6.2. Practical scenarios
In our test scenario, we considered two important aspects that are often missing from HAR positioning experiments in the related work: (1) We used real target objects (Lego structures, Fig. 5) instead of virtual ones (e.g. target zones visualized with virtual rectangles) to simulate a practical scenario where virtual objects are very often spatially dependent on the environment [40]. (2) We did not have predetermined initial positions for the virtual objects.
In practical scenarios, the initial positioning is a fundamental part of AR content creation and it should not be separated from the position adjustment. Especially in case of SlidAR, where position adjustment is highly dependent on the accuracy of initial positioning. Simply adjusting the position between two points can be unrealistic if we are unable to justify why user would have chosen the specific initial position. In addition, we forced participants to move around while doing the tasks instead of just standing still or sitting. This is important, because HAR is used in mobile context, where users often move around.
There are still few matter that should be considered when applying our findings to practical scenarios, such as creating AR annotations to machines inside of factory or to medical equipment inside a hospital. In the test scenario, the participants were aware that the real objects are not mapped by the SLAM-system. This was because we wanted to focus on the specific HAR positioning problem that can occur often, but not every time. We designed the test scenario in a way that the positioning problem occurs every time. In practical scenarios, users might not always what in the environment is mapped and what is not. Thus, we would not know if the virtual object's initial position going to be correct or is position adjustment also needed. If we use SlidAR, it is not necessary to know is the real world object mapped or not because the initial positioning is conducted in similar manner in both situations. With HoldAR, however, we might need to choose a initial position differently if it is too far away from the target position.
We did not limit the movement in any way and participants were allowed us to freely move around the scene. Even though neither method did not require users to move 360° around the target position, various environments in practical secenarios might set limitations to the movement. This could possibly affect the efficiency of both methods: SlidAR requires the user to move to a new viewpoint and HoldAR relies to movement entirely.
Our test scenario was ideal for HoldAR, because we used an easily trackable ground plane in order to correctly show the depth cues. The complexity and the structure of the environment can vary a lot depending on the scenario. This can make the correct visualization of depth cues more difficult. SlidAR does not require depth cues to be visualized on the environment, thus it can be easily used in various practical scenarios with different level of environmental complexities.
The required level of accuracy can also depend highly on the scenario and the size of the objects that are being annotated. Small objects, such as buttons or cables, require very precise positioning. Larger objects, like factory machinery, can allow more ambiguity. Positioning to a larger object is easier, regardless of the used method. Initial positioning would be easier with SlidAR and the position adjustment would be easier with the HoldAR.
We used a tablet device, but both methods could also be used on a smartphone. Tablets are beneficial because they provide more screen estate thereby easing perception and gesture-based interactions [41]. This can be beneficial in industrial or medical systems where it is necessary to view conventional 2D information, in addition to AR information. The form-factor of the device and the amount of movement needed can affect the usability of HoldAR because it relies on the physical movement. With SlidAR, the form-factor affects the initial positioning because the device had to be kept as still as possible in order to perform the position adjustment correctly. The initial positioning can be improved by adding view freezing discussed in the previous section.
We chose a generic test scenario instead of a practical one, because the positioning problem can occur in any kind of practical scenario. Conducting the experiment in a practical scenario, such as inside a hospital or a factory is risky, because the results could be affected by unique features of the scenario itself. This would steer the research focus away from the fundamental object positioning problem that is not specific for any type of scenario. A generic test scenario allowed us to focus more closely to the positioning problem and it gave us a solid implications regarding the efficiency of SlidAR. Furthermore, the whole experiment gave us important knowledge about the positioning of virtual annotations to real world objects. Practical scenarios might have some differences compared to our test scenario, but these are rather minimal. Furthermore, we believe that in practical scenarios SlidAR would provide even greater efficiency over HoldAR, because HoldAR requires more movement and virtual depth cues. Despite the possible differences between our test scenario and practical scenarios, we strongly argue that our results can be applied to various scenarios, because the 3D positioning of virtual object is a requirement and fundamental part of any kind of practical scenario where we want to create AR content to the real world.