基于多信息融合与神经网络的矿区目标体重建方法研究  

Research on Reconstruction Method of Mining Area Target Body Based on Multi-information Fusion and Neural Network

作　　者：史律[1] 李建林[1] SHI Lv;LI Jianlin(Nanjing Vocational College of Information Technology,Nanjing,Jiangsu 210023,China)

机构地区：[1]南京信息职业技术学院,江苏南京210023

出　　处：《矿业研究与开发》2020年第8期150-154,共5页Mining Research and Development

基　　金：江苏省“333工程”高层次人才培养科研资助项目(BRA2019303);江苏省高校“青蓝工程”优秀教学团队资助项目(苏教师[2019]3号).

摘　　要：为提高矿区目标体的重建效果,提出了一种将点云数据与图片进行融合的重建方法。通过MATLAB获得目标体的点云数据,对目标体的点云数据进行拼接以及表面重建。以清晰程度、纹理复杂度以及曝光率作为评价图片质量的3个因素,利用BP神经网络对目标体图片的质量进行分级,在卷积神经网络的基础上搭建BP神经网络对目标体图片中的部分进行识别。最终通过融合激光雷达、单目相机所获得的点云数据与图片来实现目标体的重建。结果显示,BP神经网络的分级准确率为96.39%,BP神经网络的最低识别率为96.4%,通过融合后重建目标体的误差小于±0.12m。研究结果为目标体的定位、矿区救援提供了参考依据。To improve the reconstruction effect of the mining area target body,a reconstruction method based on the fusion of point cloud data and images was proposed.MATLAB was used to obtain the point cloud data of target body for splicing and surface reconstruction.The clear degree,texture complexity and exposure rate were taken as the three factors for the evaluation on the quality of the image,and the BP neural network was used to classify the quality of target body image.On the basis of convolution neural network,BP neural network was used to recognize a part of the object image.Finally,through the fusion of point cloud data and image obtained by lidar and the monocular camera,the reconstruction of the target body was realized.Results show that the classification accuracy of BP neural network is 96.39%,the minimum recognition rate of BP neural network is 96.4%,the reconstruction error after fusion is less than±0.12 m.The research results can provide theoretical reference and basis for the positioning and rescue of target body.

关 键 词：矿区目标体 信息融合 重建方法 BP神经网络 激光雷达 

分 类 号：TD73[矿业工程—矿井通风与安全] TD315