A comparison of ray pointing techniques for very large displays

A comparison of ray pointing techniques for very large displays
of 10
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Similar Documents
  A Comparison of Ray Pointing Techniques forVery Large Displays Ricardo Jota 1 , Miguel A. Nacenta 2 , Joaquim A. Jorge 1 , Sheelagh Carpendale 2 , Saul Greenberg 2   1 INESC-IDTechnical University of Lisbon, Lisboa, Portugal 2 Department of Computer ScienceUniversity of Calgary, Alberta, Canada,,{miguel.nacenta,sheelagh.carpendale,saul.greenberg} ABSTRACT Ray pointing techniques such as laser pointing have longbeen proposed as a natural way to interact with large anddistant displays. However we still do not understand thedifferences between ray pointing alternatives and how theyare affected by the large size of modern displays. Wepresent a study where four different variants of ray pointingare tested for horizontal targeting, vertical targeting andtracing tasks in a room-sized display that covers a large part of the user‟s field of view. Our goal was to better unde r-stand two factors: control type and parallax under this sce-nario. The results show that techniques based on rotationalcontrol perform better for targeting tasks and techniqueswith low parallax are best for tracing tasks. This impliesthat ray pointing techniques must be carefully selected de-pending on the kind of tasks supported by the system. We also present evidence on how a Fitts‟s law analysis based on angles can explain the differences in completion time of tasks better than the standard analysis based on linear widthand distance. Author Keywords Large displays, ray pointing, distant pointing, image-plane,targeting, tracing, index of difficulty, ISO 9241, parallax. ACM Classification Keywords H5.m. Information interfaces and presentation (e.g., HCI):Miscellaneous. INTRODUCTION People often interact with a large digital display by distant  pointing, or directly pointing at the display with their fin-ger, laser pointer, or other input device.  Ray pointing is aclass of techniques that uses ray casting (the intersection of a ray with a surface) to determine where a person is point-ing to, i.e., the precise cursor position on the distant display.Ray pointing is advocated as a natural interaction techniquewith these displays[2,18]as: it does not require any physi-cal surface to operate on (as opposed to mouse-based point-ing)[26,25]; it gives access to distant locations withoutrequiring displacement (as opposed to direct-input)[19,29];it is easily understood by people as it builds upon everydaypointing gestures[29]; and it allows multiple users to inte-ract on the same display without their bodies physicallygetting in the way[16]. Thus it is no surprise that ray point-ing is increasingly used in both commercial and researchsystems, especially for large horizontal and vertical displays[28,6,19,16]. Even game consoles are shifting towards raypointing for interaction (e.g., Nintendo Wii).Digital displays are becoming increasingly large, with manysites now reporting one or more wall-sized displays. Theissue is that there are many different ray pointing tech-niques, and we do not understand the differences betweentheir performance characteristics, especially in terms of how they are affected by these very large displays. Whilethere is prior work on ray pointing, it is limited. Some con-centrate on specific type of ray pointing (laser pointing)[25,13], neglecting other alternatives. Others limit pointingactivity almost exclusively to targeting tasks[13,11].Moreover, such targeting is often studied only within smalldisplay areas, i.e., where pointing ranges cover only a smallpart of the visual angle of users[15,13].In this paper we study four different variants of ray pointingtechniques over a large display, where people perform bothtargeting and tracing tasks. Our variants cover a wide spec-trum of ray pointing possibilities: laser pointing, arrow pointing, image-plane pointing and fixed-srcin pointing .As will be explained, the first three are commonly used inreal-life activities such as art and sports. Figure 1. Four variants of ray pointing: A) laser, B) arrow, C)image-plane and D) fixed-srcin. ABCD   Jota, R., Nacenta, M.A., Jorge, J.A., Carpendale, S., and Green-berg, S. A Comparison of Ray Pointing Techniques for VeryLarge Displays.Report 2009-942-21, Department of Computer Science,University of Calgary, Calgary, AB. Canada, T2N 1N4    Specifically, we conducted three studies that tested the fourtechniques for horizontal targeting, vertical targeting, andtracing tasks on a room-sized display that covers a large  part of the user‟s field of view (>90°). Our study differs from previous work in three