+ All documents
Home > Documents > Text legibility for projected Augmented Reality on industrial workbenches

Text legibility for projected Augmented Reality on industrial workbenches

Date post: 13-May-2023
Upload: poliba
View: 0 times
Download: 0 times
Share this document with a friend
DISCLAIMER This is a post-print (i.e. final draft post-refereeing pre- copyedited) version of an article submitted for publication. The definitive publisher-authenticated version is available from the publisher. Please cite this article as: Di Donato, M., Fiorentino, M., Uva, A.E., Gattullo, M., Monno, G. Text legibility for projected Augmented Reality on industrial workbenches (2015) Computers in Industry, 70, pp. 70-78. http://www.scopus.com/inward/record.url?eid=2-s2.0- 84928473276&partnerID=40&md5=914e69cca8cbebd8a7dca0f19edb83bc DOI: 10.1016/j.compind.2015.02.008


This is a post-print (i.e. final draft post-refereeing pre-

copyedited) version of an article submitted for publication. The

definitive publisher-authenticated version is available from the


Please cite this article as:

Di Donato, M., Fiorentino, M., Uva, A.E., Gattullo, M., Monno, G.

Text legibility for projected Augmented Reality on industrial workbenches

(2015) Computers in Industry, 70, pp. 70-78.


DOI: 10.1016/j.compind.2015.02.008

Text legibility for Projected Augmented Reality on industrial workbenches Abstract

Augmented Reality is a promising technology for the product lifecycle development, but it is still not established in industrial facilities. The most relevant issues to be addressed relate to the ergonomics: avoid the discomfort of Head-Worn Displays, allow the operators to have free hands and improve data visualization. In this work we study the possibility to use projection-based Augmented Reality (projected AR), as optimal solution for technical visualization on industrial workbenches. In particular, text legibility in projected AR is difficult to optimize since it is affected by many parameters: environment conditions, text style, material and shape of the target surface. This problem is poorly addressed in literature and in the specific industrial field. We analyze the legibility of a set of colors prescribed by international standards for the industrial environments, on six widely used industrial workbenches surfaces. We compared the performance of 14 subjects using projected AR, with that using a traditional LCD monitor. We collected about 2500 measurements (times and errors) through the use of a test application, followed by qualitative interviews. The results showed that, as regards legibility, projected AR can be used in place of traditional monitors in most of the cases. Another not trivial finding is that the influence on legibility of surface irregularities (e.g. grooves, prominences) is more important than that of surface texturization. A possible limitation for the use of projected AR is given by the blue color, whose performance turned out to be lower than that of other colors with every workbench surface.


Spatial Augmented Reality; industrial applications; text legibility; visualization.


• We studied and evaluated the legibility of projected text on industrial workbenches surfaces. • Projected AR can be used effectively as an alternative to monitor for displaying technical

information. • Legibility performance is not influenced by texturization of the projection surface. • Legibility performance is reduced by tactile irregularities on the projection surface. • Projected blue text is hardly readable; solutions must be found for it.

1. Introduction

Augmented Reality (AR) is a human computer interface which improves the user perception of real world with external digital information superimposed in real time [1].

Nowadays AR is used with profit in many applications like marketing, videogames, and tourism, but it is far to be accepted by the industrial world.

However, AR can be very effective in industry and in particular in some critical phases of the product lifecycle as maintenance operations, where most of personnel time is spent retrieving technical data, task instructions and localizing parts [2]. AR can improve both preventive

maintenance by delivering the required tasks in a contextual way, but also by assisting troubleshooting with a direct access to manuals, web, documents, etc.

Previous studies have shown the potential of the use of AR in industrial applications, but also stated how there are still many issues to be addressed ([3], [4], [5]).

In the design of an optimal software interface to support technical documentation visualization, we must fulfil a major requirement: the operator must be able to use both hands to accomplish his/her tasks. Common AR applications use Head-Worn Displays (HWDs), which suffer from bad ergonomics, low resolution, excess of weight, limited/fixed focal depth [6]. Industrial operators have to wear the HWD for long sessions and for this main reason, AR is not well accepted in industry. An alternative approach to HWD is to use handheld devices like smartphones and tablets. Although this kind of AR is very easy to implement in practice due to the availability of low cost and powerful devices, it has various limitations. One of the most important is that the operator should employ one or even two hands for the visualization, thus limiting his/her ability to operate.

Considering the emerging need to have digital documents at hand in a workspace, new display paradigms must be explored. According to the authors, projected Augmented Reality (projected AR) can be an optimal solution for the visualization of both instructions and technical information directly on the industrial workbench (Figure 1). In a real working environment, the operator stands in front of the workbench and is currently assisted by instructions on monitors usually placed on their workbenches or on tool carts. Projected AR makes use of digital projectors to superimpose virtual data (text, symbols, indicators, etc.) directly on the real environment [7].

