Lighting in VR Study
12/10/2002
Paul Zimmons
This study investigated the influence of light on behavior in a virtual environment. In this case, the environment consisted of a gallery with a training room and gallery room separated by a door. The training room is on the left. The gallery room is on the right. There are 3 paintings and 6 vases in the gallery room (the 2 other vases are hidden by the dividers in this picture).
There were five different conditions explored in this experiment. All involved changes to the gallery room. The training room was the same in all the conditions seen by the participant. The different conditions involved highlighting different objects in the environment and seeing the impact it had on questionnaires and behavior recorded through tracker logs.
The conditions were as follows:
1) Area lighting (nothing highlighted)
2) Painting on the left and vase on the right highlighted (ratio 2 : 1)
3) Painting on the left and vase on the right highlighted (ratio 7:1)
4) Painting on the right and vase on the left highlighted (Ratio 2:1)
5) Painting on the right and vase on the left highlighted (Ratio 7:1)
The participant would start in the training room where they would receive instructions and get used to the head-mounted display. The participant would spend about 2:40 minutes in the training room and then the door to the gallery would open. The participant would have exactly 2 minutes in each condition.
Each participant saw all five conditions. The conditions were distributed following a balanced Latin square (with an odd number of conditions). This means that each conditions was equally likely to have another (random) condition before or behind it which minimizes the influence of order.
Before their first exposure, the participant would fill out a Participant Health Questionnaire, a Demographics Questionnaire, and then a Simulator Sickness Questionnaire.
After each exposure (5 exposures in total for one participant), the participant would fill out a Simulator Sickness Questionnaire, a Lighting Questionnaire, a Virtual Environment Questionnaire, and a PANAS Questionnaire. These forms were filled out on a computer in a room down the hall from the virtual environment tracker space.
Attention Maps
The behavioral metric used in this user study was an "attention map". This map entails recording the user's head position and orientation for every frame displayed in the head mounted display. Accumulating these readings on the geometry reveals the areas of longest dwell time within the space. In this case, tracker readings were taken 60 times a second over the entire exposure to the environment. However, only the results for the gallery room are reported here. These attentional "clouds" projected onto the geometry reveal information about the user's behavior in the environment during their exposure.
Individual Attention Maps
Before looking at average cases, let's examine the first user. This will help understand the concept of attention mapping and how it can reveal information about user behavior. The maps below were generated from one participant's run through all five conditions. The map on the left is what the user would see. The subsequent maps are attention maps derived from tracker readings for that participant.
1) Area lighting (no objects emphasized)
What the user saw:
And the resulting attention map.
2) Painting on the left and vase on the right (low contrast)
3) Painting on the left, vase on the right (high contrast)
4) Painting on the right, vase on the left (low contrast)
5) Painting on the right, vase on the left (high contrast)
Questionnaires
The two main results from initially analyzing the questionnaires are in the lighting questionnaire and the simulator sickness questionnaire.
The lighting questionnaire involves using a series of word pairs and scales between the words. Analyzing the questionnaires grouped by condition provides the following graph:
Conditions 2 and 4 are low contrast. Condition 3 and 5 are high contrast. The differences in the means are significant at the p=0.001 level. The lighting questionnaire follows the conditions well.
The simulator sickness questionnaires also reveal some interesting results. Five exposures results in a significant increase in simulator sickness (p= 0.02).
However, these questionnaires (and the others administered) need to be analyzed as within subjects to get more results.
Attention Maps for All Participants
1) Area lighting (no objects emphasized)
2) Paiting on the left, vase on the right (low contrast)
3) Painting on the left, vase on the right (high contrast)
4) Painting on the right, vase on the left (low contrast)
5) Painting on the right, vase on the left (high contrast)
For more precise comparisons, lists of object averages have been created. However, summary statistics involving these tables of values have not been processed yet.
Conclusion
Attention maps hold potential in providing insight beyond the usual analysis methods in virtual reality, gaming, training, and psychology. For virtual reality, it is easy to imagine a level-of-detail or pre-loading scheme based on average user behavior in an environment. For gaming, level design and level debugging could be enhanced with more quantitative decision making. For training, visual coverage or visual search routines can be analyzed at a much finer level. Differences between amateur and experienced users could also be evaluated. For psychology, this provides a new dimension to the conventional eye tracking methods. Attentional "BRDF"s which provide a summary of dwell time based on viewing angle and surface position can also be produced. Attention maps have applications in many different areas inside and outside of VR.