Estimating Visibility of Annotations for View Management in Spatial Augmented Reality Based on Machine-Learning Techniques

Augmented Reality (AR) is a class of "mediated reality" that artificially modifies the human perception by superimposing virtual objects on the real world, which is expected to supplement reality. In visual-based augmentation, text and graphics, i.e., label, are often associated with a physical object or a place to describe it. View management in AR is to maintain the visibility of the associated information and plays an important role on communicating the information. Various view management techniques have been investigated so far; however, most of them have been designed for two dimensional see-through displays, and few have been investigated for projector-based AR called spatial AR. In this article, we propose a view management method for spatial AR, VisLP, that places labels and linkage lines based on the estimation of the visibility. Since the information is directly projected on objects, the nature of optics such as reflection and refraction constrains the visibility in addition to the spatial relationship between the information, the objects, and the user. VisLP employs machine-learning techniques to estimate the visibility that reflects human's subjective mental workload in reading information and objective measures of reading correctness in various projection conditions. Four classes are defined for a label, while the visibility of a linkage line has three classes. After 88 and 28 classification features for label and linkage line visibility estimators are designed, respectively, subsets of features with 15 and 14 features are chosen to improve the processing speed of feature calculation up to 170%, with slight degradation of classification performance. An online experiment with new users and objects showed that 76.0% of the system's judgments were matched with the users' evaluations, while 73% of the linkage line visibility estimations were matched.