Authors:
(1) Rafael Kuffner dos Anjos;
(2) Joao Madeiras Pereira.
Table of Links
2 Related Work and 2.1 Virtual avatars
3.4 Virtual Environment and 3.5 Tasks Description
3.6 Questionnaires and 3.7 Participants
4 Results and Discussion, and 4.1 User preferences
4.3 Discussion
According to the questionnaires, we can conclude that the perspective has more effect than representation on embodiment and task execution, specially on the Abstract and Mesh representations. Also, we can state that the sense of embodiment with the Point-Cloud avatar is similar with the Abstract avatar, but significantly better when compared to the Mesh Avatar representation when the avatar is viewed through a Third-Person Perspective.
The lower sense of embodiment detected on the 3PP mesh representation, can be explained by the Uncanny Valley effect. While the point cloud directly maps the users body, both the Mesh and the Abstract representations can be considered simplifications, since not all body movements are directly mapped by the Microsoft Kinect (eg. detailed hand and head movement). This simplification is accepted on the Abstract representation, while its effect is perceived as an error in the Mesh avatar.
Regarding time efficiency on the execution of the proposed tasks, we found that the 1PP had a clear advantage over the 3PP. We argue that this result is related to the overall preference demonstrated by the users in the 1PP, as seen in Table 1. Being a more natural perspective to the user, the movement through the environment was faster when compared to the out of body experience of the 3PP.
When comparing different avatars, although we only found statistical significance in certain task-representation combinations, some considerations can be made. For the 1PP, the Abstract Avatar had the overall best performance (with statistical significance on the first task). Being a more minimalistic representation than the alternatives, less occlusion between body and obstacles were seen, with less distractions from the proposed goal.
For the 3PP, the Mesh Representation had the overall worse performance (with statistical significance on the third task). We relate this result to the lower sense of embodiment seen in Table 1and its relation to the Uncanny Valley effect, as explained above. This was noticed to frequently slow down the interaction.
Regarding the successful completion of the proposed tasks, the only task where the 3PP had the overall advantage was the first task, where spatial awareness was the key factor. Table 2 shows these results, with the Point-Cloud representation being the best performer between representations, with the lowest median number, and highest number of participants concluding the task without collisions (statistical significance found only when comparing with the mesh avatar).
For the second task, the 1PP had the advantage (statistical significance found for Mesh and Point-Cloud Avatar), except for the
Point-Cloud representation, which had worse performance in both perspectives. We attribute this to the fact that this representation is visually richer. Although this helps with the sense of embodiment, occlusions are naturally created by the rendered splats, or clothing (pants, shoes) in both 3PP and 1PP and body parts (breasts, stomach) in 1PP.
The third task was highly influenced by the wrongful estimation of the height of the obstacle. This can explain the fact that no statistical significance was found between any representation/perspective, except for the Abstract Avatar performing better in 1PP. Users would overestimate how low they needed to go to avoid the obstacle, having success most of the times. However, when analysing the questionnaires, users reported an overall preference for the 1PP for this task. This can be attributed to the fact that the virtual head in the 3PP is not exactly where the user’s real head is. The estimation of the height of the obstacle is more complicated from this point of view, and the user’s sense of balance is more heavily affected by the use of the displaced camera in 3PP, since lowering your body can affect your sense of balance.
Finally, 1PP had the advantage on the fourth task (Table 3). For this perspective, both the Point-Cloud and the Abstract Avatars had the advantage over the Mesh Avatar(with statistical significance). For the 3PP, Point-Cloud avatars had the worst performance (statistical significance found). We again attribute this to occlusions created by the users body and clothing which could inhibit the accurate visualization of the balls’ trajectories.
We noticed that the most “realistic” combination proposed (1PP Point-Cloud representation) had the fastest overall times, while having a high number of obstacles hit. This result can indicate that the user feels more confident in his distance estimation on this representation, leading him to a higher number of mistakes in avoiding obstacles.
This paper is available on arxiv under CC BY 4.0 DEED license.
