vecims 2009 - international conference on virtual

6
Numerical Input Techniques for Immersive Virtual Environments Georgios Lepouras Department of Computer Science and Technology University of Peloponnese Tripolis, Arcadia, Greece [email protected] Abstract—So far, in the literature, there exist only a few techniques for text input and none specifically for numeric input in immersive virtual environments. Initially, we evaluated two techniques for numerical input, one employing whole hand gloves gestures and a second employing Pinch gloves. While the first technique was proper only for small numbers and for limited use, the second technique performed well in all cases. The Pinch glove technique was revised to allow for hand tracking and re- evaluated with a larger group of users. Statistical analysis showed that the technique is promising and input rate is comparable to other input techniques. Keywords: virtual reality; numerical input; immersive virtual environments I. INTRODUCTION Numeric input is a sub-case of alphanumeric input where a user wishes to communicate a numerical value to the computer. Although in windows-based interaction this is carried out easily with the keyboard, in virtual reality environments this is not necessarily the case. If a keyboard is not present other means have to be devised to allow keying-in values. So far, only a few examples of interfaces in virtual reality systems exist which display some method of alphanumeric input. It could be argued that this is due to the nature of most virtual reality applications. Since in such applications most interaction is usually limited to navigation in the virtual environment and object manipulation, no need exists for methods that support alphanumeric input. However, Bowman et al. [3] provide a number of possible scenarios of use, that demonstrate the potential for alphanumeric input, such as the need for entering design annotations, filenames, labeling objects, precise object manipulation, parameter setting, communication between users and mark-up. Of these scenarios, some require only numeric input. Until now, input methods proposed cater for both alphabetic and numeric characters. This on the one hand has the advantage of being more generic and broadly applicable, but on the other hand a more focused input technique can be better suited for cases where only numeric input is needed. To this end, we describe two gesture-based techniques that support input of numbers in immersive virtual reality environments. II. EXISTING APPROACHES Of the approaches that can be employed for alphanumeric input in virtual environments some were specifically designed for virtual reality environments while others were borrowed from cases where typical desktop input devices cannot be employed. Depending on the device and the input channel, the methods proposed so far can be categorized in those which employ data gloves and gestures [5] [8], [12], [2] those which use a tablet and pen to allow handwritten input [16], those which use some type of keyboard [5] and those which are speech-based [10], [14]. Bowman et al. [4] compared four techniques: a pinch keyboard, a pen & tablet keyboard, a chord keyboard, and a speech-based approach. The results showed that while the speech-based technique was the fastest it also produced more errors than the others. The pen and tablet keyboard produced the least number of errors. However, it also produced high levels of arm strain. The Pinch keyboard was characterized by users as a natural technique, but its performance was not at the level of the other two techniques. Overall the experiment showed that none of the techniques tested was clearly the best for text input. These results further support the notion of designing a more focused numeric input technique. To this end, a gesture-based technique for an immersive virtual reality environment may have some advantages over tablets, keyboards and speech- based techniques. Speech-based techniques on the one hand can be difficult to implement, the vocabulary is usually limited, users are often obliged to speak slowly, while speaking to a computer may feel odd to some users [4]. Device-dependent techniques on the other hand force the user to hold a special device. Especially for immersive environments the use of an input device may force the user to look at the device itself to locate buttons and distract her from the task, an interaction that may feel uncomfortable. Furthermore, if both hands need to be free, a device dependent technique cannot be applied. In contrast, a gesture-based technique leaves the user’s hands free and if a suitable metaphor is selected, a gesture-based interaction can be intuitive, easy to learn and apply. Probably the main disadvantage for a gesture-based technique is the universality of its application. Depending on the culture the same gesture may have different connotations. Sign languages which employ a complex spatial grammar also differ from country to country. Even if one restricts the gestures to numbers there are still variations between cultures, although for numbers one to five the differences are minor. For example, in Germany one typically refers to number one by showing the thumb, while in England it is more likely VECIMS 2009 - International Conference on Virtual Environments, Human-Computer Interfaces and Measurements Systems Hong Kong, China May 11-13, 2009 978-1-4244-3809-9/09/$25.00 ©2009 IEEE 240

