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
4.2 Task performance
In this subsection we present the analysis of the results gathered from the users on the evaluation session. For assessing task performance of the users between the different representations we collected data through logs. This data were: time, for evaluating efficiency of the representation; number of obstacles hit ,for spatial awareness evaluation. Figures 4 and 5 show the times for each task in both perspectives and representations. Since the fourth sub-task had a fixed time of execution we chose to not use time for this task in particular. Instead, we used the number of balls caught during the fourth sub-task (Table 3). The number of obstacles hit and balls caught can be found on Table 2. Because of the small amount of obstacle in each task we also used the percentage of collision avoidance as an additional metric in the comparison between representations.
In the following sub-sections we present the results obtained for each of the metrics used (time, number of obstacles hit and balls caught) for each of the sub-tasks.
4.2.1 Task 1
Collision: Grouping results by perspective we can say that users collided with more objects in the First-Person Perspective with the Point-Cloud avatar against the Abstract Avatar (Z=-2.668,p=0.008). In Third-Person Perspective we found a statistical significance between the Mesh and the Point-Cloud representation, with the PointCloud avatar having a higher number of obstacles avoided (Z=- 2.542,p=0.011).
When grouping perspectives by representations we can say that we only found statistical significance in the Point-Cloud representation, with the Third-Person Perspective having the higher number of obstacles avoided (Z=-3.490,p<0.005).
Time: On the execution of the first task, we found a statistically significant difference in the First-Person Perspective between the Point-Cloud and the Abstract Avatar (p=0.024), with the abstract having the better performance. It was also found between the Abstract and Mesh Avatars (p=0.015), with the Abstract having the edge. When comparing the same representation for both perspectives, statistical significance was found between all avatars ( Abstract: t(18)=-6.312, p =0; Cloud: t(20)=-3.254, p= 0.001; and Mesh: t(16)=-3.254, p = .005). For the three representations, the First-Person Perspective had the better performance.
4.2.2 Task 2
Collision: In the representation-perspective test, we found statistical significance, with the Abstract avatar (Z=-2.714,p=0.007) having the most number of obstacles avoided overall in 3PP. Also, between the Mesh and Point-Cloud Avatars (Z=-3.779,p<0.005) and between Abstract and Mesh Avatars in the Third-Person Perspective. In the First-Person Perspective no statistical significant difference was found. No significance was found between 1PP representations.
Also, when comparing the different representations in both perspectives we found statistical significance for both Point-Cloud (Z=-2.941,p=0.003) and Mesh (Z=-3.673,p<0.005) Avatars, with the first person having the advantage in both cases.
Time: No statistically significant difference was found between any of the representation-perspective combinations. When comparing the same representation and two different perspectives, only in the mesh representation a statistically significant difference was found (t(18)=-2.479, p=0.023), with the first person perspective having the advantage.
4.2.3 Task 3
Collision: In the task, no statistical significant difference was found when using the representation-perspective grouping factor. However, we found statistical significance in favor of the Abstract Avatar in 1PP against the same representation in 3PP (Z=-2.714,p=0.007).
Time: On the third task, we found a statistically significant difference when comparing the abstract (p=0.05) and the point cloud (p=0.045) representations to the mesh, but not between themselves. In both situations, the mesh avatar had a worse performance. Similarly to the first task, all representations had a better performance in the first person perspective (Abstract: t(20)=-6.76, p=<0.005; Mesh t(18)=-4.276; Point Cloud: t(19)=-5.354, p<0.005)
4.2.4 Task 4
For the ball catching task we found statistical significance on the First-Person the between Point-Cloud and Mesh Avatars (Z=- 2.546,p=0.011),with better results for the Point-Cloud representation, and Abstract and Mesh Avatars (Z=-2.401) Table 3, with better for the Mesh Avatar
When comparing the representations between perspectives, we found statistically better results for the Point-Cloud in Third-Person when compared with the same representation in the First-Person Perspective (Z=-3.961,p<0.005).
This paper is available on arxiv under CC BY 4.0 DEED license.
