Drawing smooth contours using tangible input

De Ensiwiki
Aller à : navigation, rechercher
Project schedule.png
Titre du projet Drawing smooth contours using tangible input
Cadre Ensimag

Encadrants Céline Coutrix

Students

Introduction

Problem

Drawing curves using a computer drawing software can be at many times a challenging and time consuming task. Usually, the drawing tools allow the user to either draw a curve using a freehand lines tool or to draw a line, and then grab and move some of its points in order to shape the curve as he wants. However, when using a mouse - the traditional input method for these softwares - this movement becomes not intuitive. In real life, the user's hand would follow the curve he draws, movement that will hardly be truly mimicked by the mouse due to its indirect input nature.


A solution present in the industry is to use a touch screen as a form of providing a direct input interaction that imitates more accurately the real drawing process on a paper. There are products where the user uses its hands in a touch screen and others that offer a pen with which the user can draw over a tablet as canvas. However, having in mind also the creation of contours in a 3D space, we thought a tangible malleable input device could bring many advantages to the drawing process.


When analysing this problematic, it was found that an industry that faces this challenge on a daily basis is the car designing business. For better understanding of the field and validation of the hypothesis made, the expert on the car design domain Laura Pujades was consulted. The next subsection is built according to Ms Pujades’ inputs.

Background field: car design

In the car industry, apart from the technology, the design is one of the most important aspects because the sales depend on it. And, looking at any street, it is easy to see that curvy contours are heavily present in car designs nowadays. When designing these surfaces, it is of extreme importance to have contours and surfaces that look continuous, even though they will be built as different parts combined. Whilst drawing curvy contours is an easy task, doing it well and in a computer is challenging. Mathematics can be used to facilitate this process. However, sometimes, even if the mathematics are right, visually the results are not satisfactory - the surface does not look continuous.

Summarized, the car design process is as it follows:

  1. First the designer makes sketches of the new concept (Concept Car) on paper/computer (Photoshop, etc…);
  2. Then this concept is digitalized with the computer with some 3D program like Alias or Icem Surf. The first 3D model has basic/simple surfaces;
  3. At the same time, the engineering department prepares an Engineering Package, where all the technical inputs (COP pieces, sections, minimum radius, demolding angles, norms, etc.) needed for the new car can be found;
  4. Then, usually, this first model will be printed in clay, using a 3D machine with scale 1:1, so it can be reviewed. Nowadays this process spent a lot of time and money, but still important to visualize the final product before producing it;
  5. With this clay model, all the necessary changes will be made (by hand or printing only some areas). The model will, then, be scanned to generate the new version with 3D model. This will be done in an iterative process until the final approval of the design;
  6. So finally, the model will gain more and more precision, until it reaches a precision of 0.001 millimeters.


The problematic studied is predominantly present on tasks performed in the steps one and two. The difficulty of these tasks depend on many different variables, such as tools available, mastery of these tools, time available, level of difficulty of contours, precision needed, etc. Therefore, due to the vast nature of the drawing curvy contours problem and the limited time to develop this project, the research question focus on a small part of this problem that can be tested via an experiment:

When drawing curvy contours, what is the cross-over point in efficiency between the traditional (mouse) and a tangible input approach?

Objective

This project seeks to define when does it becomes more efficient to draw smooth contours using a tangible input device instead of a mouse in a traditional drawing software. This efficiency takes into account time to perform action and error. This research expands the target users to regular users (that usually are not experts on any drawing tool) and also to other applications of common users, such as small projects or drawings.


The possibility of applying this concept to 3D contours and even surfaces was considered. Nevertheless, for building a good enough prototype in reasonable time considering our previous knowledge, we decided to focus on 2D drawing of curves.

State of the art

Market / Industry

Nowadays, different software solutions are used to help in the design process of almost every industrial domain. Ms Pujades, the expert advisor in this project, said she uses Icem Surf to design surfaces when modeling new cars. Icem Surf is a software specialized in the design of curvy surfaces, more specifically A-class surfaces, like shown in Figure 1. It allows the user to control mathematically the surfaces using control points (position, tangency, curvature, torsion) and also to review certain characteristics, such as curvature, minimum radio, and highlights.


