基于语义分割的室内动态场景同步定位与语义建图

针对动态物体在室内同步定位与地图构建（SLAM）系统中影响位姿估计的问题，提出一种动态场景下基于语义分割的SLAM系统。在相机捕获图像后，首先用PSPNet（Pyramid Scene Parsing Network）对图像进行语义分割；之后提取图像特征点，剔除分布在动态物体内的特征点，并用静态的特征点进行相机位姿估计；最后完成语义点云图和语义八叉树地图的构建。在公开数据集上的五个动态序列进行多次对比测试的结果表明，相对于使用SegNet网络的SLAM系统，所提系统的绝对轨迹误差的标准偏差有6.9%~89.8%的下降，平移和旋转漂移的标准偏差在高动态场景中的最佳效果也能分别提升73.61%和72.90%。结果表明，改进的系统能够显著减小动态场景下位姿估计的误差，准确地在动态场景中进行相机位姿估计。     

基于激光SLAM的移动机器人导航算法研究

定位技术是机器人技术中导航控制和路径规划的关键问题,传统定位方式采用全球定位系统(GPS),难以完成精准的定位导航功能,不依赖于GPS的定位导航方法是目前机器人领域的研究热点。提出一种基于激光雷达采集的点云信息帧间匹配方法,根据改进式激光点云数据的位姿估计算法,结合非线性优化进行了校正和优化,完成移动机器人对未知环境的精确定位。通过ROS机器人操作系统搭建实验平台,对改进算法进行验证,证明改进后帧间匹配算法的建图和定位效果对应的鲁棒性与定位精度效果更佳,可满足工程要求。     

Sensor fusion-based exploration in home environments using information,driving and localization gains

Exploration is one of the most important functions for a mobile service robot because a map is required to carry out various tasks. A suitable strategy is needed to efficiently explore an environment and to build an accurate map. This study proposed the use of several gains (information, driving, localization) that, if considered during exploration, can simultaneously improve the efficiency of the exploration process and quality of the resulting map. Considering the information and driving gains reduces behavior that leads a robot to explore a previously visited place, and thus the exploration distance is reduced. In addition, the robot can select a favorable path for localization by considering the localization gain during exploration, and the robot can estimate its pose more robustly than other methods that do not consider localizability during exploration. This proposed exploration method was verified by various experiments, which verified that a robot can build an accurate map fully autonomously and efficiently in various home environments using the proposed method.     

总配线架(MDF)与ADSL业务发展

随着ADSL宽带接入用户的发展及系统的扩容,总配线架(MDF)作为通讯内外线连接的重要区域,其配置空间资源紧张的问题日渐突出,本文就这一问题提出自己的认识。     

A smart camera with gesture recognition and SLAM capabilities for mobile robots

This paper presents a smart camera for a real-time gesture recognition system. The smart camera has been designed and implemented as a system-on-a-chip (SoC), using reconfigurable computing technology. In this system the gesture images are captured by means of a CMOS digital camera. After some pre-processing steps, those are sent to a fault tolerant module (FTM) for the actual recognition process. The FTM implements a RAM-based neural network, using three knowledge bases. In addition, a majority voting technique is used to improve the confidence level in the recognition step. A number of experiments using a prototype implementation and selected gestures resulted in a rate of 100% true recognition. The system also showed to be robust, and flexible as new gesture patterns can be easily included by means of the on-chip training capabilities. An application for this systems is also presented, consisting of four smart cameras used in simultaneous localization and mapping (SLAM) tasks for robotics.     

基于重投影深度差累积图与静态概率的动态RGB-D SLAM算法

为了提高同时定位与建图（SLAM）系统在动态场景下的定位精度和鲁棒性,提出新的RGB-D SLAM算法. 建立基于重投影深度差值的累积模型,分割图像的动静态区域;为了避免动态区域过分割,先剔除与匹配地图点欧氏距离过大的动态区域特征点,再根据t分布估计其余特征点的静态概率;将静态区域特征点和动态区域的疑似静态点以不同权重加入位姿优化,得到提纯后的位姿. 在公开数据集上的实验结果表明,所提算法在动态场景下较改进前的RGB-D ORB-SLAM2算法的定位精度提升96.1%,较其他动态SLAM算法提升31.2%,有效提高了视觉SLAM系统在动态环境下的定位精度和鲁棒性.     

A co-processor design to accelerate sequential monocular SLAM EKF process

Extended Kalman Filter (EKF) is a non-linear state estimation technique which is used to produce values that close to the true value when given with measurement containing noise and other inaccuracies. In Monocular Simultaneous Localization and Mapping (SLAM), EKF is used to estimate position and motion information. In this paper, Monocular SLAM software implementation on a general purpose computer is studied to find the most time consuming part of the estimation program. The analysis concentrates on the Monocular SLAM EKF estimation process which involves prediction, measurement prediction, matching and update. For this purpose, a form of dynamic programming analysis tool called software profiling is utilized to determine which section of the estimation program demands the highest processing time. Based on the analysis, it is found that EKF “matching” process contribute to the highest computation time. The reason behind the time-consuming process is because for every predicted feature in the matching stage, the acceptance region and their cross correlation have to be calculated. In a typical general purpose computer software implementation, the processing is limited to sequences of operations (i.e. sequential processing). Such implementation will delay the next process until the prior process completed. However, further analysis conducted in this paper shows that each feature does not depend on the prior process and can be processed individually. This would allow several features to be processed simultaneously to improve the execution speed. Therefore, an FPGA pipelined and parallel processing architecture is proposed.     

同步定位与建图技术发展综述

同步定位与建图(simultaneous localization and mapping, SLAM)技术作为智慧交通领域研究的热点, 是无人驾驶车辆自主规划路径的关键。围绕SLAM技术相关传感器类型、定位、制图、多传感器融合四方面, 从优缺点、适用范围、概率算法、地图类型及融合方式出发, 介绍SLAM技术实现过程中的各个环节, 系统阐述了国内外相关的研究成果。基于多传感器融合SLAM, 分析了目前常见的融合SLAM技术难题, 对SLAM技术的未来发展趋势及实际工程应用做出展望。     

基于深度相机的SLAM算法评测

针对3种典型的基于深度相机的同步定位与地图构建(SLAM)算法,包括RGB-D SLAM V2,RTAB-Map和DVO SLAM,介绍这3种SLAM算法的理论特点。采用两个公开的SLAM数据集,包括TUM数据集和ICL-NUIM数据集,进行SLAM算法的评测,评测指标包括SLAM算法的精确度、运行性能以及鲁棒性。评测的实验结果表明,在选择基于深度相机的SLAM算法时:如果考虑精确度和鲁棒性优先于运行性能,则选择RGB-D SLAM V2;如果考虑运行性能和鲁棒性优先于精确度,则选择DVO SLAM;如果考虑精确度和运行性能优先于鲁棒性,则选择RTAB-Map。