Perception of Motion Snehesh Shrestha, Matthew Goldberg, Virinchi - PowerPoint PPT Presentation

Perception of Motion Snehesh Shrestha, Matthew Goldberg, Virinchi Srinivas, Yehuda Katz, Michelle Mazurek, Cornelia Fermuller

Introduction ● Optical Illusions such as Leviant Illusion ● Observation ○ Spinning/ Flickering motion in static images ○ Believed universal ● Criteria for observation Angle of intersection (~90 o ) ○ ○ Density of lines

Motivation Problems Experiment Goals: ● No systematic human study ● Validate universality with human study done ● Measure variations ● Criteria ranges unknown ○ Time taken ○ Angles of intersection ○ Density of ■ Number of lines ■ Ratio of lines and space between them

METHODS: Experiment Design ● Show participants images of varying angles, density, and illusion free images at random. ● Ask participants to press a button as soon as they see illusory effects (something moving) in the image ● For control ○ Baseline reaction time ○ Random images of cars, scenery, and random patterns known to not have illusions are shown ● Measurements of time to see illusion, if they see illusion, what type of motion do they see, screen resolution, demographics are collected ● Data is analyzed to validate universality and measure effects of variations

METHODS: Design / Interface / Pilot 1. Experiment design a. Web survey - Reach large audience fast b. Keyboard Control - less variability 2. Web interface and backend - Python/Flask, MySQL, Piwik Analytics 3. Illusion images - Generated in Matlab 4. Pilot - Friends and family a. Observed and collected feedback b. Updated the interface, images etc based on the pilot.

Experiment Setup: Reaction Time Baseline

Experiment Setup: Data

METHODS: Recruitment 1. Social media a. Facebook b. LinkedIn c. Twitter 2. Email a. Community and University Email Lists b. Emails to friends and family 3. Flyers 4. MTurk a. Threat to validity b. Worldwide representative sample 5. Reddit

METHODS: Analysis ● Research Questions (3): ○ Does variation in angles/line density/ line space ratio affect the reaction time to observe any illusion? ● Null Hypotheses (3) : ○ Variation in angles / line density / line space ratio is not related to reaction time to observe any illusion ● IVs - angles / line density / line space ratio , DV - reaction time ● Each IV : Categorical DV - Numeric (Continuous) ● Normality testing using Shapiro-Wilk test rules out using parametric test ● Within-subjects design : Friedman’s ANOVA is used ● Null hypothesis: all distributions identical ○ No difference in reaction time on varying angles / line density / line space ratio

METHODS: Analysis ● Research Question (Demographics) : ○ Does Age, Race, Gender and optical defect affect the possibility of observing illusion? ● Hypothesis : ○ Age, Race, Gender and optical defect are not related to the possibility of observing illusion. ● IVs - Categorical DV - (number of images they observe an illusion) Numeric ● Multiple linear regression ● Need linear relationship from each DV to IV as well as lack of multicollinearity; in general troublesome to assess for categorical cases for all IV ● Independence among demographic data provided by participants

METHODS: Limitations ● Sample Size : A priori power analysis for 80% power & 0.1 significance level & medium effect size ○ Requires between 97 and 117 samples for hypotheses testing relation between density, angles, spacing vs time ○ Requires 140 samples for hypothesis involving multiple linear regression ● Representative Sample Distribution : Not uniformly distributed ● Not able to control screen size ● Base reaction time could not be monitored ● Distance from screen could not be controlled ● No control over environment eg. mood ● Lighting and screen brightness not monitored ● Randomize option order ● Universal claim could not be validated ● Cannot really check if a participant is lying

METHODS: Assumption ● No interaction between IVs : Demographics such as age, gender, region, race, optical deficiencies etc. do not have effect on each other ● Lighting and screen brightness has no effect on reaction time ● Screen size and screen resolution does not affect reaction

RESULTS:: Variation: Density of Lines-Space Ratio 4 8 2 16

RESULTS: Hypothesis (1a): Ratio of lines-space ● Result of Friedman’s ANOVA hypothesis test: p-value = 2.08e-12 ● Null hypothesis The distribution is uniform ● Visually evident from differences in histograms as well