Figure 1 - Car being designed in Icem Surf
Figure 1 - Car being designed in Icem Surf


According to Ms Pujades, this software is really useful in her domain because the user can easily control a lot of design aspects of the surfaces. However, she mentions two of its flaws. Firstly, the program sometimes affirms that a surface is mathematically correct, but visually it is not perfect. And secondly, with Icem Surf the user can easily draw A-class surfaces, but it becomes difficult to make changes in it. Therefore, in case of needed changes, one needs to redo the surface if he wants the best result quality.

Research

In the HCI field, a variety of passive curve based input devices have been studied. They allow the exploration of the rich physical affordances of curves and, through bending and shaping it, users can easily directly manipulate 3d curves and splines. These advantages can be widen by active curve devices, that not only serve as input but also as output. The MIT Media Lab group explored the interaction possibilities of one of this systems - a actuated shape-changing curve interface. They built a small and a large version of their prototype, called LineFORM, both composed by a series of servo motors. The large prototype features high torque, deformation sensing. stiffness change and has the ability to render 3D structures. It smaller version had a higher resolution, but only worked in the field of 2D curves display and interaction.


Figure 2 - Users can manipulate CAD models with LineFORM. Here the curve is computationally limited to form right angles. (Image copied from original MIT Media Lab paper about LineFORM)
Figure 2 - Users can manipulate CAD models with LineFORM. Here the curve is computationally limited to form right angles. (Image copied from original MIT Media Lab paper about LineFORM)


Regarding the curvy contours designing field being studied in this project, the MIT Media Lab large LineFORM can be used to physically render and manipulate bezier or NURBs curves in 3D in Computer Aided Design applications. The researchers used the angle and stiffness control of each joint to combine direct manipulation and shape output, in such a way that users could intuitively control the model’s shape, as shown in Figure 2. The smart shape-changing ruler function for the small LineFORM (Figure 3) could also be applied in the designing field in vogue. It could be applied in the first step of the car design process, for example.


Figure 3 - LineFORM can be used as a smart ruler with different shapes. (Image copied from original MIT Media Lab paper about LineFORM)
Figure 3 - LineFORM can be used as a smart ruler with different shapes. (Image copied from original MIT Media Lab paper about LineFORM)


In this research project we also explore the Computer Aided Design domain, but with a different focus. Instead of focusing on the possible interaction techniques of a tangible approach, we pursue a cross-over point in efficiency between traditional and tangible input systems (passive approaches).

Apart from MIT Media Lab’s LineFORM, another paper served as base and inspiration for this project: "Tangible user interfaces for physically-based deformation: design principles and first prototype" by N. Takouachet, N. Couture, P. Reuter, P. Joyot, G. Rivière, and N. Verdon. The study conducted in this paper diverges from the present goal: it proposes principles for designing tangible user interfaces for physically-based deformation. However, there are two main aspects of this study that have contributed for this project.


Figure 4 - Example of use of ShapeTape tool. (Image copied from the original paper "Tangible user interfaces for physically-based deformation: design principles and first prototype")
Figure 4 - Example of use of ShapeTape tool. (Image copied from the original paper "Tangible user interfaces for physically-based deformation: design principles and first prototype")


Firstly, the prototyping tool used: ShapeTape (Figure 4). It is an “an array of fiber optic sensors fixed on a thin malleable strip of metal coated in plastic for protection” that had been previously used in a research conducted by Balakrishnan and Hinckley. The beam shape of this tool was considered a good approach for our project, because its affordance naturally invites users to bend it. In the mentioned study, the prototype’s left extremity is fixed to the table. This approach was considered for this project, but discarded after further reasoning due to its imposed limitation to user shape manipulation. The second compelling component of this study was its well-structured and thorough research process. By reading the paper, we learnt a lot about how to design a research question, a suitable prototype to assess it and an user study to validate the raised hypothesis.

Having introduced the subject of this research project, its objective and the study of the related state of the art, the next section presents the approach chosen to conduct this study.