Upload: others

Post on 03-Nov-2021

1 views

Category:

Documents


0 download

TRANSCRIPT

Numerical Input Techniques for Immersive Virtual Environments

Georgios Lepouras Department of Computer Science and Technology

University of Peloponnese Tripolis, Arcadia, Greece

[email protected]

Abstract—So far, in the literature, there exist only a few techniques for text input and none specifically for numeric input in immersive virtual environments. Initially, we evaluated two techniques for numerical input, one employing whole hand gloves gestures and a second employing Pinch gloves. While the first technique was proper only for small numbers and for limited use, the second technique performed well in all cases. The Pinch glove technique was revised to allow for hand tracking and re-evaluated with a larger group of users. Statistical analysis showed that the technique is promising and input rate is comparable to other input techniques.

Keywords: virtual reality; numerical input; immersive virtual environments

I. INTRODUCTION Numeric input is a sub-case of alphanumeric input where a

user wishes to communicate a numerical value to the computer. Although in windows-based interaction this is carried out easily with the keyboard, in virtual reality environments this is not necessarily the case. If a keyboard is not present other means have to be devised to allow keying-in values. So far, only a few examples of interfaces in virtual reality systems exist which display some method of alphanumeric input. It could be argued that this is due to the nature of most virtual reality applications. Since in such applications most interaction is usually limited to navigation in the virtual environment and object manipulation, no need exists for methods that support alphanumeric input. However, Bowman et al. [3] provide a number of possible scenarios of use, that demonstrate the potential for alphanumeric input, such as the need for entering design annotations, filenames, labeling objects, precise object manipulation, parameter setting, communication between users and mark-up. Of these scenarios, some require only numeric input. Until now, input methods proposed cater for both alphabetic and numeric characters. This on the one hand has the advantage of being more generic and broadly applicable, but on the other hand a more focused input technique can be better suited for cases where only numeric input is needed. To this end, we describe two gesture-based techniques that support input of numbers in immersive virtual reality environments.

II. EXISTING APPROACHES Of the approaches that can be employed for alphanumeric

input in virtual environments some were specifically designed

for virtual reality environments while others were borrowed from cases where typical desktop input devices cannot be employed. Depending on the device and the input channel, the methods proposed so far can be categorized in those which employ data gloves and gestures [5] [8], [12], [2] those which use a tablet and pen to allow handwritten input [16], those which use some type of keyboard [5] and those which are speech-based [10], [14]. Bowman et al. [4] compared four techniques: a pinch keyboard, a pen & tablet keyboard, a chord keyboard, and a speech-based approach. The results showed that while the speech-based technique was the fastest it also produced more errors than the others. The pen and tablet keyboard produced the least number of errors. However, it also produced high levels of arm strain. The Pinch keyboard was characterized by users as a natural technique, but its performance was not at the level of the other two techniques. Overall the experiment showed that none of the techniques tested was clearly the best for text input.

These results further support the notion of designing a more focused numeric input technique. To this end, a gesture-based technique for an immersive virtual reality environment may have some advantages over tablets, keyboards and speech-based techniques. Speech-based techniques on the one hand can be difficult to implement, the vocabulary is usually limited, users are often obliged to speak slowly, while speaking to a computer may feel odd to some users [4]. Device-dependent techniques on the other hand force the user to hold a special device. Especially for immersive environments the use of an input device may force the user to look at the device itself to locate buttons and distract her from the task, an interaction that may feel uncomfortable. Furthermore, if both hands need to be free, a device dependent technique cannot be applied. In contrast, a gesture-based technique leaves the user’s hands free and if a suitable metaphor is selected, a gesture-based interaction can be intuitive, easy to learn and apply. Probably the main disadvantage for a gesture-based technique is the universality of its application. Depending on the culture the same gesture may have different connotations. Sign languages which employ a complex spatial grammar also differ from country to country. Even if one restricts the gestures to numbers there are still variations between cultures, although for numbers one to five the differences are minor.