Cebulla [32] listed the main advantages of projected AR displays:

• Projectors can directly project onto the object.

• The eye of the observer does not need to switch focus between the image plane and the real environment.

• The image plane of projectors can have various shapes and might be non-planar.

• A projector can be much smaller than the image it projects.

He also considers as main disadvantages the low light-intensity and the displaying of objects in mid-air. However, he says that the former is not a problem for stationary projectors as those intended to be used in our context, while the latter situation is not expected in an industrial environment. Besides those highlighted in [32], we can list other advantages of the use of projected AR in industrial facilities:

• The user do not to wear HWDs or handle devices (displays, gloves, sensors, pens, etc..): ergonomics is improved.

• The tracking of the user is not required: user’s movements (especially of the head) do not affect visualization.

• Information can be displayed all around the object, since occlusion can be reduced with a well-designed multi-projection system: information can be shared by multiple users.

• There is not the tunnel vision effect induced by the view through an optical display (like in HWDs and handheld devices): this effect can be potentially harmful if one needs to be visually aware simultaneously of dangerous stimuli (safety of operators must be ensured) approaching from a peripheral position in space.

Figure 1: example of use of projective SAR on an industrial workbench.

However, projected AR, as all new technologies, requires some feasibility studies and optimization processes before it is introduced in the industrial environment. One of the most important issues is the correct visualization of technical information. In particular, in this work we want to study the legibility of projected text information in industrial applications. Gabbard et al. [8] consider text as one of the most fundamental elements in graphical user interfaces as opposed to icons, lines, or bitmap. In the specific industrial context, text is the basic of all technical data and is widely used to convey dimensions, special treatments, annotations, etc. In order to support complex maintenance or assembly processes, text can be also supported by 3D models, pictures and animations, but the visualization of this complex elements will be addressed in future works.

In the context of a first stage of research, we evaluated the possibility to project text directly on workbench surfaces (without the need to calibrate the scene), comparing users performance with that deriving from the use of a normal LCD monitor. As far as the authors know, there are few studies on this topic and there are no widely accepted guidelines to follow in the design of this kind of system.

The paper is organized as follows: in Section 2, we present related works on projected AR; then in Section 3, we describe the method; in Section 4, we illustrate the results achieved in the experiment, while in Section 5, we provide a detailed discussion. Finally, we present our conclusions and future works.

2. Related Works

Some studies in literature analyzed different AR displays technologies taking into account application, background, lighting, etc. [6] [9]. They clearly state that there is no one ideal display fitting all scenarios. The use of projected text information can theoretically solve some big issues in industrial applications, mainly because of leaving the user free from wearing or holding any device. Nevertheless, the literature on this topic is scarce.

Raskar and Low [11] proposed a Spatial Augmented Reality (SAR) framework for the integration between projective surfaces and input devices to integrate the digital information directly in the real scenario. They introduced new calibration and rendering techniques to create a simple procedure to illuminate effectively the surfaces. To create an optimal integration of virtual information on real elements, these techniques take into account: the position of the user, the projection parameters of the display devices and the shape of the real objects in the physical environment.

Bimber et al. [12] aims to find solutions to project directly on the paintings using a traditional video projector. With direct projection, the main issue was the perception of the projected color and intensity. This is caused by the physical color pigments that neutralize what projected. To solve this problem, the authors used a new film material, which is transparent and, at the same time, diffuses part of the light projected on it. The new film material is composed of very fine particles deposited on both sides of a polyester base, without visible artifacts. In this way, the 20% of the light striking the film is diffused, allowing a better view of projected information, while the remainder comes on the canvas.

Olwal et al. [13] present an industrial application of the projection on a numerical control machine CNC. They noted that, for direct projection, it is important to assess the environmental conditions and industrial surfaces. The authors highlight the importance of the use of a screen for applications that require a direct visual feedback of the worker. In particular, they used a holographic optical element overlaid onto the machine’s safety glass as optical combiner, which allows us to see simultaneously the real environment, as well as superimposed 3D graphics, when we look through this transparent display.

Servan et al. [14] proposed laser projection for supporting the mounting process in an aeronautical industry, where great precision is required. The mounting documentation provides information on the equipment to be used and the sequences of operations. Typically, this documentation is in paper format, so the Airbus Military developed a system, called SAMBA-Laser, which retrieves from the Digital Mock Up the information needed for mounting, and projects it directly on the parts of the aircraft. This study demonstrated the effectiveness of the system in reducing the time to generate and consult the documentation.

