As I was studying the many visualizations that ATC controllers were using, I came to be interested in the final steps of the design process. Jean-Luc Vinot: trained as a graphic designer, Jean-Luc was often requested for the task of tuning the final rendering of the images when there was a change, be it an addition of a new functionality on a radar image, or because the national administration planned to change the aging CRT screens for LCD ones. Though talented, Jean-Luc encountered seldom reported difficulties: the tuning was at such a level of detail that a subtle amount of difference in values (say of the luminosity of an object) would have a large impact in certain conditions on the entire image, and would make the visualization unbearable for controllers. For example, Jean-Luc designed a color for small/medium size military sectors. The color is a gray with a hint of red (it is called “lie de vin”). He later used the color palette in another control center, embedding a larger military zone. When this same color was applied to this surface, the reddish gray seemed too saturated (i.e. too red). He had to decrease the saturation in order to make sectors look grayer when they are large, but still keep a distinctly reddish nuance when they
4.4. VISUAL VARIABLE PROCESS DESIGN 35
are smaller.
In fact, visualizations of rich ATC and cockpit visualization are composed of a great amount of graphical elements that users scrutinize for long periods of time in a demanding cognitive context. Perception of graphical elements is highly dependent on multiple interactions between visual dimensions such as color, area, shape etc. and display context such as type of screens and surrounding luminosity. Understanding these interactions involves multidisciplinary knowledge: psychophysics, human computer interaction and graphical design. I hired Gilles Tabart, who had just graduated in computer science, to do a PhD about this topic. I also set up a project with Jean-Luc Vinot and Sylvie Athènes, who was trained as a neuro-psychologist.
We sought to answer those questions: How can visualization designers make sure that they minimize the risk of confusion between two objects that are supposed to differ when perceived? How can they be sure that any modification made on a 20 year old system will not hinder the perception, and hence the activity, of the users? How can users and stakeholders be convinced? In general, how can designers validate, check, assess, and justify their design? Tools and methods to design, justify, and validate user interfaces at the level of graphical rendering are still lacking. As these interactive systems are used in critical situations, the need for sensible, justified, and verified design is even more important.
We worked on many aspects underlying this topic: controlled experiments, activity analysis, design of tools and methods. We ran controlled experiments to understand the phenomenon of visual variable interaction more deeply [TACV07]. For instance, we measured how the arrangement of pixels along a diagonal lowers the discriminability compared to horizontal or vertical lines. Based on the results of the experiments, we thought that we could design algorithms that would automatically adapt the rendering according to the values of the visual variables. However, we quickly discovered that relying on algorithms would give satisfying results in only a few cases. It would be better to provide the designers with the right set of tools allowing them to control rendering instead. We interviewed designers and devised a set of requirements and implications for the design of tools and methods of what we called "the design of graphic rendering" [TCVA08]. For example, a requirement is to provide a way for designers to tune a color with direct manipulation. A usual color-design interface shows a rectangle filled with the color the user is defining, so that he can verify the results of its manipulation. However, such an interface does not allow the user to understand the impact of his modifications on the rendering of each element that uses this color. We designed a visualisation that displays multiple instances of the use of that color at once (with shapes of different size, orientation, on top of different background colors etc.) instead of controlling and having feedback on one instance only.
We ran participatory design sessions to invent new tools to support designers and specifiers [TACV09]. We devised a set of tools that help designers build the various graphical scenes involving the same set of graphical objects e.g. the representation of flight in a radar image on multiple backgrounds or with multiple configurations (figure 4.2). Figure 4.3 shows an interface and a set of interactions to manipulate of set of related colors efficiently. We were pleading for a redefinition of the design process in order to include/cope with this topic (figure 4.4). However, we did not have time to implement a system to design controlled experiments aiming at verifying that the final rendering was fulfilling its expectation in terms of perception. The design of the interactions with the color tool and its implementation eventually led to the research on linked entities and dependent graphical properties I conducted later.
Figure 4.2: The Kabuki scene builder