Predicting User Performance in Augmented Reality User Interfaces with Image Analysis Algorithms

Abstract

This study determines how to apply existing image analysis measures of visual clutter to augmented reality user interfaces, in conjunction with other factors that may affect performance such as the percentage of virtual objects compared to real objects in an interface, and the type of object a user is searching for (real or virtual). Image analysis measures of clutter were specifically chosen as they can be applied to complex and naturalistic images as is common to experience while using an AR UI. The end goal of this research is to develop an algorithm capable of predicting user performance for a given AR UI. In this experiment, twelve participants performed a visual search task of locating a target object in an array of objects where some objects were virtual, and some were real. Participants completed this task under three different clutter levels (low, medium, high) against five different levels of virtual object percentage (0%, 25%, 50%, 75%, 100%) and two types of targets (real, virtual) with repetition. Task performance was measured through response time. Results show significant differences in response time between clutter levels and between virtual object percentage, but not target type. Participants consistently had more difficulty finding objects in more cluttered scenes, where clutter was determined through image analysis methods, and had more difficulty in finding objects when the virtual of objects was at 50% as opposed to other scenarios. Response time positively correlated to measures of combined clutter (virtual and real) arrays but not for measures of clutter taken of the individual array components (virtual or real), and positively correlated with the clutter scores of the target objects themselves.