Project's Approach

General description

As mentioned in the Introduction, the proposed solution is a tangible input device. Tangible interactions with computers are proven to be more intuitive than the traditional mouse/keyboard ones. In the domain of drawing in a 3D space, the tangible input device will have the same degrees of freedom than the object being modeled. In this way, it could even be more efficient than the traditional non-digital tools, such as drawing with pencil and paper, where the lack of degrees of freedom must be compensated with more complex drawing techniques. Another task made simple by a tangible device is to modify and reshape a drawing.


Having in mind all the applications of a input device for a 3D drawing application, we chose to keep with only 2D contours, due to the feasibility of the prototype and experiment. Therefore, the chosen approach was to build a long line of a malleable material from which the shape is shown in real time in the computer screen. The generated image can be saved at any moment and, then, the user can still play with the tangible device and modify the drew shape. This allows the user to try different shapes and save all the options in a really simple way.


This approach targets novice and medium users (non-expert users). We imagine that the precision of a computer is going to be higher than the one of the tangible device. That is why this approach aims at helping the user do a rough draft of the curvy contours he wants by trying and changing his drawing as many times as he wants. Then, he saves these drafts in a format that can be easily modified with a drawing software later, such as SVG (Scalable Vector Graphics), where he can refine them.

Implementation

At first, the tangible input device was built with a roll of clay. In the first experiments, the clay became quickly dry and started to break a lot. As this problem interfered in the experiment, the material was changed: 30 cm of a round shoe lace was used and a metal wired was placed inside of it, adding some rigidity to it and allowing it to maintain a shape by itself. In order to have the real time image in the computer, we used the motion capture system called Optitrack. So, a group of optic markers was attached to the device. Three cameras were use to read and detect the position of the markers. Then, these positions served as input to our software, that fit a curve to these points.The cameras where calibrated to get all the markers’ positions in a plane, therefore generating only a contour in a 2D space.

Link to Video 1
Video 1 - Anshul explaining how the tangible input device system works. In the video he is using the clay roll that was later replaced by a shoe lace filled with a metal wire (Figure 2a) due to problems found during experimentation.

Experiment

The user study - evaluation process involving users - was designed to help answering the research question:

When drawing curvy contours, at which point it becomes more efficient to use a tangible input approach instead of a traditional input such as a mouse?

In order to have an overview of the experiment, some photos of the experiment setup can be found in Figure 5.


Figure 5 - a) Tangible input device made of show lace stuffed with a metal wire, used in the second round of experiments. b) Result images of tangible input system: on the left curve fit through points representing the optic markers and on the right final image in the predefined size for later computation of the error. c) Experimental setup with the 3 Optitrack cameras, the tangible input system on the left and bottom center, the original image in the screen in the center and the mouse input system with Inkscape on the right screen.
Figure 5 - a) Tangible input device made of show lace stuffed with a metal wire, used in the second round of experiments. b) Result images of tangible input system: on the left curve fit through points representing the optic markers and on the right final image in the predefined size for later computation of the error. c) Experimental setup with the 3 Optitrack cameras, the tangible input system on the left and bottom center, the original image in the screen in the center and the mouse input system with Inkscape on the right screen.

Assumptions and Intuitions

In this part, we aim to, through testing, try to answer the research question. That is, an experiment was built in order to gather information that, when evaluated, will help define if there is a point from which it is more efficient to draw curvy contours with a tangible input device instead of using a mouse and, if yes, determine this point.


When planning the experiment, some intuitions about the performance of each interaction were taken into consideration. First, it was considered that the traditional interaction will be more precise for contours with sharp corners. The second consideration was that the tangible device would be more efficient for smooth curves, because the lack of precision would be less important for smooth curves and would also be compensated by how fast it is to manipulate and create smooth shapes with this device.


According to the expert Laura Pujades, corners and peculiar intersections of surfaces, like the one shown in the Figure 6, are examples of elements that are complicated to construct with drawing softwares. Having this as a background, some assumptions were made about how to define the difficulty level of a task. We assumed that the steeper the peak of a curve, the more difficult it is to draw it. The number of peaks composes as well the difficulty level, in a way that the greater the amount of peaks the harder the task.


