Crossan, A. and Murray-Smith, R. (2004) Variability in wrist-tilt accelerometer bad gesture interfaces. Mobile Human-Computer Interaction – MobileHCI 2004: 6th International Symposium, Glasgow, UK, September 13 - 16, 2004. 3160:pp. 144-155.
eprints.gla.ac.uk/2952/
Glasgow ePrints Service
无为而无不为长城的历史背景eprints.gla.ac.uk死亡行军
Variability in Wrist-Tilt Accelerometer Bad Gesture
Interfaces
Andrew Crossan,1 Roderick Murray-Smith1,2
1 Hamilton Institute, National University of Ireland, Maynooth, Co. Kildare Ireland.
www.hamilton.ie/andy
自己的明天2 Department of Computing Science, University of Glasgow, Glasgow, Scotland.
rod@dcs.gla.ac.uk
www.dcs.gla.ac.uk/~rod
Abstract. In this paper we describe a study that examines human performance
in a tilt control targeting task on a PDA. A three-degree of freedom acceler-
ometer attached to the ba of the PDA allows urs to navigate to the targets
by tilting their wrist in different directions. Post hoc analysis of performance
data has been ud to classify the ea of targeting and variability of movement
in the different directions. The results show that there is an increa in variabil-
ity of motions upwards from the centre, compared to downwards motions. Also
太平洋巨型章鱼
the variability in the x axis component of the motion was greater than that in
the y axis. This information can be ud to guide designers as to the ea of
various relative motions, and can be ud to reshape the dynamics of the inter-
action to make each direction equally easy to achieve.
1 Introduction
Mobile devices are now widely ud for a variety of everyday tasks. However, due to the requireme
nt for a small screen, interacting with the devices often proves to be difficult. On-screen buttons are generally cloly grouped together making interac-tions slow and error prone. This is particularly the ca in a mobile context where the ur’s visual attention may be required elwhere.
Generally, interaction with the devices has taken the form of discrete messages pasd between the ur and device. The ur will click a button or lect a menu item, and the device will supply feedback. This method can be slow and frustrating particularly in situations requiring many button clicks such as typing with an on-screen keyboard.
孩子胆小
The development of new interaction techniques and nsors provide more oppor-tunity for a more continuous form of interaction, allowing clod loop interaction between device and the ur’s motions. In this instance, all of the ur’s movements affect the interpretation of the interaction and the device can continually change the feedback supplied to the ur accordingly. Gesture input is one form of continuous interaction that has been underud in interaction with current systems. Text entry is the one major exception to this where gesturing with a stylus is often ud for input-
ting text to a PDA. In this ca, it is ud to provide a quick, more natural alternative to a screen-bad keyboard where the keys may be required to be small and are tightly packed together leading
to high error rates. Pirhonen, Brewster and Holguin [6] demonstrate an example of gesturing as an input technique for controlling a PDA bad MP3 player. The interactions are designed to be intuitive for the task per-formed. Pirhonen, Brewster & Holguin were able to demonstrate significant usability benefits with the gesture interface over the standard interface, with urs indicating that the gesture system required a lower workload to perform the task.
Recent studies have examined the possibility of using accelerometers attached to a mobile device to provide input. Advantages over most stylus bad gesture systems are that they offer the possibility of one handed, screen free gesture control. They are often suggested as uful for continually monitoring background acceleration and providing context information for the current task. The components required for inertial input are also cheap to manufacture. (ca. $2 a device for mass production). Accelerometers allow a ur to input data and commands by tilting the device. Hinkley et al.[2] prent a study that demonstrates a tilt-bad gesture system for scrolling and automatic screen orientation of a PDA. Through ur testing, they were able to provide a system that performed screen orientation and scrolling in a manner that was uful and predictable to the ur. This study demonstrates the potential for tilt-bad gestures to provide a fast, natural method for interaction.
Rekimoto [7] explores the possibility of using tilt input to navigate menus and scroll large documents and maps. The prototype system described allowed urs to lect items in pie menus although no formal evaluation was carried out. Williamson and Murray-Smith [9] have developed the Hex system for inputting text on a PDA with accelerometer. This system allows the ur to lect letters by tilting the PDA to navigate a cursor through a ries of tiled hexagons. Through u of a language model, they were able to adjust the feedback given to the ur such that probable quences of characters were easier to perform than non-probable -quences. TiltType prented by Partridge et al. [5] is similarly a tilt bad text entry method where characters are lected by a combination of button clicks and the orien-tation of the device. The inertial control allows TiltType to be ud on devices with extremely small screens such as a watch.
2 Targeting Tasks
There is a large body of literature studying targeting tasks using many different input devices. Most common are Fitts’ Law bad studies where urs are required to con-tinuously move between two targets (an overview can be found in [3]). Timing and error rates can be gathered for different target widths and parations allowing the experimenter to calculate the comparative difficulty of the task. Most studies work with univariate targets by tting narrow target widths while allowing effectively inf
inite target heights. Accot and Zhai [1] describe a study that extends Fitts Law to take account of two-dimensional targets. Their experiment was ud to lect a model that provides the Fitts’ Law index of difficulty for two-dimensional targeting.