fundamental ways:We compare several quite different ray pointing va-riants rather than investigate just a single method;We test tracing tasks as well as the usual targeting;We specifically consider issues associated with verylarge displays, e.g., when comparing displayed objectsin front of the person vs. at the side, the person will seethese objects at quite different distances and angles.Our results show that techniques based on rotational control(laser pointing and arrow pointing) perform better for tar-geting tasks, while techniques with low parallax (arrowpointing and image-plane pointing) are best for tracingtasks. This implies that ray pointing techniques must becarefully selected depending on the kind of tasks supportedby the large display.We also found that targeting and tracing tasks are heavilyaffected by the location on the large display in which they take place. We present evidence on how a Fitts‟ s law analy-sis based on angles vs. the standard linear width and dis-tance approach explains differences in completion time of tasks better.The remaining of the paper is organized as follows. Weexplain ray pointing fundamentals, and describe the tech-niques and review previous research in the area. We thendescribe the experiment and results. We conclude with adetailed discussion of the implications of the findings. RAY POINTING FUNDAMENTALS We now describe the particular ray pointing techniques weuse and how we implemented them in our study.We define generic   ray pointing as any cursor-movementtechnique that determines the position of the cursor throughthe intersection of a ray with a distant object or surface (seeFigure 2,left). For our purposes, the distant object or sur-face is a large display. Regular Laser Pointing The most common ray pointing variant is laser pointing .Here, the ray is specified directly by the position and direc-tion of a physical device (Figure 1a). The device might ormight not be an actual laser; in fact, the only requirement isthat the computer system has a way of determining the in-tersection of the ray with the screen surface. For example,vision technology or special markers on the hand recogniz-es finger postures as a pointing device[28].Laser pointing has been proposed and implemented for cur-sor control many times (e.g.,[2,18,6,1,25,5,27]). It is oftenreferred to as distant pointing, remote pointing or virtualpointing. In our study, we implement a laser pointer via aninfrared-marked wand tracked in 6DOF. Arrow Pointing  Arrow pointing is a variant of laser pointing, where we con-straint the use of the laser pointer to be somewhat aligned with the user‟s eye  (Figure 1.B). This mimics the real lifeway people aim when great precision is required (e.g., whenusing bow and arrow, or playing darts). Our implementationis identical to laser pointing, except now people are in-structed to constrain their use of the wand by looking downits shaft at the screen, i.e., as if it were an arrow. Image-Plane Pointing An alternative ray pointing technique comes from the visual arts. Painters are often taught to place their thumb at arm‟s length between their eye and a painting to estimate the sizesand positions of painted objects. This technique has longbeen adopted in the field of virtual reality, where it is re-ferred to as image-plane manipulation , occlusion selection ,or the crushing heads technique [22,7,31,4,30, 12].The mechanism of image-plane based pointing is simple:instead of determining the ray through the position andorientation of a pointing device, the ray is determinedthrough two points in space: the use r‟s eye location, andanother point in space that the user can control (e.g., theposition of the tip of the thumb, of a pen, or the point of apointing device  –  Figure 1.C). The effect is that the user cansee the cursor aligned with the thumb (or device) in herfield of view, even if they are actually at different depths(Figure 3). To a certain extent, image-plane pointing is sim-ilar to direct-input techniques (e.g., direct-touch) in thatfeedback and input overlap in the visual space of the user.Image-plane techniques require tracking (or approximating)the eye position, and are usually calibrated so that the do-minant eye image aligns the finger or device with the cursor(however, binocular focusing on the distant surface stillimplies that two separate images of the finger or device areperceived by the user as inFigure 3). In our studies we ap-proximate eye position  –  the first point of the ray  –  in realtime by placing markers on a hat; a person calibrates thevector between hat and eye before interaction by specifying Figure 3. Image-plane pointing seen binocularly and focusedon the distant display (the cursor is displayed on the