Figure 6 - Example of peculiar intersection of surfaces that can be found in car design and that presents complications when building in a drawing software.
Figure 6 - Example of peculiar intersection of surfaces that can be found in car design and that presents complications when building in a drawing software.

Structure

In the experiment day, each participant will first hear a brief explanation about the project, then he will be asked to perform some tasks and, finally, he will have space to share any thoughts he has had about the experiment. These tasks will be divided in two batteries, each one performed using one of the input approaches in vogue. Before each battery of tests, the user will have a simple explanation about how to use the tool and some time to experiment on it. Once he is ready to start, he will execute each task of the battery in the prefixed order. Each task will be timed and its results will be saved for future calculation of the error, as in this project efficiency considers both time and error.


--- Tools ---

Two input devices are being tested: the tangible one described in the Project’s Approach section and the mouse - the traditional input device for drawing and designing softwares.

For the tangible approach, the results are produced in real time by the software we developed and are shown to the user.

Regarding the traditional approach, Inkscape was the chosen 2D drawing solution due to its robustness and the type of file it manipulates - SVG. The two tools participants were advised to use are the one for drawing Bezier curves and straight lines and the one for editing paths by nodes.


In order to neutralize the learning effect for each tool, the user will first perform all the levels with one tool and then repeat all of them with the other tool. In addition, due to statistical reasons, half of the participants will start the experiment using the mouse in Inkscape and the other half will start using the tangible system.


--- Tasks ---

A task consists in looking at the original curve in one screen and trying to reproduce it using the input device at hand. The original curve image, the Inkscape canvas and the live result of the tangible input have all the same size. Thus, the participant is asked to try to reproduce the curve with the correct proportions. He is provided with a start and a end point calculated from the original image.

In total, there are 8 (eight) tasks. The tasks have increasing levels of difficulty. This also helps to neutralize the effect that the learning of each tool may have in the results.


Considering the intuition about sharp and smooth corners, we defined these two as types of curves: sharp and smooth. In an attempt to produce unbiased results, the types of task were balanced all through the battery, in the following order: [type 1, difficulty 1], [type 2, difficulty 1], [type 1, difficulty 2], [type 2, difficulty 2], and so on. The task order can be seen in Tables I and II, by reading the read boxes that indicate T for task and its order from 1 to 8.


--- Difficulty levels ---

Two criteria were considered

difficulty level = steepness of the peak(s) of the curve + number of peaks

In order to have acceptable results, it is important to thoroughly test the relation(s) among criteria. Thus, all the criteria must be fully crossed. In this case, there are 3 criteria (type + 2 criteria of difficulty). The number of values was chosen in light of the total time we wanted the experiment to take. Having 16 tasks - eight with each input device - seemed to result in a reasonable amount of consumed time and a fair amount of produced data to analyze. Therefore, we chose two values for each one, totalizing 8 different combinations, that is, 8 tasks.

The chosen values for steepness of peak were: little and important. The values for amount of peaks were 1 and 2. The Tables I and II summarize the criteria combinations.


Tables I and II - Curves used as tasks showed as a match of 3 criteria: type of curve, steepness, and number of peaks. The number indicated in red after a T represents the order in which this curve was presented as a task for the participant during the experiment. The number indicated in black after a L is the difficulty level of the given combination of the criteria.
Tables I and II - Curves used as tasks showed as a match of 3 criteria: type of curve, steepness, and number of peaks. The number indicated in red after a T represents the order in which this curve was presented as a task for the participant during the experiment. The number indicated in black after a L is the difficulty level of the given combination of the criteria.


--- Measured parameters ---

The parameters taken into account for the efficiency evaluation are error and time.


  • Error:

    The original image is also the ground-truth imaged that will be used in the pixel level comparison. All the images were generated with the same pixel size and the participants were asked to use the provided start and end points provided. Therefore, no adjustments are necessary before comparing the result images to the ground-truth ones.

