Estimating the Depth in Three-Dimensional Virtual Environment with Feedback

Puneet Sharma and Ali Alsam


eye fixations, depth estimation, virtual environment


Visual interaction in 3-D virtual space can be achieved by estimating objects depth from the fixations of the left and right eyes. Training a PSOM neural network to estimate depth, from eye fixations, has been shown to result in good level of accuracy. Instead of training a neural network we postulate that it is possible to improve the accuracy of the fixation data by providing the observer with feedback. In order to test this hypothesis we introduce a closed-loop feedback in the environment. When the user’s visual axes intersect, within a range of the correct depth, a sound is produced. This mechanism trains the users to correct their fixations in a fashion that results in improved depth estimation. Our results show that indeed the accuracy of depth estimation improves in the presence of feedback.

Important Links:

Go Back