Research On Vision-based Pose Estimation Methods And System For Intelligent Tunnel Machine In Coal Mine

The construction of intelligent working face in coal mine is of great significance to reduce the number of workers and ensure safe production.With the continuous improvement of the intelligence level of fully mechanized coal face,the contradiction of unbalance of mining is prominent,and the intelligence and high efficiency of tunneling is imminent,which poses a severe challenge.The development of intelligent mining technology is seriously restricted by the dynamic positioning technology of underground equipment in coal mine.Vision measurement has been widely used in many fields such as dynamic target navigation,tracking and control in recent years,especially it has been rapidly developed in the underground coal mine.The non-contact pose measurement of dynamic targets with the help of visual perception and intelligent algorithm has become a research hotspot in this field at home and abroad.In view of the technical difficulties of dynamic target pose measurement on the tunneling working face in coal mine,this dissertation independently developed a monocular vision-based pose measurement system for the the tunneling equipment fuselage and cutting head.In order to improve the intelligence level of tunneling working face,the underground camera calibration model that considered the explosion-proof glasses refraction is established,the error modeling and compensation technology for the laser point-line target measurement system is investigated,the underground camera blur modeling and deblurring under vibration condition is researched,and the pose estiamtion performance evaluation method is also studied.The main research work of this dissertation includes:Image distortion caused by the two-layer planar or spherical explosion-proof glasses seriously affect the accuracy of the vision measurement in coal mine.This dissertation presents a novel geometrically-driven underground camera modeling and calibration algorithm.The underground camera model is established under coplanarity constraints,and deriving a correspondences between the image points and three-dimensional object coordinates.The refraction calibration algorithm is also proposed to realize the refraction calibration for underground camera,which effectively improving the accuracy and reliability of the vision measurement system in coal mine.In addition,a non-single viewpoint refractive calibration algorithms for cameras with an optical sphere cover is proposed.It introduces a virtual unified optical axis and make it possible to solve the severe non-linear distortion that caused by the spherical refraction,which improve the the effectiveness and robustness of vision measurement system.The equipment positioning in tunneling face is difficulty and required to be with high accuracy.However,the existing methods have accumulated errors and poor adaptability.This dissertation presents a laser beams target-based localization methods for the machine body of Boom-type roadheader.Taken the parallel laser beams that produced by the mine laser pointing instrumentas as the target,the underground laser beams target image segmentation,and feature extraction are presents,which can tackle the dense-dust and low illumination environment.A2P3L(Two-Points-Three-Lines)localization model from line correspondences is developed to obtain the global localization of the roadheader relative to the roadway.The error transfer function model of the fuselage pose measurement system is established to obtain the error propagation rule of the fuselage pose measurement system,which improve the applicability and accuracy of dynamic target detection in harsh environment.Aiming at the problems of complex structure and high failure rate of the traditional cuttinghead pose measurement systems for Boom-type roadheader,a monocular vision-based noncontact measurement method for cutting-head pose estimation is presented.Taken multi-point infrared LED on the boom-type roadheader as the target,a fast spot detection and center location algorithm based on fixed-coefficient Gaussian surface fitting method is studied.According to the space projection geometry constraint conditions.A spatial coordinate calculation model for the feature points based on the improved P4 P model is established,which can be used to realize the local localization of underground driving equipment or components,which solves the problem of pose measurement under high dust and partial LED occlusion.The error transfer function model is established to master the error propagation rule of the measurement system under driving conditions,which provides theoretical basis for improving the performance of the vision measurement system of the cutting head.The blurred image caused by the vibration interference under driving condition will lead to the pose estimation with big error and even the measurement failure,it seriously affect the stability of the visual measurement system.The iterative blind restoration algorithm based on the non-uniform blur kernel is studied to realize the single image blind deblurring for underground camera.A parameterized underground camera non-uniform blur model is established,which describe the dynamic imaging process of a blurred image caused by the roadheader's motion and vibration.The variational equation is established to evaluate and optimize the parameter distribution,and the iterative blind restoration method is studied to realize the single image blind deblurring based on non-uniform blur kernel.It is beneficial to obtain the stable and clear target image,and the accuracy of pose vision measurement system is further guaranteed.Finally,an experimental platform of roadheader pose visual measurement system is established,which is used to verify the theoretical research results,as well as the accuracy and reliability of the system.The experimental results show that the system can realize the positioning,orientation,and fixed cutting locus for tunneling equipment under the influence of explosion-proof glass,vibration,dust and water mist,it effectively solve the problems of low illumination,high dust,strong light interference and shielding in coal mine.The research is of great application value to solve the problem of pose estimation of dynamic targets in coal mine,and it is of great significance to promote the intelligentization of mining face.