MacKenzie et al. [4] describe methods that are bad on the variability in move-ment rather than error rates. They suggest task metrics suitable for measuring move-ment variability including slip off errors, mean distance from the task axis, movement variability perpendicular to the task axis, and orthogonal direction changes.
This paper is concerned with gesturing using wrist tilt motions. With all gesturing systems, there will be a degree of variability in the gesture, and therefore uncertainty about the gesture performed. This study examines the variability in movement for short gestures in eight directions. The gestures require urs to move a cursor be-tween a ries of pairs of points by tilting their wrists. The study hoped to determine areas of difficulties at the limits of comfortable movement in different tilt directions. Both error rate and variability metrics are considered. Speed and accuracy of target-ing in different directions is also examined.
3 Experimental Method
3.1 Equipment
The experiment was carried out with an HP 5450 PDA with the Xns P3C 3 degree of freedom linear acceleration nsor attached to the rial port (shown in Figure 1). Its effect on the balance of the device is negligible (its weight is 10.35g). The accel-erometer was ud to detect tilt magnitude around the x and y axis of the mobile de-vice, sampling at a rate of 35 samples per cond.
Figure 1. PDA with XSens accelerometer attached at the ba. The ur would move the cursor by ti
lting the device in the directions shown.
3.2 Task
The experimental environment ud is shown in Figure 2. Nine circular targets of radius 15 pixels were placed throughout the environment. One target was placed at the centre of the screen, and eight were spaced at 45-degree angles around the cir-cumference of a circle centred on the initial target such that the radius of this circle was 100 pixels. The gain on the cursor movement was t such that this distance corresponded to a tilt of approximately 48 degrees in the x direction and approxi-mately 36 degrees in the y direction. The difference in the values correspond to a scaling due to screen size such that the same tilt magnitude is required to move to each of the edges of the screen (for a screen of width 240 pixels and height 320 pix-els). Due to the different x-y cursor gains, the results ction considers comparisons made between targets in opposite directions only.销售技巧
The values provided a wide range of tilts while still allowing the ur to easily view his or her interaction on the screen. A pilot study suggested that screen contrast became an issue with larger tilts in the y direction. The cursor gain was deliberately t to a low value such that large tilts would be required to complete the task and the limits of the movement would therefore be explored.
The task given to participants was to lect the highlighted target (in Figure 2 the top centre target is shown to be highlighted). The cursor was controlled by a linear position control mechanism, mapping rotation of the device to movement of the cur-sor. The device accelerometer was calibrated such that the starting position of the device corresponded to the centre position on the screen. This calibration occurred at the start of each trial. To move the cursor in the x direction, the device was titled left or right, and to move the cursor in the y direction, the device was tilted up or down (shown in Figure 1). Distance of the cursor from the centre position was directly mapped to angle of rotation from the rest position. Therefore, double the rotation angle of the device would lead to the cursor being twice as far from the central posi-tion. Since a position control mechanism was employed, if the ur held the devices still at any orientation, then the cursor would remain still on the screen.
Urs held the device in their dominant hand and were instructed to sit in a com-fortable position with the device held such that they could easily e the screen. In practice, all participants sat with the device slightly tilted towards them and leaning forwards slightly over the device.
Selection required the ur to hover the cursor over the target for 1.5 conds. If the cursor slipped off the target before the lection was complete, the target timer was ret and the ur was again required to move onto and hover over the target for the full one and a half conds. Once successfu
l lection of a target was complete, a different target was then highlighted. The quence of targets was chon such that highlighted targets alternated between any of the outside target and the centre target. This ensured that all movement was either from the central target to an outer target, or from an outer target to the central target. This quence was chon to ensure that the path distance to the next target was always kept constant and that the angle to the next target was restricted to the eight equally spaced angles chon.
Two competing factors affected the chon target size. As the trajectory rather than the targeting was the main measurement for the task, the targets needed to be big
enough to allow easy targeting. However, to maintain similar path length between starting position and target position, the targets could not be made to be too large. A diameter of 15 pixels was eventually chon empirically. A bar at the top of the screen (shown in Figure 2) indicates the time the ur has spent over the target. When the bar reaches the right of the screen, target lection has been completed.人物素描
Figure 2. The experimental environment ud for the study. The top centre target is the highlighted target. The cursor is the smaller circle within this tar-get.
All participants took part in three experimental ssions with an hour break be-tween each for recovery. The first ssion was ud to train urs in the task. The cond and third ssions were eventually ud when analysing the movement charac-teristics of different participants. The ssions were designed to be short to minimi ur fatigue. No ssion lasted over five minutes.
3.3 Participants
Twelve participants took part in the training then the two experimental ssions. Their ages ranged from 23 to just under 40 and eleven were male. Two had previous experience with accelerometers and mobile devices, but none had experience with the cursor control mechanism described above. Ten participants were right handed and two were left handed, and all ud their dominant hand for this study. The effect of this factor is considered in the next ction.
3.4 Hand Ud To Tilt The Device
The hand ud by the participant to tilt the device is an important factor when it comes to analysing the results. It is not uniformly easy to tilt the wrist in all direc-
