一种室内动态场景下RGB-D SLAM的前端优化方法

针对动态物体会导致位姿估计误差过大的问题,提出一种适用于动态场景下的RGB-D SLAM算法。首先,将目标检测的结果根据物体是否具有自主移动能力划分为动态物体与潜在动态物体,并使用大津法在对应的深度图中完成检测框内区域的前、背景分割;然后,结合检测框的位置坐标与深度信息确定潜在动态物体与动态物体在三维空间内是否关联,初步筛选出具有一定运动概率的潜在动态物体;最后,通过运动一致性检测算法来最终确定是否将潜在动态物体作为动态物体处理。实验部分选用公开的TUM数据集,结果表明,该算法在保持实时性的同时具有较高的定位精度和鲁棒性。     

融合点线特征的视觉惯性SLAM算法

为使移动机器人在昏暗场景中获得更高的定位精度以实现更好的建图效果,针对特征点稀疏、相机移动过快所导致的SLAM算法初始化困难、定位精度较低等问题,提出了一种融合点线特征的视觉惯性SLAM算法。通过EDLines线特征提取匹配算法来引入丰富的线特征,进而提升特征跟踪的鲁棒性;利用惯性传感器和视觉传感器的互补优势,通过视觉惯性分步联合初始化方法在初始化阶段分三步进行优化,进而提升初始化的精度和速度。实验表明,该算法所使用的线特征提取匹配算法相比传统的LSD算法具有了更快的匹配速度和更低的误匹配率,并且该算法在视觉惯性数据集中相机位姿的估计精度更高、鲁棒性更强。     

动态环境下融合轻量级YOLOv5s的视觉SLAM

移动机器人在未知环境下依靠同步定位与地图构建（SLAM）实现自身的精确定位,目前大多数视觉SLAM系统在运行时均假设外部环境是静态的,但在实际应用场景下该假设并不成立,传统的视觉SLAM系统在动态环境下易受移动目标的影响,导致系统定位精度下降。提出一种新的视觉SLAM系统,将轻量级网络MobileNetV3作为目标检测网络YOLOv5s的主干网络,从而减少模型参数量,提高算法在CPU上的推理速度。将目标检测网络、光流法与ORB-SLAM系统相结合,使SLAM系统前端提取ORB特征点的同时能够有效剔除动态特征点。仅利用静态目标上的特征点进行帧间匹配,从而求解相机位姿,提高系统在动态环境下的定位精度。在TUM动态数据集上的实验结果表明,与ORB-SLAM3系统相比,该系统的位姿估计精度提升了80.16%,与DS-SLAM、DVO-SLAM系统等动态SLAM系统相比,该系统在定位精度上有大幅提升,相比使用MASK-RCNN的DynaSLAM系统,在保持相近ATE指标的情况下,该系统具有更高的实时性。     

动态环境下基于增强分割的RGB-D SLAM方法

目前视觉SLAM（同步定位与地图创建）方法在动态环境下易出现漏剔除动态物体的问题,影响相机位姿估计精度以及地图的可用性。为此,本文提出一种基于增强分割的RGB-D SLAM方法。首先结合实例分割网络与深度图像聚类的结果,判断当前帧是否出现漏分割现象,若出现则根据多帧信息对分割结果进行修补,同时,提取当前帧的Shi-To...     

可见光–红外特征交互与融合的YOLOv5目标检测算法

目标检测是自动驾驶系统的关键技术, 普通RGB目标检测算法在夜间和恶劣气候等场景往往表现一般, 融合可见光和红外信息的目标检测算法因而受到诸多研究关注. 现有方法通常融合结构复杂, 且忽视了模态间信息交流的重要性. 对此, 本文以YOLOv5为基本框架, 提出一种可见光–红外特征交互与融合的目标检测算法, 使用一种新的主干网络跨阶段局部(CSPDarknet53-F), 采用双分支结构分别提取可见光和红外特征. 然后, 通过特征交互模块重构各模态的信息成分和比例, 提升模态间信息交流, 使可见光和红外特征进行更充分的融合. 在FLIR-aligned和M$^3$FD数据集上的大量实验证明, 本文算法使用的CSPDarknet53-F在协同利用可见光和红外信息方面更加出色, 提升了模型精度, 同时, 拥有对抗光照强度骤变的鲁棒性     

动态环境下基于深度学习的语义SLAM算法

针对应用场景中存在的运动物体会降低视觉同步定位与地图构建（SLAM）系统的定位精度和鲁棒性的问题,提出一种基于语义信息的动态环境下的视觉SLAM算法。首先,将传统视觉SLAM前端与YOLOv4目标检测算法相结合,在对输入图像进行ORB特征提取的同时,对该图像进行语义分割;然后,判断目标类型以获得动态目标在图像中的区域,剔除分布在动态物体上的特征点;最后,使用处理后的特征点与相邻帧进行帧间匹配来求解相机位姿。实验采用TUM数据集进行测试,测试结果表明,所提算法相较于ORB-SLAM2在高动态环境下在位姿估计精度上提升了96.78%,同时该算法的跟踪线程处理一帧的平均耗时为0.065 5 s,相较于其他应用在动态环境下的SLAM算法耗时最短。实验结果表明,所提算法能够实现在动态环境中的实时精确定位与建图。     