    Two algorithms were used for the computation of the error: the Mean Structure SIMilarity (MSSIM) and the Mean Squared Error (MSE). The simplest and most widely used full-reference quality index in order to compare two images is the mean squared error (MSE) index. It is computed by averaging the squared intensity differences of the approximated image and the reference image pixels. The MSE is easy to implement, but one can run into problems when using it for similarity. The main one being that large distances between pixel intensities do not necessarily mean that the contents of the images are different. Thus, in order to assure the correctness of our similarity approach, we use a second approach called Mean Structural SIMilarity (MSSIM) in order to check our results.

    The Mean Structural SIMilarity (MSSIM) index is a method for measuring the similarity between two images. The MSSIM index can be viewed as a quality measure of one of the images being compared, provided the other image is regarded as of perfect quality. MSSIM value can vary between -1 and 1, where 1 indicates perfect positive correlation (perfect similarity), 0 indicates no correlation (no similarity) and -1 indicates perfect negative correlation. Detailed description of the MSSIM index is given in the paper "The SSIM Index for Image Quality Assessment" by Zhou Wang et al., 2004.

    Each participant will have 16 result images, 2 for each task - one with each tool. These two images for each task will be compared to the same ground-truth image, as we use the same ground-truth image for each task.


  • Time:

    The participant will be timed for each performed task with each tool. Thus, each participant will have a total of 16 recorded times, 8 for each battery.


In the end, a correlation between efficiency (time&error) and difficulty level was searched. The goal of the analysis is to study at which point it seems better to use the tangible input device instead of the traditional mouse for drawing curvy contours and even if such point exists at all.

Problems and limitations

During the experiment, three major problems were encountered. The first was that the optic markers were not always recognized as being in the same plan. This caused problems to the resulting image generated by our software, that became completely inaccurate.

The second was that the first material used - the clay - turned out to dry really fast when being handled a lot as in the case of the experiment. Once dry, the clay roll started to break a lot and this disturbed the experiment flow. So, to prevent it from altering the results, the clay roll was changed for another material: a shoe lace stuffed with a metal wire for maintaining the position of the shoe lace. In this case, to glue the markers over the show lace, double-sided tape was used. However, there is where the third and final problem appeared.

Due to the constant manipulation, the tape lost its glue and the markers started falling. We found a solution during to the experiment that allowed us to continue it without great effect on the experimental results. But, because of this issue, the experiment as a whole took much more time from the participants and slightly affected their satisfaction regarding the tangible system (that turned out to be less than we expected).


The prototype showed two major limitations. Firstly, due to the use of the Optitrack system, the manipulation of the tangible device was a bit compromised. The participants should avoid putting their hands in between the camera and the markers. Also, the markers needed to always be facing up, what limited the movement participants could do with the device.


Finally, the curve fitting had a medium accuracy. Therefore, not only the errors of the tangible input system results could be affected, but also the times due to the distance in participants’ perception of the tangible and the digital image. When asked, participants said that they took into consideration that it was a prototype, so we judge that the times recorded were not affected. However, the errors will be influenced a least a bit by this prototype limitation.

Results

The experiment had 10 participants, all non-expert users of drawing softwares. Each of the 10 subjects had to perform 8 tasks with each input system, resulting in 16 exercises per subject and 160 trials in total for all the users. Although they were all chosen randomly, the population was limited to students from Grenoble. As the research question focus on non-expert users (including novice users), this methodology was considered appropriate to choose the subjects for the experiment. Five of them started using the tangible input device and the other five started using the mouse to draw in Inkscape. In the beginning of the experiment, each one of them heard a general explanation about the experiment and had time to practice each tool right before using it. The script for the explanation can be found in Appendix A. The Videos 2 and 3 show a practice of this start point of the experiment and of a participant performing 1 task in both tested systems.


During the experiment all the tasks performed where timed. Some of the participants made comments about the experiment and the tested input device systems - summarized in Appendix B. These comments will be used further in this documents, in the Conclusions section.

Link to Video 2
Link to Video 3
Videos 2 and 3 - Experiment test: Laura giving explanation for a participant and participant performing one task with both tangible (clay roll) and traditional (mouse) input devices. In the real experiments, the participant was given time to practice each tool before performing its battery of tasks.


