【作者】 张鹏；

【导师】 李瑞；

【作者基本信息】 电子科技大学 ， 控制科学与工程， 2019， 硕士

【摘要】 微型无人机具有轻小、便携等优点,可以在未知环境中替代人工进行探测并完成一些复杂且危险的任务,从而降低人员的伤亡风险。近年来,具有未知环境自主探索功能的微型无人机更是成为军民用无人机研究的新浪潮。本文以此为应用背景,重点研究了四旋翼无人机在室内环境的实时定位、环境三维重建和自主路径规划等关键技术,并搭建了一个能在室内自主导航的微型四旋翼平台。首先,根据四旋翼机载RGB-D摄像头实现前端视觉里程计:选用小孔成像相机模型,利用均匀化的ORB算法提取图像关键帧特征点并完成特征匹配,随后利用EPnp算法实现帧间初始位姿估计;其次,针对纯视觉里程计在快速移动和纯旋转时容易造成位姿估计出错的问题,采用视觉惯导相融合的方法,利用扩展卡尔曼滤波器(EKF)对视觉里程计数据和IMU数据进行松耦合,公式推导了扩展卡尔曼预测模型和观测模型,实现了视觉-惯导里程计;接着后端优化用图的方式构建全局和局部位姿图,选用BA算法优化位姿图,从而获得长时间的最优位姿轨迹和地图;随后针对大范围相机运动过程中产生的累积误差,利用Bow词袋法完成回环检测;同时根据场景需求生成四种不同用途的环境地图,进而完成整个视觉SLAM算法。最后根据获得的Octomap地图,利用改进的RRT~*随机生成树算法实现三维空间中的路径规划。为保证系统所用算法应用于四旋翼平台的可行性,同时降低飞行事故风险,本文首先在Gazebo下构建整套半实物仿真,完成SLAM算法、RRT~*算法和整个系统稳定性验证。最终搭建了以华硕RGB-D相机、Intel NUC处理器和Pixhawk飞控为主要部件的四旋翼飞行器硬件平台,并在室内通过大量实验测试了四旋翼姿态的稳定性、SLAM算法的实时性和路径规划的准确性。结果表明,所设计自主导航四旋翼系统能完成室内定位、环境三维感知、地图构建和路径规划等功能,具有良好的实时性、动态性及鲁棒性。更多还原

【Abstract】 The micro-UAV has the advantages of light and portable,it can replace human to detect the unknown environment and accomplish some complex and dangerous tasks,which can reduce the risk of casualties.In recent years,UAV with the function of autonomous exploration in unknown environment is a new wave of research on military and civilian drones.Based on this background,this paper focuses on the technologies of real-time positioning,reconstruction of environment and path planning,then we make up a micro-UAV platform for indoor autonomous navigation.First of all,this paper completes the visual odometer based on RGB-D camera,includes using pinhole camera model,and using ORB algorithm for the feature points of adjacent key frames extracting and matching,and the initial inter-frame pose estimation is realized by EPnP algorithm.Secondly,the extended Kalman filter(EKF)is used to loosely couple the visual odometer data and IMU data to solve the camera pose estimation error easily when camera moves fast and rotates purely.Then this paper derives the extended Kalman prediction model and the observation model though the formula,and on this basis realizes the visual-inertial odometer.And then the global and local pose maps are constructed by the back-end optimization,and the BA algorithm is used to get the posture trajectory and map for a long time.Furthermore,according to the accumulated errors in the process of large-scale camera movement,the BoW algorithm is used to complete loop detection.And according to the needs of the scene,four environmental maps for different purposes are generated.Finally,the visual SLAM algorithm is completed,and the improved RRT~* algorithm is used to realize path planning in Octomap map.Before physical debugging,the hardware-in-the-loop simulation of the system is carried out under gazebo,which guarantees the stability and function of all the algorithms as well as the four-rotor platform of the system and reduces the risk of crashing.In the simulation,the SLAM algorithm,RRT~* algorithm and the stability of the whole system are validated.In the end,the micro quadrotor hardware platform was built,consists of a ASUS RGB-D camera,a Intel NUC processor and a Pixhawk flight control.And then,the attitude stability,SLAM algorithm real-time and path planning test were carried out indoors.The experimental results show that the micro quadrotor system have good performance of timeliness,dynamic and robustness in indoor positioning,three-dimensional perception of the environment,map building and path planning.更多还原