A Browser-Based Perceptual Experiment Platform for Visual Search Study in Augmented Reality System

This paper describes a real-time experiment platform for visual search in Augmented Reality (AR). Searching for a virtual object embedded in an AR environment is an example of a visual search task. However, systems implemented in the previous visual search studies have focused on static images and pre-recorded videos. Our system supports real-time video captured from a camera mounted in a Head-Worn-Display (HWD) as input. The system allows customization by the experiment administrator according to their specific requirements. A visual clutter analysis method known as Feature Congestion is utilized to analyze the visual clutter of video scene and optimized trying to meet the requirements of real-time operation. Different optimization techniques such as multithreading, general computation using GPU, fast matrix convolution are explored. The theoretical improvement and limitations of the optimizations have been discussed. By utilizing JavaScript and HTML, this browser-based platform is designed to be cross-platform and extensible. Many state-of-art HTML5 technologies are employed in the system to both improve the computation performance and strengthen the interactions between users and system.