室内动态环境下基于网格分割与双地图耦合的RGB-D SLAM算法

为解决室内动态环境下现有RGB-D SLAM（同步定位与地图创建）系统定位精度低、建图效果差的问题,提出一种基于网格分割与双地图耦合的RGB-D SLAM算法。基于单应运动补偿与双向补偿光流法,根据几何连通性与深度图像聚类结果实现网格化运动分割,同时保证算法的快速性。利用静态区域内的特征点最小化重投影误差对相机进行位置估计。结合相机位姿、RGB-D图像、网格化运动分割图像,同时构建场景的稀疏点云地图和静态八叉树地图并进行耦合,在关键帧上使用基于网格分割和八叉树地图光线遍历的方法筛选静态地图点,更新稀疏点云地图,保障定位精度。公开数据集和实际动态场景中的实验结果都表明,本文算法能够有效提升室内动态场景中的相机位姿估计精度,实现场景静态八叉树地图的实时构建和更新。此外,本文算法能够实时运行在标准CPU硬件平台上,无需GPU等额外计算资源。     

An improved PSO-based approach with dynamic parameter tuning for cooperative multi-robot target searching in complex unknown environments

Target searching in complex unknown environments is a challenging aspect of multi-robot cooperation. In this paper, an improved particle swarm optimisation (PSO) based approach is proposed for a team of mobile robots to cooperatively search for targets in complex unknown environments. The improved cooperation rules for a multi-robot system are applied in the potential field function, which acts as the fitness function of the PSO. The main improvements are the district-difference degree and dynamic parameter tuning. In the simulation studies, various complex situations are investigated and compared to the previous research results. The results demonstrate that the proposed approach can enable the multi-robot system to accomplish the target searching tasks in complex unknown environments.     

Quick segmentation for complex curved surface with local geometric feature based on IGES and Open CASCADE

Complex curved surface parts with local geometric feature are usually critical parts in high-end equipments. However, the processing for this kind of parts is usually difficult or inefficient due to the adoption of difficult-to-machine material and special structure. Current approaches cannot satisfy the rapid development of high-end equipments. Due to the existence of the local geometric feature for the parts, processing such parts with constant machining parameters is less applicative, restricting the improvement of machining efficiency. By separating the local geometric feature and generating tool path for the local geometric feature and the remaining processing area separately, the more efficient machining with variable machining parameters will be obtained for the complex curved surface with local geometric feature. In this way, the quick segmentation for the complex curved surface with local geometric feature is of great importance to the NC machining with variable machining parameters for this kind of parts, and a quick segmentation system is developed based on Initial Graphics Exchange Specification (IGES) and Open CASCADE (OCC) platform in this study. The complex curved surface model in IGES format is firstly imported into the system and then trimmed into independent surface patches. After computing the feature size of each surface patch, the segmentation for the complex curved surface is achieved by sorting and classifying the surface patches according to their feature sizes. Taking the whole impeller with small splitter blades for an example, the experimental result shows that the segmentation of small splitter blades from the whole impeller is successful and a serialized processing program could be generated, and then the whole impeller could be machined precisely and efficiently with NC equipment. In the machining experiment, it is proved that the machining with various machining parameters can improve the efficiency by 28.18% in the comparison experiment, 20.14% and 12.33% in the estimation. The research provides an important foundation for the high quality and more efficient machining of the complex curved surface with local geometric feature.     

Cooperative path planning with priority target assignment and collision avoidance guidance for rescue unmanned surface vehicles in a complex ocean environment

In this paper, a novel cooperative path planning scheme of unmanned surface vehicles (USVs) for rescuing targets in a complex ocean environment is proposed. The primary objective of the rescue USVs is to bring all targets back safely on the premise of first rescuing priority targets, while optimizing the path length, the navigation time and the angular energy. The main contributions of this paper are as follows: (1) The proposed K-means-division (KMD) algorithm is able to identify a complex ocean environment with collision-free zone and static-obstacles zone; (2) The proposed path planning method with fast-marching-method-based ellipse guidance range (E-FMM) is able to optimize the angular energy while ensuring safety; (3) The proposed cooperative management system (including priority-target-assignment (PTA) with reward-mechanism genetic-optimization (RM-GO) and collision-avoidance (CA) guidance law with Tangent-based surge-varying wave-disturbances-observer (Tangent-SV-WDO)) can accomplish the mission of the rescue USVs. Comparative studies with the state-of-the-art methods demonstrate that the proposed cooperative path planning scheme is superior in terms of priority-target-assignment (PTA) and collision-avoidance (CA) of the actual rescue work.     

Adaptive fuzzy dynamic surface control for uncertain nonlinear systems in pure-feedback form with input and state constraints using noisy measurements

This paper is concerned with the design of an adaptive fuzzy dynamic surface control for uncertain nonlinear pure-feedback systems with input and state constraints using a set of noisy measurements. The design approach is described as follows. The nonlinear uncertainties are approximated by using the fuzzy logic systems at the first stage, secondly the adaptive fuzzy dynamic surface control is introduced to remove the problem of the explosion of complexity for the derivation of the adaptive fuzzy backstepping control, thirdly a new saturation function for state constraints is proposed to design the controllers based on the Lyapunov function, fourthly the number of the adjustable parameters is reduced by using the simplified extended single input rule modules, and finally the weighted least squares estimator to take the estimates for the un-measurable states and the adjustable parameters is in a simplified structure designed. The proposed approach provides effective system performance in the simulation experiment.