动态环境下的实时高清大规模三维地形重建研究

项目名称： 动态环境下的实时高清大规模三维地形重建研究

项目编号： No.61503005

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 宋伟

作者单位： 北方工业大学

项目金额： 21万元

中文摘要： 三维地形重建是机器人环境感知的重要研究方向，本项目将开发一套基于多种传感器融合的实时高清大规模三维地形重建系统，利用局部三维点云、二维图像、惯性数据重建全局三维地形模型。该系统分为四大功能模块：1)多种传感器数据融合模块，基于最小二乘法估算出装载的激光雷达和摄像机间的投影矩阵，使局部的点云数据精确地映射到拍摄的视频流中；2)大规模数据加载模块，提出基于去重原理的点云和视频图像压缩技术，利用有限的内存空间加载大规模点云和高清的图像信息；3)三维前景背景分离模块，结合高度直方图、三维像素密度图、马尔科夫随机场模型和卡尔曼滤波器算法，将地形数据划分为可行走的地表、静态物体和动态移动物体三大类；4)高清三维地形重建模块，利用多种渲染方法，实现大规模地形在虚拟环境中的实时真实感展现。此外，本项目采用GPU编程方法对图形图像算法并行计算处理，以实现地形重建的实时性。

中文关键词： 多传感器信息融合；即时定位与地图构建；环境感知；三维背景建模；实时地形重建

英文摘要： 3D terrain reconstruction is an important researching area of environmental perception for mobile robots. This project develops a real-time high-resolution large-scale 3D terrain reconstruction system using multiple sensors integration. The system utilizes local 3D point clouds, 2D images, and inertial data to reconstruct a global terrain model. The system includes four functional modules: 1) Multiple Sensor Data Integration Module estimates the projection matrix between the mounted LiDAR and cameras based on least squares, so as to accurately map the local point clouds onto the captured video images. 2) Large-scale Datasets Registration Module proposes a compression method for point clouds and video images based on the redundancy removal principle. This method registers large-scale point clouds and high-resolution images using limited memory. 3) 3D Foreground/Background Segmentation Module integrates height histogram, 3D voxel dense map, Markov Random Field model, and Kalman filter algorithm to classify the terrain data into three classes, which are traversable ground, static object, and dynamic moving object. 4) High-resolution 3D terrain reconstruction Module utilizes multiple rendering techniques to realize real-time photorealistic representation of the large-scale terrain in a virtual environment. Furthermore, this project applies the GPU programming method to implements the applied graphics and image processing algorithms in parallel, which realizes real-time approach of terrain reconstruction.

英文关键词： Multiple Sensor Information Integration;SLAM;Environmental Perception;3D Background Modeling;Real-time Terrain Reconstruction