For example, in Germany one typically refers to number one by showing the thumb, while in England it is more likely

VECIMS 2009 - International Conference on Virtual Environments, Human-Computer Interfaces and Measurements Systems Hong Kong, China May 11-13, 2009

978-1-4244-3809-9/09/$25.00 ©2009 IEEE 240

to show the index finger [15]. In China there exists a special set of gestures for denoting numbers [6], although there exist differences in the gestures used in Taiwan, Hong Kong, and Mainland China. Furthermore, possible gestures are limited by the device used to recognize them. To this end, we have opted for gestures that were intuitive and easy to memorize and for the employment of affordable devices. The techniques proposed, address the problem of character input for a character set of limited size. To this end, the solutions can also be applied to other small-alphabet domains such as numbered menu items or simple system control commands.

III. PROPOSED TECHNIQUES

A. Design Background A common device for recognizing hand gestures in virtual

environments is a pair of virtual reality gloves. Virtual reality gloves can be light and flexible, rendering them comfortable for extended use and are being produced for long enough to offer a stable and quite affordable solution for gesture recognition.

For implementing numeric input two types of virtual reality gloves were considered. The first type measures bending and abduction of the fingers, with Immersion CybergloveTM and Fifth Dimension DataGloveTM being frequent examples. The second type reports contact between the tips of the fingers and is commercially available by Fakespace. Fakespace’s Pinch GlovesTM although a simpler device has the advantage of not requiring calibration prior to use as is the case with whole hand gloves which report continuous joint angle data.

The differences between the functionality of the two types of gloves lead to the design of two different techniques for numeric input. A pair of Fifth Dimension 14 Ultra virtual reality gloves was selected as a representative of the first type of virtual reality gloves. It has to be noted that these gloves provide bending and abduction data for all fingers (14 sensors) but no posture data for the whole hand. Although a tracker can be added to acquire hand orientation data, use of a simple pair of gloves was thought to be a more affordable albeit limited solution. For the implementation of the second input technique a pair of Pinch Gloves was employed. One can note that the first technique employs gestural input, while the second single touches between fingers.

B. 5DT Gloves Numeric Input Technique 1) Initial considerations

In everyday life we often use hands to indicate numbers between one and ten. A straightforward technique for number input would be to key-in a digit at a time by employing a finger extension. For example, in order for the user to input number ‘2356’, she could consecutively extend two, three, five and six fingers. However, this simplistic approach raises some issues. The system has to be able to discern between successive digit gestures. For example, if the user extends two fingers just to extend a third some milliseconds later, or if the user wants to enter twice the same digit, the system may misinterpret the gesture as a ‘23’ in the first case or only a single digit in the second case. Either a time limit or an ‘accept-digit’ gesture has

to be set. Additionally, in some cases the user has to be able to enter real numbers, including negative numbers and numbers with decimals. The user should also be able to gesture the initiation and completion of the input procedure. Finally, any input technique has to allow for digit correction. The user should be able to gesture the deletion of wrongly keyed-in digits. For each of the issues raised alternative designs exist. We will next outline the possible variations.

2) Technique design Gesture – digit correspondence. Probably the most

prominent solution would be to start by extending the thumb as a gesture to denote the digit ‘one’ and consecutively extend the rest of the fingers to denote other digits, with the extension of all fingers denoting zero instead of ten. For example, extending thumb and index would correspond to two, while extending all fingers of the one hand plus the thumb of the other hand would correspond to six. Gestures like extending the index of the one hand and the thumb of the other will not correspond to any digit and could be assigned to other tasks such as navigation. This approach does not require the definition of the dominant hand, allowing the user to start with whichever hand she feels more comfortable with. Two other variations to this technique were considered. The first assigned a digit to each finger. Starting with the thumb of the dominant hand the user extends one finger at a time to denote a digit. Therefore, the thumb of the dominant hand would signify 1, the index of the dominant hand 2, the thumb of the non-dominant hand 6, etc. This design suffered from two possible drawbacks. The first is ergonomic in the sense that it is not very comfortable to extent any finger keeping the rest closed in a fist (for example try to extent the ring finger keeping the rest in a fist). The second drawback is that the system has to be informed of the user’s dominant hand, requiring some initialization prior to first use. Another variation considered was to extend any combination of fingers to denote a digit. In such a case, extending the thumb and the index would be equivalent to extending both thumbs and would correspond to two. Although this approach offered increased freedom to the user, it also limited the set of available gestures for other tasks. Furthermore, initial testing with a small group of users revealed that the enhanced sense of freedom rather complicated some users’ interaction.

