Absolute Scale Estimation of 3D Monocular Vision on Smart Devices

Abstract

This paper presents a novel solution to the metric, scaled reconstruction of objects using any smart device equipped with a camera and an inertial measurement unit (IMU). We propose a batch, vision centric approach which only uses the IMU to estimate the metric scale of a scene reconstructed by any algorithm with Structure from Motion like (SfM) output. IMUs have a rich history of being combined with monocular vision for robotic navigation and odometry applications. These IMUs require sophisticated and quite expensive hardware rigs to perform well. IMUs in smart devices, however, are chosen for enhancing interactivity—a task which is more forgiving to noise in the measurements. We anticipate, however, that the ubiquity of these “noisy” IMUs makes them increasingly useful in modern computer vision algorithms. Indeed, we show in this work how an IMU from a smart device can help a face tracker to measure pupil distance, and an SfM algorithm to measure the metric size of objects. We also identify motions that produce better results and, using a high frame rate camera, gain insight to how the performance of our method is affected by the quality of the tracking output.