Metric Calibration of Aerial On-Board Multiple Non-overlapping Cameras Based on Visual and Inertial Measurement Data

Abstract

Recently, the on-board cameras of the unmanned aerial vehicles are widely used for remote sensing and active visual surveillance. Compared to a conventional single aerial on-board camera, the multi-camera system with limited or non-overlapping field of views (FoVs) could make full use the FoVs and would therefore capture more visual information simultaneously, benefiting various aerial vision applications. However, the lack of common FoVs makes it difficult to adopt conventional calibration approaches. In this paper, a metric calibration method for aerial on-board multiple non-overlapping cameras is proposed. Firstly, based on the visual consistency of a static scene, pixel correspondence among different frames obtained from the moving non-overlapping cameras are established and are utilized to estimate the relative poses via structure from motion. The extrinsic parameters of non-overlapping cameras is then computed up to an unknown scale. Secondly, by aligning the linear acceleration differentiated from visual estimated poses and that obtained from inertial measurements, the metric scale factor is estimated. Neither checkerboard nor calibration pattern is needed for the proposed method. Experiments of real aerial and industrial on-board non-overlapping cameras calibrations are conducted. The average rotational error is less than $0.2^{\circ }$, the average translational error is less than 0.015 m, which shows the accuracy of the proposed approach.

Supported in part by the National Natural Science Foundation of China under Grant 61902322, in part by the Doctoral Fund of Southwest University of Science and Technology under Grant 19zx7123, in part by the Open Fund of CAUC Key Laboratory of Civil Aviation Aircraft Airworthiness Certification Technology under Grant SH2020112706.