Accept digit. A problem identified by Mapes and Moshell [13] is ‘the fact that natural gesturing involves a series of transitions from gesture to gesture essentially creating a continuum of gesturing’. For the system to recognize that the gesture is completed and accept this gesture as input a time limit can be set. For example, if the gesture remains stable for a second then the system may try to interpret the gesture. This has a potential drawback of restricting the input rate and tiring the user who will have to keep her fingers still. An alternative would be to use a gesture as an indication of a ‘neutral’ state. Closing both fists was selected to be the accept digit gesture that has to be carried out after each input gesture (including deletion).

Other gestures. Gestures to denote fractions (comma or dot sign), minus sign, deletion of last digit and beginning/ending of the input procedure were also devised. The number of these gestures is small enough to help memorization. Minus sign is denoted by extending both index fingers, comma delimiter by

241

extending index and middle fingers and delete last digit by extending both pinkies. Signaling the beginning and the completion of the input procedure can be carried out by extending both thumb fingers and keeping the rest of the fingers closed, a gesture interpreted as ‘two-thumbs-up’ or ‘okay’.

3) Usage scenario A simple scenario illustrating the technique is the

following. If we assume that the user wants to enter number 84, she would initiate the input procedure by extending two thumbs. The system would respond by displaying “ready” and the user would use all fingers of one hand plus three fingers of the second to denote 8. The system would display 8 and if the user closes both fists the number is accepted. The user would then extend 4 fingers of one hand, keeping the rest closed and by closing both fists the system would accept 4 as the next digit. Finally, the user would extend both thumbs to terminate the input procedure.

C. Pinch Gloves Numeric Input Technique Since pinch gloves report only contact between fingers

(single touch) they can be employed to recognize postures instead of gestures. To this end, the technique devised for whole hand data gloves cannot be directly applied. Instead we propose a technique similar to the one used by children when they learn to count [9], [7]. Using the index of the dominant hand, they count the fingers of the non-dominant hand (one to five), and then they use (if the number is greater than five) the index of the non-dominant hand to count (six to ten) the fingers of the dominant hand. This technique cannot be directly implemented as it has an inherent ambiguity: touch between the indexes of the two hands may denote a 2 or a 7. To overcome this limitation/problem the method proposed uses the index of the dominant hand to touch fingers denoting numbers one to five and the middle finger to denote numbers six to zero.

Due to the discrete nature of the pinch gloves output there is no need for a special ‘accept digit’ gesture, simplifying thus the number of gestures that have to be memorized. To minimize the set of gestures, minus sign and fraction sign can be denoted by the same gesture, by touching the two pinky fingers. If this is the first digit then it denotes a minus sign otherwise it denotes the fraction delimiter. The backspace gesture was assigned to both thumbs touching since it is a gesture that is not easy to perform accidentally. Finally, start and completion of the input procedure is denoted by thumb and index touch, a gesture associated with ‘Okay’ sign in many cultures. This gesture also signifies the dominant hand, alleviating the need for additional initialization gestures.

1) Usage scenario A simple scenario illustrating the technique is the

following. Assuming the user wants to enter number -3.7. The user would start by touching the thumb and the index of the dominant hand to initiate the input procedure and denote the dominant hand. The system would respond by displaying “ready” and the user would touch both pinkies to denote minus sign, then touch with the index finger of the dominant hand the middle finger of the other hand to denote 3. The user would then touch both pinkies to denote the fraction sign and touch

the middle of the dominant hand to the index finger of the other hand to denote 7. Finally the user would touch the index and the thumb of the dominant hand to denote the termination of the input procedure.

