无人机软管式自主空中加油视觉导航技术

近年来,机器视觉技术的迅速发展给无人机自主空中加油提供了一种新思路,克服了传统GPS动态适应能力差的不足。基于双目视觉,提出了一种实用的无人机自主空中加油导航方案,并提出了一种利用深度学习进行检测跟踪配合几何约束进行位姿计算的技术。根据加油机和受油机距离的不同,将导航分为远端导航和近端导航。全程对加油锥套进行检测和跟踪,并在近端进行锥套位姿计算。实验结果表明,该算法满足加油的鲁棒性要求,稳定跟踪锁定目标距离范围大于40m。当与目标距离小于10m时,距离误差优于±10cm。说明本文提出的方法具有较强的应用前景。     

RTK辅助的单目视觉地图构建与恢复方法

针对单目视觉SLAM在复杂场景下跟踪失败导致地图丢失的问题,提出了一种利用RTK信息辅助单目视觉的自动化地图构建与恢复方法。在视觉跟踪失败后,设计了一种自动化恢复地图构建方法,基于RTK信息的辅助,采用坐标变换算法处理丢失前的地图并将其融合到当前地图构建中,在提高建图效率的同时,最大程度地减少了地图信息的缺失。实验仿真结果表明,该方法可以解决系统无法继续跟踪建图的问题,与原始地图相比,其融合后构建的地图在保证全局一致性和完整性的基础上,可以保持米级的估计精度。     

基于红外视觉/激光雷达融合的目标识别与定位方法

近年来,基于可见光图像的目标识别在无人车感知领域得到了广泛应用.然而,可见光图像目标识别无法应用于弱光和黑暗环境.针对于此,提出了一种基于红外视觉/激光雷达融合的目标识别与定位算法.首先,通过基于颜色迁移的数据增强训练方法,提高了红外目标识别算法的泛化性能.继而,提出了一种基于激光雷达修正的单目深度估计方法,通过视觉图...     

Monocular Vision-based Two-stage Iterative Algorithm for Relative Position and Attitude Estimation of Docking Spacecraft

Visual sensors are used to measure the relative state of the chaser spacecraft to the target spacecraft during close range rendezvous phases. This article proposes a two-stage iterative algorithm based on an inverse projection ray approach to address the relative position and attitude estimation by using feature points and monocular vision. It consists of two stages: absolute orientation and depth recovery. In the first stage, Umeyama's algorithm is used to fit the three-dimensional (3D) model set and estimate the 3D point set while in the second stage, the depths of the observed feature points are estimated. This procedure is repeated until the result converges. Moreover, the effectiveness and convergence of the proposed algorithm are verified through theoretical analysis and mathematical simulation.     

Drogue detection for autonomous aerial refueling based on convolutional neural networks

Drogue detection is a fundamental issue during the close docking phase of autonomous aerial refueling(AAR). To cope with this issue, a novel and effective method based on deep learning with convolutional neural networks(CNNs) is proposed. In order to ensure its robustness and wide application, a deep learning dataset of images was prepared by utilizing real data of ‘‘Probe and Drogue" aerial refueling, which contains diverse drogues in various environmental conditions without artificial features placed on the drogues. By employing deep learning ideas and graphics processing units(GPUs), a model for drogue detection using a Caffe deep learning framework with CNNs was designed to ensure the method's accuracy and real-time performance. Experiments were conducted to demonstrate the effectiveness of the proposed method, and results based on real AAR data compare its performance to other methods, validating the accuracy, speed, and robustness of its drogue detection ability.     

