【作者】 潘志国；

【导师】 罗光春； 段琢华；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2018， 硕士

【摘要】 随着传感器技术和人工智能技术的发展,智能移动机器人逐渐进入到使用阶段,代替人类进行重复劳动以及危险工作。而在机器人众多研究方向中有关移动机器人同时定位与建图(Simultaneous Localization And Mapping,SLAM)问题的解决是移动机器人实现完全自主工作的重要一环。SLAM主要应用场景是在陌生环境中,机器人最初没有周围环境的信息,这时就需要机器人通过安装在本体上的传感器获得周围环境和自身状态的信息,然后对信息进行分析加工,从而得到陌生环境的地图以及自己所在地图的位置。但是单一的传感器误差较大,由此本文进行了基于多传感器融合的SLAM算法的研究与应用。双目相机可以通过立体匹配恢复出空间路标点的深度信息,相比于激光等其它传感器,鲁棒性好获取信息更加丰富。惯性测量单元(Inertial Measurement Unit,IMU)在较短时间内精度较高。通过结合IMU和双目相机的优点,本文提出了一种融合IMU和双目相机的改进FastSLAM算法。首先,研究了双目立体视觉工作原理。通过实验分析比较了立体匹配常用的特征算法,决定选用ORB(Oriented FAST and Rotated BRIEF)算法。本文对于ORB算法的提取和误匹配筛选做了改进。针对光线不好时图像噪声较多影响特征点的采集和匹配的问题,在采集特征点之前先通过对比度被限制的直方图均衡化方法对图像进行预处理,减少噪声的影响。针对SLAM系统对于特征点匹配精度要求较高的问题,通过结合汉明距离设定阈值方法,随机采样一致性算法以及匹配对间距离比和角度比误匹配筛选算法进行误匹配剔除,明显减少了误匹配的数量。然后,分析了SLAM算法的运动过程模型和观测模型。针对现有的基于粒子群改进的FastSLAM算法可能会使采样粒子丧失多样性,使粒子向局部最优解聚集的问题,在已有的粒子群FastSLAM算法的基础上做出改进。通过引入免疫算法增加粒子的多样性,改善了FastSLAM在迭代过程中粒子退化问题和由于多次重采样造成的粒子耗尽问题。最后,在深入研究改进的SLAM算法的基础上,实现了双目相机和IMU融合的SLAM系统。该系统是在机器人操作系统(Robot Operating System,ROS)下开发的,并且进行了系统测试,对于改进的Fast SLAM算法的有效性进行了验证。更多还原

【Abstract】 With the development of sensor technology and artificial intelligence technology,intelligent mobile Robots gradually replace humans for repeated work and dangerous work.In many research fields of robots,the problem of Simultaneous Localization And Mapping(SLAM)is the most important part of mobile robot’s complete autonomous work.The SLAM problem can be described as: In an unknown environment,the mobile robot determines the environmental map and its position in the map by the sensors it carries.However,the single sensor has a large error.This thesis implements SLAM algorithm based on multi-sensor fusion.The binocular camera can recover the depth information of the spatial landmarks through stereo matching.Compared with other sensors such as laser,the binocular camera has better robustness and richer information.The inertial measurement unit(Inertial Measurement Unit,IMU)has a higher accuracy in a shorter time.By combining the advantages of IMU and binocular camera,an improved FastSLAM algorithm for fusion of IMU and binocular cameras is proposed in this thesis.Firstly,This thesis studies the principle of binocular stereovision.This thesis compares the commonly used feature algorithms of stereo matching and finally the ORB(Oriented FAST and Rotated Brief)algorithm is chosen as the feature point extraction algorithm.This thesis improves the extraction of ORB algorithm and reduces false matching.This thesis improves the extraction of ORB algorithm and reduces false matching.Before the feature points are acquired,the image is preprocessed by the histogram equalization method with limited contrast,which reduces the influence of noise.This paper uses the Hamming distance setting threshold method,Random Sample Consensus(RANSAC)and matching pair distance ratio and angle ratio mis-matching screening algorithm to eliminate false matching,which obviously reduces the number of mis-match.Secondly,the motion process model and the observation model of the SLAM algorithm are analyzed.The FastSLAM algorithm based on particle swarm optimization may make the sampling particles lose diversity and make the particle swarm to the local optimal solution.So this thesis makes improvements based on existing particle swarm FastSLAM algorithm.By introducing immune algorithm to increase mutation and multiplication of particles and increase particle diversity,the problem of particle degradation during Fast SLAM iteration and particle exhaustion caused by multiple resampling is improved.Finally,based on the improved FastSLAM optimization algorithm,the SLAM system with binocular camera and IMU is realized in this thesis.And this system was developed under the Robot Operating System(ROS).This system was tested in a realworld scenario and the effectiveness of the improved FastSLAM algorithm was verified.更多还原