IV. INITIAL EVALUATION In order to identify strengths and weaknesses of both

techniques a preliminary experiment took place. Five users, four males and one female, with ages ranging from 22 to 42 participated. All the users had no prior experience with the gloves. The setup included a PC with the two pairs of 5DT gloves attached to it, eMagin 3DVISOR goggles and a program running in Vizard Toolkit using VRPN [17] to connect to Pinch Gloves. The program displayed 12 numbers, one at a time, which users had to key-in. Users inputted all numbers first with 5DT gloves (Figure 1) and then performed the same operation with the Pinch gloves.

Figure 1. 5DT Gloves number input technique

The program recorded the time required for inputting each number, the number of corrections made as well as whether the user managed to input it correctly. The set of numbers included positive and negative integers and real numbers and was created to contain numbers with digits uniformly distributed in the range of 1-5 and 6-0, with 3 to 9 characters length and mean length of 4 characters. The program also displayed the gesture recognized. The first three numbers were considered to be the training set and the results from keying-in the rest nine numbers were used for statistical purposes. Prior to the experiment each user was briefed in the way each technique worked. After the experiment users were interviewed to record their comments.

From this initial evaluation some preliminary observations can be drawn. With the first technique a mean value of 7.2 chars/min was achieved, while the second achieved a mean value of 18.8 chars/min. Most users did not have a problem completing the tasks with the second technique. Of the 60 numbers assigned, they managed to input correctly 58, while the mean error rate during key-in was 11%. For the first technique the results were not that good. Of the 60 numbers assigned they managed to input correctly 55 numbers with a

242

mean error rate of 27%. These findings were also reflected in the users’ responses during the interview.

On the positive side, users found techniques intuitive and thought of the gestures used to depict numbers as ‘natural’. While users found both gestures devised to denote start/ending of the input procedure easy to memorize, the same did not happen for other gestures. To this end, users found the second technique easier to learn, as it required only one extra gesture to remember, that of fraction/minus sign.

Users found the first technique tiring for prolonged use and for non-positive real numbers. This can be derived by the evaluation results. While input rate was around 11 chars/min for positive integers, rate dropped around 5.5 chars/min for real numbers and negative integers. Also after repeated input, error rate increased with a consequence of most errors occurring during keying-in the last numbers. Users also remarked on the need for calibrating whole data gloves prior to use. Almost all of the users faced problems with the data gloves auto-calibration routine.

The data glove auto-calibration routine works by comparing the raw values read during every update cycle from the gloves’ sensors to the current minimum and maximum values. If the current minimum and maximum values are exceeded, they are overwritten. After inputing the first six or seven numbers the gloves seemed to read a very small or very big value from a finger’s sensor, altering thus the value corresponding to closed or open finger. As a consequence, while the actual finger was closed (or open) the virtual finger would appear as semi-closed (or semi-opened). Once this happened, users had to struggle to perform the gesture required.

In regard to the second technique, the experiment results coincide with the users’ opinion. Users found it easier to perform gestures with the pinch gloves, less tiring as it required a single touch between fingers and more efficient as they did not have to perform an accept gesture to input a digit. Two were the major problems faced by users during their interaction. The first was that they did not have visual feedback on their hand movements. This was due to the fact that the pinch gloves were not being tracked and there was no point of displaying static hands. The second point made by users was that pinch gloves did not fit well. In contrast to 5DT data gloves, which are made to fit differently-sized hands, Pinch gloves are available in different sizes and unfortunately there was only one size available for the experiment and it did not fit well all participants.

A minor comment made by users for both techniques was that of the training. Since none of the users had prior experience with data or pinch gloves, users found the training session too fast and time-constrained to allow them to get familiar enough with glove usage and leave time for gesture memorization.