单目相机物体位姿估计方法研究

现有的机器视觉通常以边缘轮廓和角点作为特征,因此要求背景单一,对环境结构化依赖程度高。为了拓展机器人的应用范围,使其脱离结构化的环境,提出了一种基于SIFT特征点和PNP技术的单目相机估计目标物体位姿的方法。以BumbleBee双目相机为硬件基础,以C++为开发平台,结合了Eigen计算库、OpenCV图像处理库和Triclops库,开发了单目视觉位姿估计算法,实现在复杂背景下对表面纹理较为丰富的物体的位姿估计。利用试验对所提方法进行了验证,试验结果表明,该算法具有较高的估计精度,可以作为机器抓取的依据。     

基于迭代算法的复杂曲面零件三坐标测量快速精确定位方法

针对复杂曲面零件在三坐标测量过程中存在着定位难、定位精度低的问题,基于改进的迭代最近点算法,提出了考虑半径补偿的预定位与精定位的多级定位算法;并通过建立定位点几何误差与叶片定位精度的数学关系模型,结合贪婪算法与六点原则,给出了近优的定位点序列生成方法。结合定位算法与定位点序列,给出了复杂曲面零件定位迭代流程,并开发了三坐标测量定位系统。以涡轮叶片为例,通过的定位仿真和定位实验,结果表明:该系统只需测量12～18个点即可使得叶片定位误差在0.1mm以内,证明该系统能有效的提高复杂曲面零件的定位精度和效率。      

Modeling and simulation of bow wave effect in probe and drogue aerial refueling

In a probe and drogue aerial refueling system, the bow wave of the receiver aircraft will produce a strong aerodynamic effect on the drogue once the receiver follows the drogue at a close distance. It is a major difficulty of docking control in the probe and drogue refueling. This paper analyses the bow wave effect and presents a simple method to model it. Firstly, the inviscid flow around the receiver is modeled based on the stream function defined by basic stream singularities.Secondly, a correction function is developed to eliminate the error caused by the absence of air viscosity. Then, the aerodynamic coefficients are used to calculate the induced aerodynamic force on the drogue. The obtained model is in an analytical form that can be easily applied to the controller design and the real-time simulations. In the verification part, computational fluid dynamics(CFD)simulation tests are conducted to validate the obtained flow fields and aerodynamic forces. Finally,the modeling method is applied to an F-16 receiver aircraft in a previously developed autonomous aerial refueling simulation system. The simulations results are analyzed and compared with the NASA flight-test data, which demonstrates the effectiveness of the proposed method.     

单目主动视觉无人机导引中摄像机内参数标定的线性方法

提出一种单目主动视觉无人机导引中摄像机内参数标定的线性方法,在摄像机主点位置进行预标定的条件下,摄像机只需拍摄一幅着舰平面靶标上的一组正方形图像,无须知道该组正方形的任何几何信息,通过计算圆环点,建立绝对二次曲线对摄像机内参数的约束方程即可线性求解摄像机内参数,仿真实验和真实图象实验表明:该方法在摄像机内参数(如焦距等)需要经常改变的视觉任务,如基于视觉的无人机着舰导引、运动参数估计、三维视觉重建及视觉监控中,使得摄像机内参数的标定具有简便、实时和鲁棒性强的特点。     

无人车基于激光雷达/视觉的目标体积自动测量方法

近些年,基于激光雷达和视觉的目标感知在无人系统中得到了广泛应用。目标的体积测量在很多应用场景可以发挥极其重要的作用,然而对识别感知目标的体积测量,目前尚无大量研究。首次提出了一种基于激光雷达/视觉的无人车目标体积自动测量方法,实现了无人车与目标体积测量功能的结合。通过在LeGO-LOAM算法中加入点云畸变补偿,相较于原始LeGO-LOAM算法,无人车在高速情况下的构图精度得到提升;通过将激光雷达与视觉进行深度融合,实现了目标的自动识别与全局定位;通过基于平面拟合的地面分割与欧式聚类,实现了目标点云轮廓的实时获取;通过设计一种基于切片法的不规则物体体积测量方法,实现了无人车在运动情况下对目标体积的自动估计。最终,分别通过Gazebo仿真和实际试验验证了算法的有效性。试验结果表明,所提算法在无人车运动的情况下对静态目标物的实时体积测量精度优于3%,具有较好的工程应用价值。     