One interesting study on text legibility with projected AR was made by Iwai et al. [10]. In this work, the authors first proposed a new legibility estimation method, with a genetic algorithm, to evaluate the ease of reading of projected characters by considering their deformation, contrast loss, and the occlusion caused by non-planarity and spatial variance of the reflectance of the projection surface. Then, they proposed a new label layout technique for projection-based augmented reality. It determines the optimal placement of each label directly projected onto an associated physical object whose surface is normally inappropriate for projection. Furthermore, they also highlight the lack of readability studies with projective techniques respect to optical see-through HWDs.

Not all the presented solutions are directly applicable to ensure legibility of text with direct projection on industrial workbenches. In fact, it is not possible to change the layout of the text label [10] because the workbench is made up of a single material and there are no other possible projection areas. One could think to project outside the workbench, for example on an ad hoc screen as made by Olwal et al. [13], but in this way we are not exploiting the advantage of having information directly on the work area. Use of special coatings [12] on the workbench surface is not a viable solution for our purpose because they, if usable, would be subject to wear and tear. Finally the number of colors available, and their cost, limit the use of laser projectors.

As final conclusion, we collected from literature the main factors which can influence the correct perception of projected information [7]:

1. keystone and distortion in oblique projection on planar displays; 2. user’s point of view; 3. distortion on non-planar target surfaces; 4. limited depth focus for displaying information on screen surfaces that are extremely curved; 5. text contrast; 6. material and texture of projection surfaces.

The first two points can be controlled operating on the setup, thus reducing their influence. In the following section, we detail how we addressed these parameters in the specific workbench scenario. The third and fourth points do not concern our setup because our workbench is planar. Finally the fifth and sixth points are the main variables of the experiment, since text contrast changes according to the color displayed (for a fixed luminance of the projector), and six different projection surfaces were tested.

3. Material and methods

For this study, we used a low-level identification task without semantics (e.g. understanding the contents/meaning of the text). We evaluated how quickly and accurately users could read text with two devices, five text colors and six different projection surfaces. Prior to conducting the study, we formulated the following hypotheses.

H 1. Different projection surfaces affect legibility performance differently;

H 2. Different text colors affect legibility performance differently;

H 3. The monitor performs better than projection surfaces regardless of the color.

These hypotheses were tested for both the response times and the error rates.

3.1. Participants Fourteen unpaid participants were recruited for the study. They were undergraduate or graduate students in technical subjects. They were 12 males and 2 females. The mean age is 30.9 with a standard deviation of 9.2 in the range from 23 to 53. We also tested every participant for color perception deficiency through Ishihara test. We also checked if their height was within the range of 1706 mm to 2162 mm calculated to ensure the right visual angle.

Figure 2: user performing the legibility test with monitor (left), and projective SAR setup (right).

3.2. Test setup We built in our laboratory a test workspace composed of three main systems: the workbench, the lighting system, and the displays.

We simulated the workbench with a normal office desk raised to a height of 900 mm (industry standard) and holding the materials used in the specific experiment or the LCD monitor. Users stand in front of the workbench at a horizontal distance of about 700 mm from the center of the projection window (Figure 2). Although the text appears slightly flattened in this position, users did not considered this influencing legibility. Anyway, this is a pejorative condition compared to the ideal case. On the workbench, outside the projection area, we placed an alphanumeric keyboard where users gave their answers during the experiment.

Figure 3: the workbench setup

The lighting conditions are critical because they affect contrast, so we controlled the lighting of the room. We used only two lamps with 5300 K temperature. Both lights are equally directed on the work plane and are provided with a diffuser to uniformly illuminate the entire work surface without light spots. On the worktop we guarantee an average illuminance of 500 lx, in according to ISO 8995-1 [15] in all testing conditions.

The projector used is a EB-1775W (Epson Co.) LCD Projector, projection technology 3LCD 0.59ʻʻ with MLA, resolution WXGA - 1280 x 800 x 60 Hz (16:10 aspect ratio), contrast 2000:1, white light output (ISO 21118 based measurements) 3000 lumens, color light output 3000 lumens, zoom ratio optical x1.2, keystone correction vertical ±30° and horizontal ±20°, connected to computer by VGA. The projector, supported by a tripod, was placed over the workbench, orthogonally and at a distance of 850 mm from the projection plane. In this position the keystone effect was minimized, and we further adjusted the projected image setting the keystone correction of the projector to -2 vertical and 0 horizontal. Furthermore, in this position there are no shadows produced by the operator, so it would also be the typical installation of a projector in a projective SAR industrial workbench (Figure 3).

The LCD monitor is a SyncMaster 920n (Samsung Co. Ltd.), LCD display, 1280 x 1024 x 60 Hz, contrast 700:1, brightness 250 lm, viewing angle (H/V) 160° / 160°. We placed the monitor on the

lamp lamp



projection surface

workbench, in front of the user, who was allowed to set the screen tilt to achieve his/her best viewing condition.