screen).Figure 2  . Left) Ray pointing specifies the location of the cursorthrough the intersection of a ray (s) with the display surface. Center) the ray (s) can be specified through a point (A) and adirection,   or  Right) through two points (A and B) ACACBC  the position of their dominant eye with another marker. Weuse the tip of a wand to specify the second point of the ray. Fixed-Origin Pointing We can relax image-plane pointing by placing one of thetwo points of the ray onto any fixed location (instead of theeye). The user still controls the other point, or may evencontrol the two points separately (e.g., one with each hand).The former alternative  –  which is the one we study  –  wasexplored somewhat by Shoemaker and colleagues in sha-dow reaching  [23] . Shadow reaching allows the control of alarge display through the shadow cast by a person on alarge display illuminated from a fixed point behind the per-son. Because shadows are cast in straight lines, shadowreaching is geometrically equivalent to fixing point A onthe location of the light source and using the pointing ges-ture of the person (usually the finger) as point B. Shoemak-er et. al. also speculate using a virtual light source thatwould move with the user at a certain fixed distance. Thelatter alternative (allowing the user to control both points) isakin to pointing using the position and direction of an elas-tic band held with two hands.We tested only  fixed-srcin pointing , where the srcin pointof the ray is fixed in space. The user controls the other pointto specify the ray‟s direction. We use an srcin point nearthe bellybutton of the user so that the required pointing de-vice movements are somewhat similar to shadow reaching,where the light source is located close to the floor and be-hind a person. RELATED STUDIES Previous research in ray pointing falls mostly into two cate-gories: laser pointing studies for interaction on distant dis-plays, and virtual reality techniques for manipulation of objects in 3D environments. At the end of this section wealso discuss enhancements to ray pointing techniques. Laser Pointers for Large Displays  Thanks to the studies in this first category we now know agreat deal about laser pointing. For example, MacKenzieand Jusoh[13]and Stuerzlinger and Oh[25]showed that laser pointing targeting performance is poor compared tothe mouse (and around 1.4b/s or 3.0b/s respectively). Peck [21]parameterized the jitter of a laser pointer spot in termsof angle, and suggests that grip affects it. Myers and col-leagues[14]studied the effect of different grips and post-ures, and found reduced jitter with a PDA-pointer held withtwo hands. Kopper and colleagues[11]proposed severalmodels for pointing tasks on very large displays that takedistance into account.Most studies of pointing for large displays, with the excep-tion of [9], only test laser pointing techniques. Our studycompares a broader range of ray pointing techniques. At thesame time, we pay special attention to the effects of verylarge displays in performance (similar to[11]). Pointing in VR  The variety of pointing techniques studied in the VirtualReality literature is broader, since image-plane techniquesare easy to implement (the required tracking of head or eyesis already present). In general, studies comparing image-plane selection to ray casting (laser pointing) for manipula-tion of 3D objects in 3D spaces have found that the imageplane method is generally faster[12, 3, 4, 31, 30]. This ledHill and Johnson[7]to propose an interaction manipulationframework based on image-plane techniques. However,most of the above-mentioned studies concern 3D tasks,which can be radically different to the 2D tasks that are ourconcern. It is not yet clear whether image-plane techniqueswill provide performance advantages for pointing to large2D surfaces. Laser Pointer Enhancements  Several enhancements have been proposed that modify orimprove the operation of ray pointing and distant pointinginteraction. For example, laser pointers are often filtered[6, 28], its CD gain altered[10], alternative modalities blended into the action[27], and snapping mechanisms added[27,  15]. We know that some of those mechanisms may improvepointing (e.g., fitering) but these also imply trade-offs (e.g.,filtering implies delay[20], and semantic snarfing  [15] makes it harder to operate with empty space). We chose notto alter the basic elements of pointing partly because thesemodifications can introduce a large number of parametersthat can obscure fundamental effects we are after. RAY POINTING PERFORMANCE FACTORS There are many possible factors that might affect ray point-ing performance (e.g., grip, number of hands and filtering[14, 6]). In our study we concentrate only on control type  and  parallax , as described below. Table 1. Technique classification according to the factors.    