For the data analysis, all the times and errors were complied and an average per task was calculated. For the error two methods were used. The MSSIM represents the structure similarity, this means, the higher the value, the more precise the tool. The MSE represents the average error, so the lower the value, the more precise the tool is. Regarding the time, the lower the time, the more efficient the tool was for a certain task.


Moreover, in an attempt to try to find the cross-over point (the one that delimits when it becomes more efficient to use the tangible device than a mouse) the average times per curve were grouped per tool and analyzed according to the curve type, as seen in Graph 1.


Graph 1 - Average time per curve for both input methods. Curves ordered per type (sharp or smooth) and per difficulty level (1, 2, 3 or 4). All times given in seconds.
Graph 1 - Average time per curve for both input methods. Curves ordered per type (sharp or smooth) and per difficulty level (1, 2, 3 or 4). All times given in seconds.


As it can be observed in Graph 1, for the sharp curves there was no concluding difference in time. Nevertheless, the tangible input system performed, in average, 2.64 times better than the mouse&Inkscape system for smooth curves. The curve difficulty level did not have a significant influence on the time efficiency of the tools. Therefore, one could say that a cross-over point was not found for a curve difficulty, but only for the curve type: for smooth curves the tangible system has a better time efficiency.


Regarding the error, as predicted, the mouse system led to more precision in all tested cases. This result was confirmed by both used algorithms: MSSIM (Graph 2) and MSE (Graph 3).


Graph 2 - Average for Mean Structural SIMilarity (MSSIM) per curve. The higher the value of MSSIM, the closer the resulting image was from the ground-truth one for that specific task.
Graph 2 - Average for Mean Structural SIMilarity (MSSIM) per curve. The higher the value of MSSIM, the closer the resulting image was from the ground-truth one for that specific task.


Graph 3 - Average for Mean Squared Error (MSE) per curve. The higher its value is, the more differences (error) there were between the resulting image and the ground-truth one for that specific task.
Graph 3 - Average for Mean Squared Error (MSE) per curve. The higher its value is, the more differences (error) there were between the resulting image and the ground-truth one for that specific task.


Nonetheless, no relations were found between precision and curve type or difficulty level. Hereafter, the average MSE per task will be used when discussing the tasks’ error due to the corresponding results reached through both methods.


So far, the tangible input system has better time efficiency and the traditional input system is more precise. Thus, we searched for a correlation between error and time. However, as it can be observed in Graphs 4 and 5, no clear correlation was found, regardless of the input method or curve type. The only information we extract from these correlations is the confirmation that the traditional method was more precise than the tangible one.


Graph 4 - Correlation between error (computed using MSE) and time for all participants in tasks performed using the traditional input system - mouse&Inkscape.
Graph 4 - Correlation between error (computed using MSE) and time for all participants in tasks performed using the traditional input system - mouse&Inkscape.


Graph 5 - Correlation between error (computed using MSE) and time for all participants in tasks performed using the tangible input system.
Graph 5 - Correlation between error (computed using MSE) and time for all participants in tasks performed using the tangible input system.


The influence of learning effect of each tool on the results was also checked. According to the Graphs 6 and 7, no learning effect could be observed, neither for time efficiency nor for precision.


Graph 6 - Average time per task for both input methods. Curves ordered per experiment order in for analysis of tool learning curve during experiment. All times given in seconds.
Graph 6 - Average time per task for both input methods. Curves ordered per experiment order in for analysis of tool learning curve during experiment. All times given in seconds.


Graph 7 - Average error (MSE) per task for both input methods. Curves ordered per experiment order in for analysis of tool learning curve during experiment. All times given in seconds.
Graph 7 - Average error (MSE) per task for both input methods. Curves ordered per experiment order in for analysis of tool learning curve during experiment.


From Graph 6 we can also observe that the tangible input system allows a better consistency in regards of the time spend per task.