The projector and the LCD monitor were connected to a workstation, located outside the dark room. During the experiment, a supervisor guided the test and arranged the workbench according to the testing surface.

3.3. Legibility test Our legibility test follows Gabbard approach [8]. It shows the user two different text blocks: the upper is composed by three random generated strings with alternating uppercase and lowercase letters, while the lower block consists of three strings of capital letters. All the letters are displayed using a unique style we call presentation mode. The alphabet is restricted to the following letters “C, K, M, O, P, S, U, V, W, X, Z”, because these letters have graphical similarity in uppercase and lowercase, therefore this restriction uniforms the difficulty associated with the target identification. In the upper block, there is a single couple of equal uppercase and lowercase consecutive letters (i.e., target letter). The user has to identify the target letter and she/he has to count out how many times it appears in the lower block. The participants should input the result on a provided numeric keypad. The possible answers are 1, 2, 3, or 0 in case of unreadable/not found letter.

In order to better investigate the design space of styles and colors, we developed a software tool, called HMD test [16] written in C++ and Qt. The software has two functions: set the text style parameters (editor mode) and run the test (player mode). In the editing phase, the test designer can change interactively all the text style parameters (font, size, color and transparency) with a simple GUI and preview the final effect in real time. Once these parameters have been set, the designer can choose the number of repetitions for every single text visualization modality, and finally save the test configuration. In player mode, the test configurations are retrieved, shuffled randomly and then displayed on the monitor and projector.

In summary, each users executes the following tasks:

• scan meaningless random text strings;

• identify a target letter;

• count letters;

• provide a numeric response with the keyboard keypad.

3.4. Text Style Settings For our experiment, we chose a sans serif non proportional typeface: “Monospace Typewriter” freely available on web, [17]. Sans serif typeface is easier to be read on a computer display [18] and non-proportional typeface arise from preliminary tests with a proportional one, in which we realized that large letters as “M” were easier to identify than others like “V.”

We set the font height in order to have a visual angle of 30' (arc minutes), for both the devices. This value is larger than the 20’ required by the symbol recognition standard [19]. The height of the font that provides a visual angle of 30' for the projection surface is 12.4 mm, while for the LCD monitor it is 8.0 mm. The height of the font was computed assuming the average height of the Italian population (1750 mm) and a fixed (700mm) user position relative to the workbench. The visual angle is granted for user’s heights between 1706 mm and 2162 mm, therefore we selected only subjects in this range.

3.5. Colors Differently from other application fields, in industrial applications color coding is very important and often regulated by international standards (for example “ASME A13.1, 2007” for piping, “ISO 3864” for safety symbols or internal practices (e.g., 5S, a common workplace organization method [20]). Considering the ISO 3864 directives, we chose the following colors for our test:

• White: RGB (255,255,255);

• Blue: RGB (0,0,255);

• Yellow: RGB (255,255,0);

• Red: RGB (255,0,0);

• Green: RGB (0,255,0).

3.6. Projection surfaces Text was projected on planar surfaces so we did not have problems of distortion (no calibration needed) and of depth focus (all projected characters were into focus). However, worktops chosen are far from being considered ideal projector screens: they show macroscopic surface irregularities (those tactile texture that we call “3D texture”), 2D texturization (that only visually perceivable, and that we call “2D texture”) and are made of materials with different radiometric properties.

In a preliminary research we found the materials most commonly used in industrial workbenches. We selected this set for our tests:

• Embossed Steel (in the following called “Embossed”) with a marked 3D texture given by the prominences on the surface, about 1.5 mm high (a);

• Stainless Steel (“Steel”) with a 2D texture given by the “engine turning” finish of the surface (b);

• Butcher Block Maple (“Wood”) with a 2D texture given by the wood grain (c);

• Radial Runner PVC (“Rubber”) with a slight 3D texture given by the squared prominences on the surface, about 0.1 mm high (d);

• Heavy Duty PVC Corrugated Runner Vertical (“Vertical”) with a marked 3D texture given by the vertical grooves, about 1.5 mm deep and 2 mm wide (e);

• Heavy Duty PVC Corrugated Runner Horizontal (“Horizontal”) same as above rotated by 90°(f).

Readability is particularly influenced by the 3D texture. Regular patterns can influence the perception of a letter shape, generating unintended occlusions or shifting of character contours (Figure 4a,e,f). However, we quantified this effect for the projection surfaces with 3D texture and it was almost three times smaller than stroke width of the characters, so legibility was not compromised. All materials used have a similar surface finish which allow them to diffuse a good portion of incident light, so legibility was always ensured.

