Adaptive User Perspective Rendering for Handheld Augmented Reality

TL;DR

This paper introduces a lightweight method for user perspective rendering in handheld AR that reduces computational load by using optical flow and motion estimation, improving performance on limited mobile devices.

Contribution

It presents a novel, resource-efficient approach for user perspective rendering in handheld AR, reducing reliance on intensive face-tracking algorithms.

Findings

The proposed method significantly lowers computational demands.

It maintains accurate user perspective rendering on mobile devices.

Performance compares favorably to traditional head-tracked methods.

Abstract

Handheld Augmented Reality commonly implements some variant of magic lens rendering, which turns only a fraction of the user's real environment into AR while the rest of the environment remains unaffected. Since handheld AR devices are commonly equipped with video see-through capabilities, AR magic lens applications often suffer from spatial distortions, because the AR environment is presented from the perspective of the camera of the mobile device. Recent approaches counteract this distortion based on estimations of the user's head position, rendering the scene from the user's perspective. To this end, approaches usually apply face-tracking algorithms on the front camera of the mobile device. However, this demands high computational resources and therefore commonly affects the performance of the application beyond the already high computational load of AR applications. In this paper, we present a method to reduce the computational demands for user perspective rendering by applying lightweight optical flow tracking and an estimation of the user's motion before head tracking is started. We demonstrate the suitability of our approach for computationally limited mobile devices and we compare it to device perspective rendering, to head tracked user perspective rendering, as well as to fixed point of view user perspective rendering.