WebPoseEstimator: A Fundamental and Flexible Pose Estimation Framework for Mobile Web AR

PubDate: May 2021

Teams: Beijing University of Posts and Telecommunications

Writers: Yakun Huang; Xiuquan Qiao; Zhijie Tan; Jianwei Zhang; Jiulin Li

PDF: WebPoseEstimator: A Fundamental and Flexible Pose Estimation Framework for Mobile Web AR

Abstract

Exploring immersive augmented reality (AR) on the cross-platform web has attracted a growing interest. We implement WebPoseEstimator, a fundamental and flexible pose estimation framework, running on the common web platform, and providing the core capability to enable true Web AR. WebPoseEstimator provides a first real-time pose estimation framework that optimizes the loosely coupled multi-sensor fusion framework to flexibly adapt heterogeneous mobile devices. We also introduce how to optimize and compile such computationally intensive pose estimation from C++ source code into JavaScript profiles of less than 2 MB, thus supporting the fundamental and underlying capability for implementing true Web AR.