Figure 4: white text projected on the projection surfaces used in the experiment: Embossed (a), Steel (b), Wood

(c), Rubber (d), Vertical (e), and Horizontal (f).

3.7. Metrics The test follows a ‘within-subjects’ design: all users ran the readability test for each target material and for each device following a Latin square order [21]. Participants received adequate instructions for the experiment and before the start of each test, they executed a training trial to get used to each device and projection surface configuration. In these training trials, they accomplished the same main task. The training trial ended when either they answered correctly to two consecutive queries in a reasonable time (less than about 25 s per query, based on our previous experience), or when 15 queries were displayed. If the latter condition had been reached, the user would have been discarded from the sample; however this possibility has never occurred in the experiment.

In the readability test, each text color was randomly shown five times, so each user saw, for each configuration of projection surface and device, a total of 25 visualization queries.

At the end of each legibility test, after a 2 minutes break, each participant was asked to fill in a first questionnaire. He/she assigns first votes on a 5-point Likert scale (1=poor, 5=excellent) about the visibility of each text color shown in real time on the worktop/device, and then an overall judgment about the visibility with that worktop/device regardless color. Then, at the end of the experiment,

a b

d c

e f

each participant was asked to detect and suggest particular problems and opinions. The single trial lasted from a minimum of about 3 minutes to a maximum of about 5 minutes, for a maximum registered time of 54 minutes for the whole experiment.

The independent variables of the experiment are:

• projection surface;

• text color.

The test allowed us to measure the following dependent variables:

• Time;

• Error rates;

• User ratings.

At the end of each trial, (i.e., the execution of the designed test with a specific device on a specific projection surface), the application acquires and stores in a simple ASCII text file the following data: subject username, time and date of the trial, device, projection surface, text style including font and size, response time and result (OK/KO), displayed text strings.

4. Results

Our purpose was to evaluate the main effects of the projection surface and of the text color on readability performance. We had two types of data: the quantitative values of completion time and error rate, and the qualitative votes of subjects.

To make statistical inferences, we started to enquire whether the completion time sample followed a normal distribution. We used the Shapiro-Wilk normality test, AS R94 algorithm, on all samples. If the data set rejected the normal distribution, we used the Box-Cox transformation.



x = response variable,

λ = transformation parameter.

To determine the optimal value of lambda λopt, which allows the normalization of the samples with the Box-Cox transformation, we used the MAX Log-Likelihood Function [22]. On the transformed data, we repeated the test of normality, so we performed the outlier detection with the Tukey's outlier detection filter. To evaluate the homoscedasticity, we applied Levene test, which does not require equal dimensions for all the groups.

As regards the error rate, we used the method of “nx2 contingency tables” to do statistical inference. We used the following error rate definition:


4.1. Projection surface comparison To test the Hypothesis H1, we collected all the response time for the six projection surfaces and the monitor into seven samples. The value of λopt was -0.3934. However, after having transformed the

samples with the box-cox transformation with λopt, Horizontal and Vertical remains not normal and there was no value of λ which normalizes them. Samples were homoscedastic (F(6,2396)=0,794; p=0,574), then we used the Kruskal Wallis test to verify whether there are significant differences among the samples. The result of the test (H(6)=59,854; p<0,001) revealed that the differences among the samples times are significant.

To evaluate the relative performance among the seven samples, we used the pairwise multiple comparison with Bonferroni correction. We estimated the relative performance as:



= mean time of the best performing sample;

= mean time of the least performing sample;

The pairwise comparison tests showed a statistical difference. Analysis of the data shows that the Wood performs better than Rubber (Δ=10.6%), Vertical (Δ=14.9%), Embossed (Δ=15.8%) and Horizontal (Δ=16.7%); Monitor performs better than Vertical (Δ=8.6%), Embossed (Δ=9.6%) and Horizontal (Δ=10.6%); Steel performs better than Embossed (Δ=8.4%). In other cases there is not a significant difference.

As to error rates, there is a significant difference among all samples (χ2(6)=24.241, p<0.001). However, we observed that there is not a significant difference among Monitor, Rubber, Steel and Wood (χ2(3)=3.665, p=0.300). They have lower error rates than the other projection surfaces.

Figure 5: medians of users’ ratings about legibility on the visualization modes tested (1 poor, 5 excellent)

As to the qualitative results consequent upon the questionnaire (Figure 5), we could say that users gave good opinions on the use of Monitor, and the projective technique on Wood and Steel. This result confirm that users are familiar with monitor vision, thus confirming our choice to use it as a baseline.

4.2. Text color comparison To test the Hypothesis H2, we performed different analyses for the seven target materials, considering as independent variable in each analysis the text color. For each projection surface and Monitor, we determined the value of λopt, which normalized the major part of the samples to be compared (text colors). After having evaluated the homoscedasticity of the samples, we tested the Hypothesis H2 for each projection surface and Monitor with the proper test, as shown in Table 1.

