The research presented in this thesis (titled Evaluating Non-Photorealistic Rendering for 3D Urban Models in the Context of Mobile Navigation) examines the legibility of using different rendering styles in the context of 3D navigation on contemporary mobile devices. The styles examined deviate from the usual representation of 3D urban models on current applications, which at the moment strive for an approximation of realism. Other than the technical resource limitations posed by this, the use of this photorealistic approach, in the context area given, eliminates the possibilities of abstraction plus thematic highlighting and connectivity, obstructs inference to traditional cartography and flexibility in visualization. Finally, visual attention overload can be observed, which can be very detrimental to navigational decision-making. The alternative approach explored is the use of non-photorealistically rendered stylizations, of differing styles, suggested to possess benefits such as communicative aspects, which can influence low-level perceptual processes and emotional responses to cognitive workloads, interactivity motivation, the succinct presentation of context-assistive information and decreasing user effort. These styles are influenced by 2D human artwork, as simulated by computer algorithms.
The investigation consists of four empirical studies, conducted in a range of conditions, across a combined total of 211 participants. The first and second studies are based on the use of self-reported rankings for crucial spatial attributes of the, respectively, 2D and 3D scene stimuli. For the third experiment, following the Nielsen usability taxonomy and based on a route retracing task, objective measures are collected. Similarly, for the final experiment a route retracing task is repeated, which on this occasion contrasts the learning the rendering styles provide for a realworld urban location, rather than a virtual one. Across all experiments standardized additional tests such as the Santa Barbara Sense of Direction scale, the NASA TLX workload assessment tool and the IBM Computer Usability questionnaire provide additional information on both the user experience and the population sample itself. A three-tier architecture-based prototype pedestrian navigation application with GPS and digital compass sensor integration has been used in the studies, capable of the display and interaction of egocentric, real-time, georeferenced 3D urban location models.
Findings suggest a consistent common thread across all four experiments of strong support, based on extensive statistical analysis, in both user preference and user performance to non-photorealism over the verisimilar representation. This elicits the emergence of certain non-photorealistic styles, such as cel-shading and sketched shading, as effective new user interface paradigms for mobile navigation. Key outcomes from the experiments include, amongst others, the ranking of the sketched style ahead of photorealism in finer detail distinction and interaction engagement (first and second studies respectively). Additionally, another important observation is the superior performance of cel-shading in task time completion, error rate, disorientation and context-switching in the final experiment, signifying how easier it is for users to learn and remember a mobile virtual route rendered in this manner. All results are discussed and explained in terms of applicable existing perception work. The research concludes with a discussion on the suggested technical benefits of the non-photorealistic approaches, such as level-of-detailing and mip-mapping, progressive network transmission and generalisation as well as the potential use of biometrics-related modalities to collect additional information in future experiments.
It can be found here on the EThos system and here on the City Uni library.