants (6 male and 6 female) ‘SIRCA’ driving simulator SR-Research EyelinkTM system was used to monitor participant’ eye movements. changes of the virtual environment was about 70 ms. 50 different real life photographs (persons, animals, traffic signs, etc.) 65% of changes appeared on the right side.

12. Experiment 2: Change detection in a dynamic road traffic scenario Changes were presented randomly every 16– 24 s either after a blanked screen or a fixation, in proportion 40:60. Four 12 min-trials in the driving simulator.

13. Experiment 2: Change detection in a dynamic road traffic scenario 3.2 Results –Detection rate (F1,11 = 21.896; p <.002) and detection time (F1,11 = 6.253; p <.030) were significant for the blank condition compared to the fixation condition. –The average false alarm rate in blank situation was 1.9%.

14. Experiment 2: Change detection in a dynamic road traffic scenario

15. The change detection rate be lower in saccades than in blinks (87% vs.100%, respectively), participants also needed 80 ms or more to react to changes if they were produced during saccades.

16. Experiment 2: Change detection in a dynamic road traffic scenario 3.3 Discussion –In the dynamic setting, the blank occlusion leads to an improvement in accuracy and speed of detection. –Occlusions of about 100 ms and more disrupt the perceptual continue of visual events, transforming them to a short static shoot. –Dynamic data in line with the confusing masking-of-masking effects related to a recovery the visual target.

17. Conclusions Changes to environment in dynamic road-traffic scenarios are more difficult to detect in static situations. only a changes to a scene such as relevant insertions is importance for driving which are highest accuracy and fastest reaction times. Saccadic eye movements and blinks, selectively interfere with perception and attention in motion flow fields which may even improve detection.