Laser beams target-based monocular vision measurement system self-calibration and localization method for boom-type roadheader in large-scale underground complex scene

Abstract

The monocular vision measurement system is an effective approach to obtain accuracy and reliable pose estimation for underground tunneling, especially with a specially designed laser beams targets for the far-distance and large-scale scene localization in the tunneling face. However, the conventional manual off-line calibration with a total station is complex and difficulty resulting unsuitable for the automatic underground excavation. This article presents a novel online self-calibration method for the monocular vision measurement system based on the double laser beams targets, in which one is as the benchmark target, the other is as the movable target. The double 3D targets-based monocular vision geometric constraint model is established to connect the 3D targets spatial distribution and the laser beams image points and lines features, considering explicitly the 2D-3D correspondences between laser beams image and object coordinates of the double 3D targets. On this basis, we perform the extrinsic parameters self-calibration to reconstruct the laser beams spatial vector of the movable target from its multi-view image sequences. With no need to gain explicit knowledge of object space coordinates, the movable target multi-station online self-calibration can be performed with known initial benchmark target as well as the multi-view image sequences. The experimental results demonstrated the proposed self-calibration approach to be effective and the accuracy of the fuselage pose estimation has increased by 67.7%, meeting the far-distance localization requirements for underground excavations in coal mine.