Observing the results, we can say that for every projection surface and Monitor, differences among the groups are significant.

Table 1: Time and error analyses for text color comparison.

Time Analyses

Embossed Horizontal LCD Rubber Steel Vertical Wood

λ (Box-Cox Trans.)

-0.3142 -0.4817 -0.1834 -0.4187 -0.3370 0.0633 -0.3349

Levene Test F(4,337)=0,289; p=0,885

F(4,339)=0,276; p=0,894

F(4,341)=6,951; p<0,001

F(4,337)=0,941; p=0,440

F(4,339)=0,609; p=0,656

F(4,322)=2,889; p=0,023

F(4,344)=2,105; p=0,080

Mean Comparison


One-way ANOVA

One-way ANOVA


One-way ANOVA

One-way ANOVA

Welch ANOVA Welch


F(4,337)=8,587; p<0,001

F(4,339)=7,875; p<0,001

H(4)=13,687; p=0,008

F(4,337)=8,179; p<0,001

F(4,339)=8,715; p<0,001

F(4,153,450)=4,745; p=0,001

F(4,344)=5,944; p<0,001

Error Analyses

Chi-Squared Test

χ2(4)=17.070, p=0.002

χ2(4)=22.444, p<0.001

χ2(4)=8.358, p=0.079

χ2(4)=13.407, p=0.009

χ2(4)=11.362, p=0.023

χ2(4)=49.946, p<0.001

χ2(4)=23.082, p=0.300

To evaluate the relative performance of the five different colors in the six projection surface and Monitor, we used the post-hoc tests for the ANOVA and the pairwise comparison with Bonferroni correction for the Kruskal Wallis test. These tests showed that the blue has significantly lowest performance on all projection surface with the exception of yellow on wood material. The worsening of time performance with the blue color ranges from a minimum of Δ=16.3% (against white in Steel) to a maximum of Δ=28.4% (against yellow in Horizontal). For Monitor, we can see that the yellow performs worse than blue (Δ=23.7%) and white (Δ=24.5%).

As to the error rates, we have for all the projection surfaces a significant difference (Table 1): blue and yellow have the highest error rates in all the projection surfaces, except for wood where yellow has the highest error rates. On the contrary, for Monitor there are not significant differences among error rates.

As to the qualitative results consequent upon the questionnaire (Figure 6), we could say that users felt better with the green color, while misjudged the use of blue on almost every projection surface.

Figure 6: medians of users’ ratings about legibility of different text colors, on the visualization modes tested (1

poor, 5 excellent).

4.3. Blue text color issue Given that the previous analyses showed that the blue color has a negative impact on the legibility performance of text projected on the projection surfaces, we decided to make a further comparison among them by removing the blue color from the samples. However, we preserved the blue color for Monitor because its performance are not significantly different from that of the other colors. The new value of λopt was -0.3638. However, after having transformed the samples with the box-cox transformation with λopt, Vertical remains not normal and there was no value of λ, which normalizes it. The samples were homoscedastic (Levene’s test F(6,2010)=0,550; p=0,770), so we used the Kruskal Wallis test, which showed that there is a significant difference among the samples (H(6)=47,375; p<0,001).

The pairwise comparisons showed that the Wood performs better than Rubber (Δ=10.1%), Monitor (Δ=11.0%), Embossed (Δ=15.0%), Vertical (Δ=15.1%) and Horizontal (Δ=15.4%); Steel performs better than Vertical (Δ=15.0%). In the other cases, there is not a significant difference.

As to the error rates there is not a significant difference between the samples (χ2(6)=8.777, p=0.118).

5. Discussion

From the presented results, we can confirm hypothesis H1 (different projection surfaces affect legibility performance differently) both for time and accuracy performance. In particular, Wood and Steel are the projection surfaces with the best performance, whereas Horizontal, Vertical and Embossed have the worst performance. This result allows us to say that with projective SAR the influence of 3D texture on legibility is more important than that of 2D texture. This is an important finding since other displaying technologies, as HWDs, are influenced by 2D texture of the background [23], [24], [25]. In the case of projective SAR, this does not happen because luminance of text is much higher than that of background, so perceived text contrast [26] has always optimal values and human eye is not disturbed by background 2D texturization. On the contrary, 3D texture

produces slight distortions of the characters causing a sort of masking effect on text, similar to that theorized by Petkov and Westenberg [27], and Solomon and Pelli [28] . Furthermore, we observed that there are no significant differences between the performance of surfaces with 3D texture, so we can say that legibility is not influenced by the shape of the irregularities tested in our experiment. However this is indeed a delicate aspect to be further investigated.