P  a  r  a   l   l  a  x   Control  Rotational Positional None Laser pointing Image-plane Some Arrow pointing Fixed-srcin Control Type As explained previously, the ray of ray pointing can be spe-cified through two points, or through a point and a rotation.Although geometrically equivalent, our four control types  (Table 1) result in different types of end-user movement.For example, people that use laser pointing and arrowpointing specify the position of the cursor mostly throughthe rotation of the device (we call this the rotational controltype ), whereas image-plane and fixed-srcin techniquesonly require the specification of a single position on spaceand the orientation of the limbs or the device is mostly irre-levant (we call this the  positional control type ). Withinthese categories, we studied our four previously describedmethods, chosen as they represent design points in the de-sign space defined by these two factors(Table 1). Parallax Our other factor of interest is visual parallax. We definevisual parallax as the distance between the center of rota-    tion used to specify the pointing direction (usually a device)and the point of view of the user. In real-life aiming activi-ties, parallax is usually avoided if precision is important.For example, sharp shooters align themselves in the direc-tion of their weapons so that the line of view coincides withthe shooting direction.Our four techniques vary how people perceive parallax. Theimage-plane technique is, by definition, devoid of parallax.Arrow-pointing transforms laser pointer into an almost pa-rallax-free technique, as the person aligns the pointing de-vice with her line of sight. EXPERIMENT 1: HORIZONTAL TARGETING Our first experiment tested targeting in the horizontal di-mension. We were interested in testing targeting separatelyon this dimension because large displays (e.g., room-sizeddisplays) tend to be much broader than tall, which impliesthat any effects due to the size of the display and the obliq-uity of distant areas would be most evident in these tasks,especially if the participant is close to the screen(Figure 5). Apparatus We used a large wall display (292cm x 109cm) composedof 4x2 modular back-projected displays, each one with aresolution of 1024x768px (for a total of 4096x1536px). Themodular displays are adjacent to each other with just-perceptible but very narrow image seams (under 2mm). Thedisplays rest on a table 76cm high so that the participant‟s head lines up approximately with the center of the top rowof displays (seeFigure 6).To accentuate the effects of large display widths, we askedparticipants to stand on a location approximately 73cmfrom the display and 36cm from its right edge  –  seeFigure6). From this point of view, the display covered approx- imately 100º of the user‟s ho rizontal field of view, and 68ºvertically.We implemented the ray pointing variants using a 25cmwand and a cap equipped with reflective markers, whosepositions were tracked by a VICON® motion capture sys-tem. The position of the dominant eye of the user was up-dated in real time by suing the position and orientation of the cap and the calibration data obtained before each block that involved the image-plane technique. Participants se-lected targets by clicking a mechanical button held in theirnon-dominant hand (we used a separate button, as pressinga button on the wand could affect its stability).Experimental software ran on a Core 2 Quad PC runningWindows XP. Software was built on the .NET platform andused WPF for presentation. Both image and input were re-freshed at a rate well above interactive rate (approx 50Hzrefresh rates for display and input). Task The horizontal task follows the ISO 9241-9 one-directiontapping test recommendation[8]. Participants had to alter-nately move the cursor onto each of the target positions thatcomposed a path, and click within its boundaries. The tar-gets were vertical bands that covered the whole height of the display (seeFigure 5andFigure 6). Targeting tasks varied in the width of the targets (100, 200and 400px  –  7.1, 14.2 and 28.5cm), the distance betweentarget centers (1024, 2048 and 3072px  –  73, 146 and219cm), the position of the targets along the screen, and thedirection (left-right or right-left).Figure 6shows a diagramwith the four different paths, which multiplied by three dif-ferent widths and two directions result in 24 distinct target-ing tasks. Visual feedback of errors was provided in theform of color changes of the target. Participants Twelve participants recruited from a local university (4female, 8 male; 24 to 36 years old) took part in the study for$15 remuneration. All participants were right-handed. Procedure and Design After signing a consent form each participant providedsome basic demographic information, was tested for eyedominance (to determine the dominant eye for the image-plane technique), and received instruction in the four raypointing techniques. Figure 6. Experimental setup and horizontal paths 36cm76cm    7   3   c   m    Path 1Path 3Path 4Path 2   Figure 5. Horizontal task with two distant targets as seen frombehind the participant (approximation).Figure 4. Parallax causes that α ≠ α’.   