Based on the results of the initial evaluation, the second technique was re-implemented to allow for the real-time tracking of the gloves’ position and orientation and offer visual feedback to the users, in the form of two static (since the Pinch gloves do not provide finger bending or abduction data) 3D hands (Figure 2). A Polhemus Patriot system with a pair of

trackers was used; one for each Pinch glove. Following the implementation a second experiment was set up to evaluate the revised technique.

V. PINCH GLOVE TECHNIQUE EVALUATION For this evaluation a total of 16 users participated, 12 males

and 4 females (different from the first set of subjects). Twelve of the subjects were undergraduate students and the rest were postgraduate. None of the users had any experience with virtual reality gloves. Each user had a small training session with the Pinch gloves to get accustomed to the technique’s gestures. Time for character input during the training session was increased. In this setup users had to input 12 numbers to get accustomed to the revised technique (instead of 3 numbers during the initial evaluation). During the training session both time and errors were recorded to offer an assessment for the importance of learning in user’s performance. Apart from these changes the experiment setup remained the same as in the initial evaluation. Data recorded included input time and errors per number and once all users performed the experiment a statistical analysis was carried out on the evaluation data to reveal possible trends.

Figure 2. Pinch gloves revised technique

Overall, the mean time for character input was 2.4 seconds per character (spc) or 25 characters per minute (cpm). This is comparable to the Pinch Keyboard technique (31.69 cpm) and chord keyboard (21.13 cpm) as found by Bowman et al [4].

There was a statistical difference between input rate during training and input rate after training (p < 0.0001). Mean time was in the first case 2.93 spc or 20.48 cpm and in the second case 2.4 spc or 25 cpm signifying that learning is an important factor.

243

TABLE I. INPUT TIME (SECS) PER DIGIT DURING AND AFTER TRAINING

Time to input digit During Training After Training Mean: 2.925 2.398 # of points: 192 192 Std deviation: 2.098 1.592 Std error: 0.1518 0.1155 Minimum: 0.7820 0.3008 Maximum: 15.254 14.198 Median: 2.305 2.019 Lower 95% CI: 2.628 2.172 Upper 95% CI: 3.223 2.624 Two-tailed P < 0.0001

On the other hand, error rate per number before and after training remained almost constant (0.9 numbers inputted wrong per user) denoting that training did not affect the number of errors.

Although time per character seems to decrease for longs numbers (2.78 spc for 10-digit long numbers vs. 3.02 spc for 3-digit long numbers), there was actually not a statistical difference between long numbers (10 characters length) and short numbers (3 characters long) (Mann-Whitney test, p=0.3791).

TABLE II. INPUT TIME (SECS) PER DIGIT FOR 10 AND 3 CHARS LENGTH NUMBERS

Time to input digit 10 Digit Numbers 3 Digit Numbers Mean: 2.778 3.017 # of points: 16 16 Std deviation: 1.413 3.105 Std error: 0.3649 0.7762 Minimum: 1.420 1.271 Maximum: 6.461 14.198 Median: 2.259 2.135 Lower 95% CI: 1.995 1.363 Upper 95% CI: 3.561 4.671 Two-tailed P 0.3791

However, there was statistical significance between numbers with digits in the range 1-5 and numbers with digits in the range 6-0 (Mann-Whitney test, p=0.0260). This was in accord with users’ comments stating that touch between middle finger of the dominant hand and the fingers of the non-dominant hand was not as easy as to touching the index of the dominant hand with the fingers of non-dominant hand. Some users also attributed this difference to the gloves’ size, which was much larger than their hands.

TABLE III. SUMMARY OF DATA FOR INPUT TIME (SECS) PER DIGIT

Time to input digit Digit Range 1-5 Digit Range 6-0 Mean: 1.882 2.398 # of points: 16 16 Std deviation: 0.6800 0.6215 Std error: 0.1700 0.1554 Minimum: 1.122 1.519 Maximum: 3.281 3.878 Median: 1.759 2.242 Lower 95% CI: 1.520 2.067 Upper 95% CI: 2.245 2.729 Two-tailed P 0.0260

The problem with the gloves’ size became apparent when the input rate between females and males was compared. This problem did not affect only females but some males as well who had usability problems with the Pinch gloves due to the large size of the gloves, forcing them to delete and re-input digits.