Results confirm also hypothesis H2 (different text colors affect legibility performance differently) both for time and accuracy performance, except for Monitor LCD where we have no significant differences among error rates of the five text colors. In particular, we can say that with projective SAR blue text color has bad performance because of the low contrast between text and the background, since the blue is a dark color. This result is not confirmed for Monitor because text is always displayed on a black darker background. On the contrary, with projective SAR, as well as with Optical See-through HWDs, the text luminance is summed to that of the background. In this way, colors with a lower luminance, as blue, are more difficult to read. This result is in agreement with previous studies on legibility with Optical See-through HWDs [3]. This is the same reason why Gabbard did not use the pure blue in his experiments with HWDs [8], replacing it with cyan. For Monitor we expected to have no significant differences among text color, but this was confirmed only for error rates. For response times the yellow color had lower performance, but this result could be mainly due to the warm light used which cause a luminance value on the black screen of the Monitor similar to that of the yellow text. However, we did not consider this as a valuable result even because it was not confirmed for the error rates analysis.

We can partly reject hypothesis H3: Monitor performs better than projection surfaces with 3D texture (Embossed, Horizontal, Vertical). However, if we do not consider the performance of the blue text, so removing the effect of contrast, we observed that there is not a significant difference between the performance of Monitor and the performance of all the projection surfaces. Indeed, there is also a projection surface (Wood), whose performance is better than that of Monitor.

Literature on this topic is very scarce but our results confirm all the potential of projective SAR to provide operators with technical information directly on their workbenches. However, legibility was confirmed to be an issue with this kind of interface, since there were difficulties with the legibility on some projection surfaces or with the blue color. All the techniques proposed in the literature to improve the perception of projected images, cannot directly applicable to our case. Then, with our study we provided some guidelines for the choice of an optimal material of the workbench.

6. Conclusion

With this work, our goal was to make a study of text legibility for projective SAR on industrial workbenches, assessing whether a simple LCD projector is able to guarantee the same performance of a traditional LCD monitor. This study provides guidelines for the choice of an optimal material of the workbench. A user test was created to evaluate five text colors, projected onto six projection surfaces and displayed on a monitor, which was used as a baseline.

The results of the study show that materials that present a 3D texture should not be used, because their performance is worse than that of the monitor. In fact, on those surfaces, characters appear deformed by the irregularities on them, so their perception is compromised.

On the contrary, a very good projection surface is wood, which despite wood grains, has a better performance than monitor.

As regards colors, with projective SAR blue has bad performance because of the low contrast between text and the background. In our future works, we aim to solve this problem providing

methods to improve legibility of blue text. These includes the techniques of contrast enhancement via software, like outline and billboard, already tested with success on HWDs [29], [30], [31]. This is a crucial problem, because we observed that, if we do not consider the blue color, every projection surface used have similar performance of the monitor.

Another future work is the study of the interaction between the 3D texture of the projection surface and the projected text. We aim to create a legibility predictor, which will take into account the shape, the height, the spacing, and the orientation of the surface grooves, as well as the size, the spacing and the stroke of the text characters.

7. References

[1] R. T. Azuma and others, “A survey of augmented reality,” Presence, vol. 6, no. 4, pp. 355–385, 1997.

[2] A. E. Uva, S. Cristiano, M. Fiorentino, and G. Monno, “Distributed design review using tangible augmented technical drawings,” Computer-Aided Design, vol. 42, no. 5, pp. 364–372, 2010.

[3] M. Fiorentino, A. E. Uva, M. Gattullo, S. Debernardis, and G. Monno, “Augmented reality on large screen for interactive maintenance instructions,” Computers in Industry, vol. 65, no. 2, pp. 270–278, 2014.

[4] S. Henderson and S. Feiner, “Exploring the benefits of augmented reality documentation for maintenance and repair,” Visualization and Computer Graphics, IEEE Transactions on, vol. 17, no. 10, pp. 1355–1368, 2011.

[5] “ManuVAR, http://www.manuvar.eu.” [6] D. Van Krevelen and R. Poelman, “A survey of augmented reality technologies, applications

and limitations,” International Journal of Virtual Reality, vol. 9, no. 2, p. 1, 2010. [7] O. Bimber and R. Raskar, Spatial augmented reality. Peters, 2005. [8] J. L. Gabbard, J. E. S. II, D. Hix, S.-J. Kim, and G. Fitch, “Active Text Drawing Styles for

Outdoor Augmented Reality: A User-Based Study and Design Implications,” in VR, 2007, pp. 35–42.

[9] R. Azuma, Y. Baillot, R. Behringer, S. Feiner, S. Julier, and B. MacIntyre, “Recent advances in augmented reality,” Computer Graphics and Applications, IEEE, vol. 21, no. 6, pp. 34–47, 2001.