αα '  Technique order was counterbalanced across subjects usinga random Latin square design. Participants underwent ablock of training for each technique (24 individual trials pertechnique involving all distances, positions and targetwidths), and then, in the same order, two separate blocks of actual trials for each technique with three repetitions perindividual task. Tasks were presented in order of increasingdistance between targets and decreasing target width. After the end of the each technique‟s trials of the second block, the participants were asked to rate the perceived workloadthrough a NASA TLX questionnaire[17].At the end of the experiment, participants were asked torank the techniques according to speed, accuracy, physicaleffort, and general preference. The entire experimental pro-cedure took approximately 1hour. Measures and Analysis Methodology For each trial we measured completion time, location of thecursor during the click, and whether it missed the target(error). We designed the experiment and the analysis toconform to the ISO 9241-9 recommendations[8]as well as the Fitts‟s study guidelines provided in[ 24].As Soukoreff and MacKenzie recommend, we planned er-ror and completion time comparisons as well as throughputcomparisons. This requires the calculation of the index of d ifficulty of each task according to Fitts‟ s law:D is the distance between targets and W the width of tar-gets. However, in a very early stage of the research we rea-lized that targeting tasks have different difficulties depend-ing on their location on the display and the direction of tar-geting. ConsiderFigure 5:at the very least, targeting intothe farthest region of the display (a distant target) should beharder than targeting onto the near target. Following rea-soning parallel to Kopper and colleagues[11], we antic- ipated that the standard Fitts‟s model would not capture targeting time differences that can be derived from thegeometrical relationships between the person, the display,and the target. Therefore we performed two regressions onth e data, one with the standard (linear) version of Fitts‟s index of difficulty (  ID Linear ), and one with a variant of theformula that substitutes D and W for the subtended anglesof D and W from the location of the user (seeFigure 7):The subtended angles are calculated through standard trigo-nometric procedures with the generic formula:  x 1 and  x 2   correspond to the horizontal coordinates of theextreme points of the linear distance whose angle we arecalculating (seeFigure 8). In our experimental setup,  ID Li-near and  ID Angular calculations proved substantially differentfrom each other because of the large size of the display andthe position of the user.Figure 9plots the  ID Linear of alltasks against their  ID Angular .  If, as we hypothesize,  ID Angular   predicts performance signifi-cantly better than  ID Linear across participants, it would makesense to use this instead to calculate throughput. In eithercase, for the throughput calculation we apply the effectivewidth corrections as argued in[24]. The calculation of theangle was done using a point 73cm in the direction perpen-dicular to the top right modular display, which approx-imates the position of the head of the user. Results We begin with our analysis of fit of the linear and angularmodels, follow by the performance analysis, and end with asummary of the subjective measures results. We performedanalysis on throughput, time and error for all tasks. Due tospace restrictions, we omit reporting those analyses that areredundant. Analysis of fit  We did a per-participant, regression analysis of trial com-pletion time for each technique. Using  ID Linear as a predictor Figure 9. Relationships between linear IDs and angular IDs. 12345678123456    A   n   g   u    l   a   r   I   D Linear IDTarget Distance (D)Long (Path4)Target Width (W)Large (28.5 cm)Medium (14.1 cm)Small (7.1 cm)Medium (Path3)Short (Path1 & 2)   Figure 8. Calculation of subtended angle between  x 1 and  x 2   d p α X 1 X 2   Figure 7. Geometrical relationships between D, W, δ and ω.   δω DW
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!