基于平面模板自由拍摄的双目立体测量系统的现场标定

 摄像机标定是基于光学摄像的立体测量技术中的一个十分重要的步骤，标定精度直接影响系统测量的精度和稳定性。本文提出并实现了一种简便易行的双目立体测量系统现场标定方法：利用一块特殊设计的具有不同大小圆形特征点的平面标定板，在无需控制任何运动参数的情况下，双目测量系统只需对标定板在不同角度自由拍摄一组图像即可方便地实现系统标定。该方法在已有单摄像机标定算法的基础上，加入对双摄像机相对位置和姿态的优化，同时考虑了镜头的非线性畸变，达到了较好的标定结果。另外，提出的图像点与其空间点的对应算法具有良好的稳定性，即使由于环境或摄像机摆放位置等因素的影响，一些标定板上的特征点不能被摄像机拍摄到，或者不能被正确识别时，仍然能进行标定工作。标定实验和系统标定后的三维重建结果验证了该方法的有效性。     

结构光三维视觉系统研究

介绍了结构光三维视觉原理,在广义坐标系中采用投影和透视变换理论建立了结构光三维视觉模型。在此基础上,提出了基于非线性最小二乘法的模型参数标定方法,并建立了结构光三维视觉系统,给出了实验结果。     

基于Faster-RCNN 的机场跑道异物识别定位技术研究

针对机场跑道异物(FOD) 识别检测，分析了机场跑道环境特性，建立了针对机场FOD图像识别数据库， 提出了一种基于Faster-RCNN 的机场跑道异物识别定位系统，通过无人机系统对机场跑道进行图像采集，运用 Faster-RCNN 算法框架完成异物种类识别，并结合无人机地理位置信息进行位置换算求解，得到机场FOD 的 类别信息和位置信息。经验证，该系统可高效完成机场FOD 检测识别及定位任务。     

基于单目视觉的机器人避障方法研究

避障问题一直是机器人领域研究的热点和重点。设计一种基于单目视觉的避障系统,能够同时检测机器人正前方和两侧的障碍物,且对低纹理障碍物同样具有效果。该系统采用特征尺度检测器完成机器人视野正前方障碍物的检测,对于视野两侧的障碍物运用光流平衡策略进行规避,而对于低纹理的障碍物则利用图像熵的方法进行检测。整个系统在地面移动机器人上进行了测试,实验结果表明,该方法能在复杂的环境下可靠地完成机器人的避障任务。     

基于合作目标和视觉的无人飞行器全天候自动着陆导引关键技术

 GPS受他国控制和易受干扰对基于该导航系统的无人飞行器导航和着陆具有潜在危险,为了使其安全自动着陆,提出通过在着陆跑道(广场或公路)上预设具有受控发射红外光的地面合作目标,实现无人飞行器的视觉全天候自动精确着陆新技术。首先研究得到切实可行的工作过程;然后研制了合作目标,通过理论分析和实验得出,无论能见度高低,8~12 μm中远红外视觉系统探测到的合作目标图像质量远优于可见光视觉系统,研究得到合作目标和背景的温差在170~200 ℃时,成像效果最好。在识别研究中发现,基于传统的不变矩算法对合作目标的识别可靠性稍差,为了进一步准确地识别合作目标,又提出了基于方向链码的合作目标识别算法,该方法的耗时为不变距的96%,但是识别的可靠性比前者有显著提高。     

基于视觉的大尺度部件相对位姿实时测量方法研究

相对位姿是装配过程中的一项重要监控项。针对大尺度部件对接过程中的相对位姿测量需求,提出了一种基于视觉的相对位姿实时测量方法。该方法利用单目视觉技术,通过采集合作靶标的图像,实时解算大部段间的相对位姿,用于辅助大尺度部件的装配。首先,设计了一套相对位姿实时测量系统,包括搭载单目相机的视觉测量单元,以及用于辅助位姿解算的合作靶标;其次,对相对位姿测量的完整流程进行了研究,包含系统标定方法与实时位姿解算方法;最后,在实验室环境下对位姿测量系统的精度进行测试。试验结果表明,位姿测量系统在垂直于光轴方向的重复精度可达0.02mm,沿光轴方向重复精度优于0.2mm,输出位姿结果时间低于0.3s;对多测量单元组网测量进行了仿真计算,垂直于光轴方向的重复精度优于0.1mm,沿光轴方向重复精度优于0.2mm,输出位姿结果时间优于1.3s。试验结果表明,提出的方法可满足一般大尺度部件对接过程实时位姿监控与对接状态评估的需求。     

