Report

Non-Overlapping Aggregated Multivariate Glyphs for Moving Objects Roeland Scheepens, Huub van de Wetering, Jarke J. van Wijk Presented by: David Sheets Problem • Address Visual Clutter in… – High density areas – Low resolution screens (e.g. mobile phones) • Clutter makes it difficult to… – Identify points of interest – Find objects that are occluded by other objects Before After Requirements 1. No overlap or occlusion between visual representations of the subsets 2. Subsets as small as possible 3. User can estimate point density of areas 4. User can recognize patterns in the attributes 5. User can see more detail by zooming in 6. Areas of inﬂuence of different subsets do not overlap 7. Small changes in object positions have small effects on the partition Other Requirements • Position of objects must be maintained – Or at least close • Support real-time streams of data Related Work • Clutter Reduction – Resampling to approximate original – Interactions to explore dense regions – Displace objects to prevent overlap – Clustering to reduce clutter – Aggregation (using Multivariate Glyphs) Related Work • Multivariate Glyphs – Replace a large collection of crowded glyphs with a single, larger glyph – Glyphs are stacked to represent multiple objects – Glyphs represent multiple dimensions • x, y, direction, average, variance, etc. – Pie chart glyphs to show distribution Technique • Divide object set O into a partition {S1,…,Sm} of non-empty disjoint subsets Si that span O. • Each subset has a circular area of influence defined by the centroid cS and radius rS=r(|S|) – r is a function of the number of elements in S • Radii are projected into screen space to deal with zoom • Can now define measures of overlap Measures of Overlap A. Overlap of the area of influence B. Penetration depth (easier to calculate) Measures of Overlap • If , > 0, subsets S and T overlap Partitioning • INIT: First create partition K containing a singleton {o} for each ∈ • MERGE: While there exists pairs of subsets S and T in K where , > 0, merge the pair with maximum overlap. 1. Subsets as small as possible 2. Areas of inﬂuence of different subsets do not overlap • Update addresses each subset in partition K by running INIT and then MERGE Partitioning Visualization • Encoded values – x(t), y(t), hdg(t), vessel type, velocity – Size of glyph represents number of objects it represents – Represent distribution in a pie chart – Heading encoded as oriented bar chart – Velocity reduced to moving | stationary Visualization A. Distribution of object types B. Direction of objects C. Proportion of objects that are stationary Visualization Variations: Visualization A. Mouseover a glyph shows the spatial distribution of objects represented by the glyph B. Clicking a glyph shows statistics for the glyph Animations • Visualizing moving objects – Objects split from a merged glyph – Objects merge into a glyph • Animation is used to illustrate the change – Linearly interpolated between states Interaction • Mouseover • Click • Panning and Zooming – Zoom changes screen space and recalculates merges Evaluation • Proposed (Mpart) • KDE (Mdens) • Single Point (Msingle) Evaluation • Proposed (Mpart) • KDE (Mdens) • Single Point (Msingle) Evaluation • Proposed (Mpart) • KDE (Mdens) • Single Point (Msingle) Evaluation • Proposed (Mpart) • KDE (Mdens) • Single Point (Msingle) Evaluation • Tested – Static Visualization • Ability of subjects to recognize density & patterns – Dynamic Visualization • Situational awareness • Tasks (3 static, 1 dynamic) 1. Which square contains more points? 2. Which square contains more blue points? 3. Which square contains more blue points heading approximately North-East? 4. When a quadrant no longer contains both blue and red objects, press its number. Evaluation • Task 1 given with varying number of points – 50, 500, 1000 • Task 2 and 3 use random number of points – Points to identify based on % of points • Small (5%), Medium (10%), Large (15%) – Difference between left and right vary • Small (5%), Medium (10%), Large (15%) 1. Which square contains more points? 2. Which square contains more blue points? 3. Which square contains more blue points heading approximately North-East? Evaluation • Task 4 – Three variations to distribute data in each quadrant • • • • 4. Green,Red&Blue 100,1 200,2 300,4 When a quadrant no longer contains both blue and red objects, press its number. Results Results 1. Which square contains more points? 2. Which square contains more blue points? 3. Which square contains more blue points heading approximately North-East? n, number of points ns, number of special points p, percent difference special points left & right ps, percentage special points Results 1. Which square contains more points? 2. Which square contains more blue points? 3. Which square contains more blue points heading approximately North-East? n, number of points ns, number of special points p, percent difference special points left & right ps, percentage special points Results • Tukey’s HSD post hoc at 5% significance level Results • Participant Questionnaire Summary – Mpart • Intuitive and less clutter • Mixed on heading ring • Animation at high speeds is distracting – Msingle • Simple and intuitive at low density • Occlusion is a problem • Easy to visualize moving objects – Mdens • • • • Distribution of objects is easy Low clutter Direction is difficult to interpret Moving objects difficult to interpret Author’s Conclusion • Benefits – Method is comparable and competitive to existing methods. – Clutter is reduced – Positive feedback from users • Future Work – Heading ring needs improved – Aggregation makes comparing individual items more difficult. Additional interactions may improve that. – Animation needs improvement for faster moving objects – Test using domain experts Other Thoughts • Change the heading ring to triangle instead of bar chart to better represent direction. • Using domain experts for evaluation.