基于四旋翼平台的融合单目视觉与惯性传感的里程计方法研究

发布时间：2018-01-17 19:10

本文关键词：基于四旋翼平台的融合单目视觉与惯性传感的里程计方法研究　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：本论文关注融合单目视觉和惯性传感的里程计方法研究,旨在实现能在四旋翼飞行器平台上实时运算,并且具备高精度和鲁棒性的自主定位算法,主要分为两个场景。首先在笔者参加的国际空中机器人大赛(the International Aerial Robotics Competition,IARC)中要求飞行器在不依赖外界辅助导航的条件下实现自主定位。针对该比赛的特殊场景(地面具有丰富纹理信息和规则网格特征),提出了一种基于光流法和网格信息,同时融合惯性测量单元(Inertial measurement unit,IMU)的定位方法。.首先在传统光流法上作了改进,基于固定块匹配方法,实时获取相机平动速度,然后将其积分作为初始估计,并通过地面网格信息来校正光流积分得到的位置信息,最后融合IMU数据进行信号平滑,确保飞行器位姿无累积误差。该方法在2016年IARC亚太赛区比赛中得到成功应用。其次针对一般场景,选择基于稀疏直接法的单目视觉里程计算法,无需计算每帧图像的特征描述子,计算速率提高,并且设计了模块化的扩展卡尔曼滤波(extend Kalman Filter,EKF)框架,融合单目视觉里程计计算得到的相机位姿和IMU数据。对于预测部分,基于IMU驱动系统的误差状态运动学实现,对于测量部分,由视觉里程计提供的位置和姿态作为量测,另外进行了测量量和状态量的时间同步处理,以及视觉算法位姿检测失败时的校正处理。在开源数据集测试本算法的准确性和鲁棒性。
[Abstract]:In this paper, we focus on the research of odometer based on monocular vision and inertial sensing, aiming to achieve real-time computing on the platform of four-rotors, and have high accuracy and robustness of autonomous positioning algorithm. It is divided into two main scenes. Firstly, I participated in the International Air Robot Competition (. The International Aerial Robotics Competition. IARC requires the aircraft to achieve autonomous positioning without relying on external navigation. The special scene of the game (ground has rich texture information and regular grid features). Based on optical flow method and grid information, an inertial measurement unit is proposed. Firstly, the traditional optical flow method is improved. Based on the fixed block matching method, the camera translational velocity is obtained in real time, and then the integral is used as the initial estimation. And through the ground grid information to correct the optical flow integration of the position information, finally fusion of IMU data for signal smoothing. This method has been successfully applied in the 2016 IARC Asia Pacific Competition. Secondly, for general scenarios, a single vision mileage calculation method based on sparse direct method is selected. The computation rate is improved without calculating the feature descriptor of each frame image, and a modular extended Kalman filter extend Kalman filter (EKF) framework is designed. For the prediction part, the error state kinematics of the IMU drive system is realized, and the measurement part. The position and attitude provided by the vision odometer are used as the measurements, and the time synchronization of the measurement and the state is also carried out. The accuracy and robustness of this algorithm are tested in open source data sets.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【参考文献】