The statistical analysis revealed that male users had a much better mean time per character (2.01 spc or 29.85 cpm) resulting in a statistical difference between males-females (Mann-Whitney test, p=0.0011). It has to be noted that for two of the male users that the gloves fitted well the average input rate was less than 1.5 spc or more than 40 cpm. Similar problems have also noted in [4]: “Some subjects had trouble making contact between two fingers, forcing them to type the same character several times”.

TABLE IV. SUMMARY OF INPUT TIME PER DIGIT - FEMALES VS. MALES

Time to input digit Females Males Mean: 3.561 2.014 # of points: 4 12 Std deviation: 1.017 0.3350 Std error: 0.5086 0.09672 Minimum: 2.515 1.469 Maximum: 4.912 2.486 Median: 3.408 2.111 Lower 95% CI: 1.943 1.801 Upper 95% CI: 5.179 2.227 Two-tailed P 0.0011

VI. DISCUSSION The experiment showed that the choice of input device is at

least as crucial as the design of the input technique. In the case of 5DT gloves technique, although users found the selection of digit input gestures natural they had trouble performing the gestures, mainly due to the auto-calibration problems previously described. The “accept-digit” employed for dividing the continuum of gesturing and differentiating between gestures also affected performance. While most users managed to input the first few numbers correctly and rather fast, performance dropped significantly towards the end of the task. User fatigue and finger strain had a direct impact on users’ performance. To become efficient the first technique has to overcome the problems related to gesture recognition. Even in this instance, finger strain should be considered and interaction should be better restricted to the input of a small set of numbers.

Pinch gloves offer an alternative to data gloves that report continuous measurements of the fingers’ bending. The discrete nature of the data provided alleviates the need for an ‘accept-digit’ gesture and for initial calibration. A single touch between fingers is enough to input a digit. If the gestures are intuitive the user can easily memorize them and with some initial training perform comparably fast. Furthermore, since only touch between fingers is required, finger strain is minimized and in our experiment none of the users complained of fatigue, even though they had to input 24 numbers (12 for training and 12 for the actual experiment).

The main advantage of the Pinch gloves is also the source of their main drawback. Since no bending data is reported from

244

the gloves, the visual feedback is minimal. During the second experiment we elected to add tracking of gloves to provide a static visualization of the users’ hand. Even though some users complained that visual feedback for the Pinch gloves technique was distracting, overall results showed that there was significant improvement over the initial implementation. Probably this drawback is not that critical in a virtual environment where the user can view her hands, but if the user wears goggles or an HMD it can make a difference.

Overall when choosing an interaction technique the designer has to balance factors such as efficiency (usually measured by input rate and error rate), cognitive load, muscular strain and perceived user satisfaction. Orthogonal to these factors is the context of interaction. If the user needs to have both hands free during interaction, gesture-based techniques can be an advisable solution. In the case of character input for a character set of limited size such as numeric input the revised pinch glove technique offers a prominent choice.

VII. CONCLUSIONS AND FUTURE PLANS A conclusion that can be drawn is that research in the area

of character input should continue. One possible enhancement of the pinch glove technique will be to include capabilities for reading bending data as suggested in [11]. This will allow the use of simple touch gestures without the need for recognizing continuous gestures and at the same time will provide accurate visual feedback to users.

To this end, we plan to customize the pair of 5DT gloves with a series of electrical contacts at each fingertip. 5DT gloves will provide bending data to implement a real time visualization of fingers flexing in the virtual environment. A pair of trackers like those employed for the revised pinch glove technique will offer the necessary data for monitoring hand movement. Numeric input will be implemented as previously using touch between fingers.

Since 5DT gloves are made of stretch lycra permitting comfortable fitting for differently-sized hands, it is anticipated that customizing the 5DT gloves with touch functionalities will also alleviate the problems faced by users due to the large size of Pinch gloves.