A mask R-CNN based method for inspecting cable brackets in aircraft

In the aviation industry, cable bracket is one of the most common parts. The traditional assembly state inspection method of cable bracket is to manually compare by viewing 3D models. The purpose of this paper is to address the problem of inefficiency of traditional inspection method. In order to solve the problem that machine learning algorithm requires large dataset and manually labeling of dataset is a laborious and time-consuming task, a simulation platform is developed to automatically generate synthetic realistic brackets images with pixel-level annotations based on 3D digital mock-up. In order to obtain accurate shapes of brackets from 2D image, a brackets recognizer based on Mask R-CNN is trained. In addition, a semi-automatic cable bracket inspection method is proposed. With this method, the inspector can easily obtain the inspection result only by taking a picture with a portable device, such as augmented reality (AR) glasses. The inspection task will be automatically executed via bracket recognition and matching. The experimental result shows that the proposed method for automatically labeling dataset is valid and the proposed cable bracket inspection method can effectively inspect cable bracket in the aircraft. Finally, a prototype system based on client-server framework has been developed for validation purpose.     

A wall-thickness compensation strategy for wax pattern of hollow turbine blade

As an important index affecting the aerodynamic performance and the structural strength of hollow turbine blades, the wall-thickness precision of the blade is mainly inherited from the positional relationship between the corresponding wax pattern and the internal ceramic core. However, due to locating errors, the actual position of ceramic core is always deviated from the ideal position, which makes it difficult to guarantee the wall-thickness precision of the wax pattern. To solve this problem, a wall-thickness compensation strategy is proposed in this paper. Firstly, based on the industrial computed tomography (ICT) technique and curve matching algorithms, a model reconstruction method is developed, with which the 3D model of a trial wax pattern can be easily constructed. After that, focusing on eliminating the wall-thickness errors of the trial wax pattern, an optimization method for the pose of the ceramic core in the wax pattern is proposed. Then, by mapping the optimal pose of the ceramic core to length adjustments of the locating rods, the wall-thickness errors of the wax pattern can be greatly reduced. A case study is also given to illustrate the effectiveness of the proposed compensation strategy.     

基于布谷鸟搜索算法的一类变体飞行器容错控制

针对一类存在执行机构故障的分布式结构变体飞行器的控制分配问题,结合整数规划理论,提出一种基于布谷鸟搜索算法的容错控制方法。首先,设计虚拟控制指令,使得系统状态能够很好地跟踪参考模型;然后,将执行器概率性故障与饱和约束转换为整数规划问题中决策变量的约束,从而将执行器控制分配问题转化为一类整数规划问题;最后,采用改进的布谷鸟搜索算法进行求解,得到实际的执行器控制分配指令。仿真结果表明,在执行器存在概率性故障的情况下,该容错控制方法较无容错策略的情况能够有效提升系统的跟踪性能;与遗传算法相比,该算法得到的执行器控制分配结果更加精确。     

仿射坐标系下多视点的三维物体识别方法研究

多个视角下三维物体的识别一直是计算机视觉和模式识别领域的重要研究方向。对平面区域或适当的变换区域中以不同视角观察时图像中的点集所具备的性质进行了研究．并得出了在仿射变换下点模式匹配所具备的重要性质，即基于不同视角下平面图形的傅立叶描绘矩阵，得到形状匹配的数学约束，并将其应用于多个视角下物体的形状识别。理论分析及仿真实验表明，该方法在形状识别中是简便、快速、有效的。