基于雷达和视觉复合传感器的无人机障碍物检测研究

发布时间：2019-06-14 09:12

【摘要】：具有自主飞行能力的无人机由于其成本低,机动性能好,效费比好,生存能力强,无人员伤亡风险等优点,在军事、民用以及科学研究中均具有重要的应用价值。然而现有的无人机缺乏自主探测及躲避障碍物的能力,在可视范围以内可以通过人为的遥控飞行,可视范围以外障碍物会对无人机飞行产生安全威胁。因此需要构建一套具有感知能力,并使无人机能够实现障碍物自主检测的方法。本文提出了基于毫米波雷达和视觉传感器融合进行无人机障碍物的检测的方法。利用毫米波雷达获取前方障碍物的距离、角度等信息,根据毫米波雷达获取的信息和图像的底层颜色信息在视频图像上建立起障碍物的感兴趣区域,然后利用SURF(SpeededUp Robust Feature)算法对感兴趣区域进行验证,判断是否为建筑物等障碍物。主要研究内容如下:1.建立雷达坐标和图像坐标转换关系。根据无人机平台高度、姿态变化的特点,雷达和图像坐标转换关系需要满足对其变化具有自适应性。传统的毫米波雷达和视觉传感器融合通常基于二维运动平台的假设,对平台的姿态变化非常敏感。本文基于相机的针孔模型推导并建立与融合系统的姿态和高度关联的雷达坐标和图像坐标转换关系,并通过融合IMU的姿态数据和差分GPS的高度数据精确计算出目标障碍物的图像坐标。2.障碍物候选区域分割。由于毫米波雷达获取的为环境的稀疏信息,不能得到障碍物的全面位置信息。因此根据雷达反射点的图像坐标和图像颜色特征进行障碍物的候选区域分割。3.基于SURF算法的障碍物候选区域判别。通过提取候选区域的SURF特征关键点与障碍物样本的关键点数据库匹配来实现障碍物的分类识别。4.融合系统开发及实验验证。利用VS2008和Opencv 2.43开发了基于毫米波雷达、视觉传感器、IMU(Inertial Measurement Unit)和 GPS(Global Position System)多传感器融合的无人机障碍物检测系统。系统运行环境为研华Celeron 2.3GHz处理器、2G内存的单板计算机,多传感器融合障碍物检测系统搭在无人机上进行低空障碍物检测实验。实验结果表明,该方法可以自适应无人机的姿态变化以及能够快速、准确地实现无人机障碍物的在线检测,具有较好的实时性和准确性。
[Abstract]:UAV with autonomous flight ability has important application value in military, civil and scientific research because of its low cost, good maneuverability, good efficiency-cost ratio, strong survivability, no risk of casualties and so on. However, the existing UAV lacks the ability to detect and avoid obstacles independently, and can fly by man-made remote control within the visible range, and obstacles outside the visible range will pose a safety threat to UAV flight. Therefore, it is necessary to construct a set of perceptual methods, which can enable UAV to detect obstacles autonomously. In this paper, a method of UAV obstacle detection based on millimeter wave radar and vision sensor fusion is proposed. The millimeter wave radar is used to obtain the distance and angle of the obstacle in front of the video image, and according to the information obtained by the millimeter wave radar and the color information of the bottom of the image, the region of interest of the obstacle is established on the video image, and then the region of interest is verified by SURF (SpeededUp Robust Feature) algorithm to determine whether it is an obstacle such as a building. The main research contents are as follows: 1. The transformation relationship between radar coordinates and image coordinates is established. According to the characteristics of UAV platform height and attitude change, the coordinate conversion relationship between radar and image needs to be adaptive to the change of UAV platform. The fusion of traditional millimeter wave radar and visual sensor is usually based on the assumption of two-dimensional motion platform and is very sensitive to the attitude change of the platform. In this paper, based on the pinhole model of the camera, the transformation relationship between radar coordinates and image coordinates related to the attitude and height of the fusion system is derived and established, and the image coordinates of the target obstacle are calculated accurately by combining the attitude data of IMU and the height data of differential GPS. 2. Obstacle candidate region segmentation. Due to the sparse information obtained by millimeter wave radar for the environment, the comprehensive position information of obstacles can not be obtained. Therefore, according to the image coordinates and image color characteristics of radar reflection points, the candidate regions of obstacles are segmented. Obstacle candidate region discrimination based on SURF algorithm. The classification and recognition of obstacles is realized by extracting the SURF feature key points of the candidate region and matching the key points of the obstacle samples. 4. Development and experimental verification of fusion system. An obstacle detection system for UAV based on millimeter wave radar, vision sensor, IMU (Inertial Measurement Unit) and GPS (Global Position System) is developed by using VS2008 and Opencv 2.43. The running environment of the system is Yanhua Celeron 2.3GHz processor, 2G memory single board computer, multi-sensor fusion obstacle detection system is built on UAV to carry out low altitude obstacle detection experiment. The experimental results show that the method can adapt to the attitude change of UAV and realize the on-line detection of UAV obstacles quickly and accurately, and has good real-time and accuracy.
【学位授予单位】：沈阳理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：V279;TP391.41

【参考文献】