RESULTS:: Variation: Density of Lines

RESULTS:: Variation: Density of Lines 2 4 8 32 96 120 =2

RESULTS: Hypothesis (1b): Number of lines ● Result of Friedman’s ANOVA hypothesis test: p-value < 2.2e-16 ● Null hypothesis Distribution is uniform ● Evidence that categories are not all identically distributed

RESULTS:: Variation: Angles

RESULTS: Hypothesis (1c) : Angles ● Result of Friedman’s ANOVA hypothesis test: p-value = 0.9637 ● Extremely bad p-value; null hypothesis unaffected ● Supported by histogram examination of distribution location

RESULTS: Hypothesis (2): Demographics lm(formula = one ~ AgeGroup + Gender + Race + lensGroup, data = new_data) Residuals: Min 1Q Median 3Q Max -3.8256 -0.7904 0.0000 1.0105 2.5935 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.8904 1.7822 -0.500 0.619895 AgeGroup2 0.8904 0.6258 1.423 0.162014 AgeGroup3 -0.1399 0.9528 -0.147 0.883913 AgeGroup4 -0.8936 1.0421 -0.858 0.395906 AgeGroup5 -0.9142 0.8124 -1.125 0.266730 AgeGroup6 4.0209 1.7723 2.269 0.028358 * Gender2 -0.2030 0.5003 -0.406 0.686961 Gender3 -2.6200 2.2434 -1.168 0.249302 Race2 0.3830 1.3837 0.277 0.783235 Race3 1.0439 0.5839 1.788 0.080854 . Race4 2.8404 1.9643 1.446 0.155425 Race5 3.5914 1.4505 2.476 0.017301 * lensGroup2 8.9647 1.8474 4.853 1.64e-05 *** lensGroup3 12.7401 2.5890 4.921 1.31e-05 *** lensGroup4 9.0286 1.7929 5.036 9.01e-06 *** lensGroup5 9.2940 1.9775 4.700 2.69e-05 *** lensGroup6 11.1672 1.9605 5.696 1.01e-06 *** lensGroup7 10.4496 2.1821 4.789 2.02e-05 *** lensGroup8 9.7607 2.2331 4.371 7.70e-05 *** lensGroup9 8.8904 1.9638 4.527 4.68e-05 *** lensGroup10 6.1810 2.1344 2.896 0.005922 ** lensGroup11 11.2030 2.4124 4.644 3.22e-05 *** lensGroup12 9.1591 2.4994 3.664 0.000676 *** lensGroup13 7.0934 2.4894 2.849 0.006697 ** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 1.669 on 43 degrees of freedom Multiple R-squared: 0.6209, Adjusted R-squared: 0.4181 F-statistic: 3.062 on 23 and 43 DF, p-value: 0.0007505

RESULTS: Demographics Distribution: VISITORS Total Participants 260 Valid Participants 67

RESULTS: Demographics Distribution: RACE

RESULTS: Demographics Distribution: GENDER

RESULTS: Corrective Lens Effect

RESULTS: Demographics Distribution: AGE

RESULTS: Age vs Reaction & Illusion Seeing Time

DISCUSSION and CONCLUSION 1. Results show that changes in density is related to how fast and more people observing the illusion. Changes in angles does not seem to affect. 2. There were many challenges , however, we learned a lot and esp. How to reduce the risk to validity to our tests. 3. Even though we did not have sufficient samples , the results look promising and with lessons learned from this full process, we can use this as a pilot and conduct a larger and stronger experiment. 4. Future work : This summer we plan to continue this work under Dr. Fermuller and Dr. Mazurek's guidance.

Q&A Thank You!

Appendix

Histogram of did NOT see across different params

RESULTS: Did NOT see distribution

Rough outline for to follow 1. Intro/Motivation/ Background 2. Method Details a. Overview/ Plan b. Design/ Interface/ Pilot c. Recruitment d. Methods analysis (Assumptions…) 3. Results a. Data Overview (distribution, results of the tests and interpretations) b. Discussion on the hypothesis 4. Discussions a. Implication of the results b. Limitations (Challenges, …) c. Next Steps/ Future Work d. Lessons Learned from this study as pilot (Coulda/Shouda, ...) 5. Conclusion