Further research is also planned to evaluate the input technique in different situations requiring input of limited-size character sets. The first will be in the context of numbered menu items, where the user will be able to select choices in a multi-level menu by inputting the corresponding number. The second will be in the context of a mobile phone type of keypad, where the user will be able to input text in virtual environments. This novel technique will be evaluated against other input techniques, such as pen and tablet, to offer a comparative assessment of its performance and usability as an interaction method for virtual environments.

VIII. REFERENCES [1] Adamo-Villani, N., Heisler, J., Arns, L., Two gesture recognition

systems for immersive math education for the Deaf. ACM Proceedings of IMMERSCOM 2007, 10-12 October 2007, Verona, Italy.

[2] Braffort A., A gesture recognition architecture for sign language, ACM ASSETS ‘96, Vancouver, 1996, pp.102-109.

[3] Bowman A.D., Kruijff E., LaViola J. J. Jr., Poupyrev I., 3D User Interfaces – Theory and Practice, Addison-Wesley, 2005.

[4] Bowman, A.D., Rhoton J. C., Pinho, S. M., Text input techniques for immersive virtual environments: an empirical comparison, in Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 2002, pp. 2154-2158

[5] Bowman, A.D., Wingrave, C. A., Campbell, J. M., Ly, V. Q., Rhoton, C. J., Novel uses of pinch gloves for virtual environment interaction techniques, Virtual Reality, vol. 6, no.3, 2002, pp. 122–129.

[6] Chinese Number Gestures, Wikipedia, available at http://en.wikipedia.org/wiki/Chinese_number_gestures, Last accessed 30 March 2009

[7] Count me in project. 2007 parent newsletter available at http://www.carleton.ca/cmi/CMIFiles/PublicationsFiles/E-Newsletter%2007%20CMI.pdf, Last accessed 30 March 2009

[8] Fels, S. S., Hinton, E. G., Glove-TalkII: A neural network interface which maps gestures to parallel formant speech synthesizer controls, IEEE transactions on neural networks, 1998, vol. 9, no1, pp. 205-212.

[9] Geary, D.C., Brown, S.C., Cognitive addition: strategy choice and speed-of-processing differences in gifted, normal, and mathematically disabled children, Developmental Psychology, Vol. 27(3), 1991, pp. 398-406.

[10] Laviola, J.J. Jr., MSVT: A virtual reality-based multimodal scientific visualization tool, in Proceedings of the Second IASTED International Conference on Computer Graphics and Imaging, 1999, pp. 221-225.

[11] LaViola, J., and Zeleznik, R. "Flex and Pinch: A Case Study of Whole Hand Input Design for Virtual Environment Interaction", In Proceedings of the Second IASTED International Conference on Computer Graphics and Imaging, 221-225, October 1999.

[12] Lee, S., Hong, S. H., Jeon, J. W., Designing a universal keyboard using chording gloves, ACM CUU’03, November 10–11, 2003, Vancouver, British Columbia, Canada

[13] Mapes, D.P. and Moshell, J. M. A two-handed interface for object manipulation in virtual environments. Presence: Teleoperators and Virtual Environments, 4, 4 1995, pp. 403-416.

[14] Muller, J., Krapichler, C., Lam Son Nguyen, Hans Englmeier, K., Lang, M., Speech interaction in virtual reality, in Proceedings of the ICASSP '98, Vol.6, 1998, pp. 3757-3760.

[15] Pika, S., Nicoladis, E., Marentette, P. A cross-cultural study on the use of gestures: Evidence for cross-linguistic transfer? Bilingualism: Language and Cognition, 2006, 9, 319–327.

[16] Poupyrev, I., Tomokazu, N., Weghorst, S., Virtual notepad: handwriting in immersive VR, in Proceedings of the Virtual Reality Annual International Symposium, Atlanta, 1998, pp. 126-132.

[17] Taylor, R. M. II, Hudson, T. C., Seeger, A., Weber, H., Juliano, J., Helser T. A., VRPN: A device-independent, network-transparent VR peripheral system, Proceedings of the ACM VRST 2001. Banff Centre, Canada, November 15-17, 2001.

245