Conclusion

About research question and experiment

Through the analysis of the presented results, we can find a partial answer for the research question, just in terms of time efficiency. We found that, in terms of time, the tangible input system was more efficient for smooth contours than the traditional one, and that, in these cases, the tangible was in average 2.64 times faster than the mouse. Taking into account the prototype limitation mentioned related to the curve fitting procedure used, one could consider satisfactory the answered reached for the time efficiency.

Concerning the precision, the traditional approach had a better performance. During the experiments, we observed that, the fact of having a more accurate tool (mouse and Inkscape) led people to try to have a better result, taking a bit more time. We also observed that due to its complexity, this tool had a minimum time necessary to complete no matter which task of the type smooth. We, therefore, wonder, how this sense of accuracy could influence in time efficiency and in the real precision attained. Having not been able to reach any conclusions about this with our experiment, we see the study of this aspect as a possible future development.

A limitation of these results is, however, that all the tasks studied were the creation of a new curve. No emphasis was given to the problem of opening a already created design and trying to modify it. The prototype was not prepared for recognizing a digital curve and reproducing it, like the LineFORM from MIT Media Lab can do.


Regarding the time x error correlation, more studies would need to be carried out in order to better determine this correlation and its effects on the studied input device systems. We also ponder that a more accurate prototype would be needed in this type of studies. One important improvement would be to use another fitting technique for generating the curves from the tangible input device. At the end of the project, we came across the NURBS (Non-Uniform Rational B-Spline) - a mathematical model generally used for generating and representing curves and surfaces. It is commonly used in computer-aided design (CAD), manufacturing (CAM), and engineering (CAE) and can be found in various 3D modeling tools. The NURB model is considered to be a good option for a next exploration on the present topic, and is thought to be useful to overcome the just presented shortcoming of this project.


Overall, we mature the use of this tangible input device as a way to have a fast rough draft of curvy contours that could be later modified in order to improve its precision and add details. The time consistency of the tangible approach also points towards this direction, where for a first draft it is fast no matter the contours' difficulty level. The participants’ comments also confirm this point of view and add that, if a final version of the tangible device could be produced, it would be really useful for many different tasks. This final version should be entirely trackable and use materials that allow better precision.

We also consider this tool to have a ludic, playful characteristic that could be explored in educational applications and applied in studies for digital animation systems controlled by a tangible device.


With respect to future developments, three thoughts stand out. The first is to redo this study, but now considering 3D contours and not only 2D ones. Our intuition is that the potential of the tangible input system will be better exploited in this case, generating a bigger performance difference in favor to the tangible device.

The second would be also consider a mixed task in the seek of a cross-over point in efficiency between the traditional and the tangible input system. This mixed task would be create a curve with the tangible device and then improve it in a drawing system. This would embrace the opportunity brought up by the participants - tangible for rough draft and traditional for refinement.

And finally, it would be interesting carry a study that would not have the limitation mentioned in this section. Thus, the research would also consider tasks where a ready digital curve would need to be altered by the tangible system and try to find a cross-over point in efficiency between the two input systems.

Auto-evaluation about the project

In general, it was a really good project to understand methodology of a study oriented by human centered interaction. Also, it helped us understand better the daily routine of a researcher and its challenges, from delimiting a well-defined research question, to considering all the possible variable when planning the experiment, finding participants and selecting and analyzing all the resulting data. We could see in our experience what the professors had already warned us about: experiments go wrong, so it is important to schedule a good amount of time to carry all the necessary sessions.


We acknowledge that a lack in more profound statical knowledge may have affected the results analysis. A bigger amount of conclusions or of details in the raised conclusions might have been possible for students with more practice in statistical analysis of experiment results.

We also believe that a better scheduling of the work by our part, the students, could have led to more interesting results. We might have had time to conduct other rounds of experiment to asses small changes of structure coming from partial results analysis.


Two remarks about the project organization that we would like to mention. The first is about the project’s tutor. Our professor was really helpful. However the fact that she was not physically present or available (she was living in another country) during the project duration affected a bit negatively the development of the project and the learning experience.

The second is about the final document format. Having never created or modified any pages on Ensiwiki, doing so caused a bit of stress. Even after reading many pages of instructions about the system, we still faced problems to understand how to add images and videos. Maybe an idea for next year would be to create a Ensiwiki template for this project and a page explaining in a few lines these most important elements for the present project.

