Occlusion handling for augmented reality environment using neural network image segmentation: A review.

Abstract

Recently, the advancements of handling occlusions for Augmented Reality (AR) introduces neural network-based image segmentation methods. However, it comes with increased computational costs. There has been some research that try to tackle this issue. Therefore, this paper compiles and reviews the recent articles that covers on this topic in order to identify the current achievements and the future works. The structure of this paper consists of explaining the basics of neural network-based image segmentation, discussing the methods in the published articles and suggesting expected future improvements. 14 articles were identified and 6 of them are discussed in this article. From the discussions, it is found that the common image segmentation methods comprise of semantic and instance segmentation, with instance segmentation giving more accurate and robust results for tracking objects but with higher computational costs. Some methods also incorporate depth-based techniques and 3D reconstruction to improve the accuracy. Based on the advancements, it is concluded that future works on this topic will be more on improving instance segmentation methods in order to reduce the computational costs.