[10] D. Iwai, T. Yabiki, and K. Sato, “View Management of Projected Labels on Nonplanar and Textured Surfaces,” Visualization and Computer Graphics, IEEE Transactions on, vol. 19, no. 8, pp. 1415–1424, 2013.

[11] R. Raskar and K.-L. Low, “Interacting with spatially augmented reality,” in Proceedings of the 1st international conference on Computer graphics, virtual reality and visualisation, 2001, pp. 101–108.

[12] O. Bimber, F. Coriand, A. Kleppe, E. Bruns, S. Zollmann, and T. Langlotz, “Superimposing pictorial artwork with projected imagery,” in ACM SIGGRAPH 2006 Courses, 2006, p. 10.

[13] A. Olwal, J. Gustafsson, and C. Lindfors, “Spatial augmented reality on industrial CNC-machines,” in Electronic Imaging 2008, 2008, pp. 680409–680409.

[14] J. Servan, M. Oliva, F. Mas, and J. Rios, “Proyección Laser como soporte al proceso de montaje del ala del Airbus A400M,” 2012.

[15] W. ISO, “iso8995-1:2002 Lighting of indoor work places,” The international organization for standardization. 2002.

[16] “http://sourceforge.net/projects/hmdtexttester/.” [17] “http://www.dafont.com/monospace-typewrite.font, 2013.” [18] M. L. Bernard, B. S. Chaparro, M. M. Mills, and C. G. Halcomb, “Comparing the Effects of

Text Size and Format on the Readability of Computer-displayed Times New Roman and Arial Text,” Int. J. Hum.-Comput. Stud., vol. 59, no. 6, pp. 823–835, Dec. 2003.

[19] W. ISO, “ISO 9241-8:1997 Ergonomic Requirements for Office Work with Visual Display Terminals (VDTs) - Part 8: Requirements for Displayed Colors,” The international organization for standardization. 1997.

[20] H. Hirano, B. Talbot, and N. Bodek, 5 Pillars of the Visual Workplace: The Sourcebook for 5S Implementation. Productivity Press Tokyo, 1995.

[21] E. Williams, “Experimental designs balanced for the estimation of residual effects of treatments,” Australian Journal of Chemistry, vol. 2, no. 2, pp. 149–168, 1949.

[22] J. C. Pinheiro and D. M. Bates, “Approximations to the log-likelihood function in the nonlinear mixed-effects model,” Journal of computational and Graphical Statistics, vol. 4, no. 1, pp. 12–35, 1995.

[23] A. Leykin and M. Tuceryan, “Automatic determination of text readability over textured backgrounds for augmented reality systems,” in Proceedings of the 3rd IEEE/ACM International Symposium on Mixed and Augmented Reality, 2004, pp. 224–230.

[24] K. Tanaka, Y. Kishino, M. Miyamae, T. Terada, and S. Nishio, “An information layout method for an optical see-through head mounted display focusing on the viewability,” in Proceedings of the 7th IEEE/ACM International Symposium on Mixed and Augmented Reality, 2008, pp. 139–142.

[25] L. F. Scharff and A. J. Ahumada, “Predicting the readability of transparent text,” Journal of vision, vol. 2, no. 9, p. 7, 2002.

[26] C. Chubb, G. Sperling, and J. A. Solomon, “Texture interactions determine perceived contrast,” Proceedings of the National Academy of Sciences, vol. 86, no. 23, pp. 9631–9635, 1989.

[27] N. Petkov and M. A. Westenberg, “Suppression of contour perception by band-limited noise and its relation to nonclassical receptive field inhibition,” Biological cybernetics, vol. 88, no. 3, pp. 236–246, 2003.

[28] J. A. Solomon and D. G. Pelli, “The visual filter mediating letter identification,” Nature, vol. 369, no. 6479, pp. 395–397, 1994.

[29] M. Fiorentino, S. Debernardis, A. E. Uva, and G. Monno, “Augmented reality text style readability with see-through head-mounted displays in industrial context,” Presence: Teleoperators and Virtual Environments, vol. 22, no. 2, pp. 171–190, 2013.

[30] S. Debernardis, M. Fiorentino, M. Gattullo, G. Monno, and A. Uva, “Text readability in head-worn displays: color and style optimization in video vs. optical see-through devices,” IEEE transactions on visualization and computer graphics, 2013.

[31] J. L. Gabbard, I. J Edward Swan, and D. Hix, “The effects of text drawing styles, background textures, and natural lighting on text legibility in outdoor augmented reality,” Presence: Teleoperators and Virtual Environments, vol. 15, no. 1, pp. 16–32, 2006.

[32] A. Cebulla, “Projection-Based Augmented Reality,” 2013.