References

[1] Nakagaki, K., Follmer, S., & Ishii, H. (2015, November). Lineform: Actuated curve interfaces for display, interaction, and constraint. In Proceedings of the 28th Annual ACM Symposium on User Interface Software & Technology (pp. 333-339). ACM.


[2] Takouachet, Nawel & Couture, Nadine & Reuter, Patrick & Joyot, Pierre & Rivière, Guillaume & Verdon, N. (2012). Tangible user interfaces for physically-based deformation: Design principles and first prototype. The Visual Computer. 28. . 10.1007/s00371-012-0695-y.


[3] Danny Leen, Raf Ramakers, Kris Luyten StrutModeling: “A Low-Fidelity Construction Kit to Iteratively Model, Test, and Adapt 3D Objects”. In Proceedings of the ACM symposium on User interface software and technology (UIST '17).


[4] Arika Yoshida, Buntarou Shizuki, Jiro Tanaka: “Capacitive Blocks: A Block System that Connects the Physical with the Virtual using Changes of Capacitance”.


[5] Moritz B¨acher, Benjamin Hepp, Fabrizi Pece Paul G. Kry, Bernd Bickel, Bernhard Thomaszewski, Otmar Hilliges “DefSense: Computational Design of Customized Deformable Input Devices”.


[6] How-to: Phython Compare Two Images: https://www.pyimagesearch.com/2014/09/15/python-compare-two-images/


[7] Michel Alves: "Similarity Maps Using SSIM Index" on https://www.slideshare.net/michelalves/similarity-maps-using-ssim-index


Appendixes

Appendix A - Script of experiment explanation

Hello,


First we would like to thank you for your time and explain you a bit of what we are doing. We are developing a project in University to study human-centered interaction. We are studying the efficiency of a new tangible input device we created. For that, we will compare its performance with the one of a traditional drawing software - Inkscape.

To test our creation, we will ask you to draw 8 different curves, first using our device and then using Inkscape. We ask you to try to be the fastest and more accurate possible. Once you have finished each drawing, tell us I am finished.


For each task, we ask you to draw the curve being shown to you in the screen of the computer.


With the clay, you will have a start and end point indicated in the sheet of paper in front of you. You should use them for your drawing. The sheet of paper is proportional to the canvas with the resulting curve you can see in the screen of this notebook. You should try to meet the proportions of the original curve. The resulting curve is calculated based on the position of the markers on the clay. These positions are captured by these cameras, so please be aware of your hands, trying not to cover the markers too much. You can add and remove parts of the clay if you want to change its length.


In Inkscape you will also have a start and end point for each curve.

Appendix B - Summary of participants' opinions about experiment and tested system

Participant 5 found the traditional input method using Inkscape and a mouse more efficient to draw any shapes, but more specifically for straight lines and sharp corners. He also mentioned that the prototype contributed for him making some mistakes due to its limitations. He added that the way to see the resulting image from the tangible input device was also not that great and that it caused some problems when trying to be accurate in the matching with the original image.


Participant 6 believed the tangible input system to be more efficient for smooth curves, mostly if the prototype was improved or in a hypothetical final version where all the tangible device position would be detected, not only the markers. She stressed that a big advantage she saw on the tangible system is that it is faster to learn how to use it - its learnability. She also believed the traditional method to be more efficient for lines and sharp corners. She concluded that the experiment was okay and would be really good once the problems with the markers falling would be fixed.


Participant 8 had an opinion similar to the one form Participant 6. He also thinks the traditional mouse&Inkscape system is better when precision is needed and the tangible one, regardless of its limited precision, is really good for drawing a first draft of a shape. In his opinion, for the tangible device to become usable in a real situation, the entire wire / roll would need to be trackable. He added a suggestion of using a mix of hard and soft parts made of different materials that would maintain the time efficiency and allow to improve the precision of this tool. Finally, he suggested that both methods should be used together for complicated shapes: the tangible would create a first rough draft and then the traditional would be used to improve and add details to it.