动态场景下基于视觉的SLAM技术研究

针对视觉同时定位与地图构建(SLAM)技术在动态环境中存在定位精度低、地图虚影等问题,提出了一种基于深度学习的动态SLAM算法。该算法利用网络参数少且目标识别率高的YOLOv8n改善系统的视觉前端,为视觉前端增加语义信息,提取动态区域特征点。然后采用LK光流法识别动态区域的动态特征点,剔除动态特征点并保留动态区域内的静态特征点,提高特征点利用率。此外,该算法通过增加地图构建线程,剔除YOLOv8n提取的动态物体点云,接收前端提取的语义信息,实现静态语义地图构建,消除由动态物体产生的虚影。实验结果显示,在动态环境下该算法与ORB-SLAM3相比,定位精度提升92.71%,与其他动态视觉SLAM算法相比,也有小幅度改善。     

动态环境下基于深度学习的视觉SLAM研究综述

目前的同时定位与地图构建(Simultaneous Localization and Mapping,SLAM)研究大多是基于静态场景的假设,而实际生活中动态物体是不可避免的。在视觉SLAM系统中加入深度学习,可以协同剔除场景中的动态物体,有效提升视觉SLAM在动态环境下的鲁棒性。文章首先介绍了动态环境下基于深度学习的视觉SLAM分类,然后详细介绍了基于目标检测、基于语义分割和基于实例分割的视觉SLAM,并对它们进行了分析比较。最后,结合近年来视觉SLAM的发展趋势,通过对动态环境下基于深度学习的视觉SLAM存在的主要问题进行分析,总结了未来可能的发展方向。     

视觉SLAM在动态场景下的图像处理方法

SLAM一直是机器人领域的研究热点,近年来取得了万众瞩目的进步,但很少有SLAM算法考虑到动态场景的处理。针对视觉SLAM场景中动态目标的处理,提出一种在动态场景下的图像处理方法。将基于深度学习的语义分割算法引入到ORB_SLAM2方法中,对输入图像进行分类处理的同时剔除人身上的特征点。基于已经剔除特征点的图像进行位姿估计。在TUM数据集上与ORB_SLAM2进行对比,在动态场景下的绝对轨迹误差和相对路径误差精度提高了90%以上。在保证地图精度的前提下,改善了地图的适用性。     

面向室内动态场景的VSLAM

传统VSLAM算法基于静态场景实现,其在室内动态场景下定位精度退化,三维稀疏点云地图也会出现动态特征点误匹配等问题.文中在ORB-SLAM2框架上进行改进,结合Mask R-CNN进行图像的语义分割,剔除位于动态物体上的动态特征点,优化了相机位姿,得到了静态的三维稀疏点云地图.在公开的TUM数据集上的实验结果表明,结合...     

YOLO-NKLT视觉SLAM回环检测方法

针对存在明显光照变化或遮挡物等室外复杂场景下,现有基于深度学习的视觉即时定位与地图构建(visual simultaneous localization and mapping,视觉SLAM)回环检测方法没有很好地利用图像的语义信息、场景细节且实时性差等问题,本文提出了一种YOLO-NKLT视觉SLAM回环检测方法。采用改进损失函数的YOLOv5网络模型获取具有语义信息的图像特征,构建训练集,对网络重训练,使提取的特征更加适用于复杂场景下的回环检测。为了进一步提高闭环检测的实时性,提出了一种基于非支配排序的KLT降维方法。通过在New College数据集和光照等变化更复杂的Nordland数据集上进行实验,结果表明：室外复杂场景下,相较于其他传统和基于深度学习的方法,所提方法具有更高的鲁棒性,可以取得更佳的准确率和实时性表现。     

动态场景下融合YOLOv5s的视觉SLAM算法研究

为了解决视觉同步定位与建图(Simultaneous Localization and Mapping, SLAM)系统在动态场景下容易受到动态物体干扰,导致算法定位精度和鲁棒性下降的问题,提出了一种融合YOLOv5s轻量级目标检测网络的视觉SLAM算法。在ORB-SLAM2的跟踪线程中添加了目标检测和剔除动态特征点模块,通过剔除图像中的动态特征点,提高SLAM系统的定位精度和鲁棒性。改进了YOLOv5s的轻量化目标检测算法,提高了网络在移动设备中的推理速度和检测精度。将轻量化目标检测算法与ORB特征点算法结合,以提取图像中的语义信息并剔除先验的动态特征。结合LK光流法和对极几何约束来剔除动态特征点,并利用剩余的特征点进行位姿匹配。在TUM数据集上的验证表明,提出的算法与原ORB-SLAM2相比,在高动态序列下的绝对轨迹误差(Absolute Trajectory Error, ATE)和相对轨迹误差(Relative Pose Error, RPE)均提高了95%以上,有效提升了系统的定位精度和鲁棒性。相对当前一些优秀的SLAM算法,在精度上也有明显的提升,并且具有更高的实时性,在移...     

基于Deeplab V3+的机器人语义SLAM算法研究

针对机器人在同步定位与地图构建(SLAM)系统中受几何场景信息计算力和带宽负载的限制，对ORB-SLAM2框架进行改进，提出语义跟踪和语义建图线程，语义跟踪线程通过Deeplab V3+对图像语义分割，同时提取该图像特征点，进行移动一致性检查来剔除动态噪声点，结合一致性检查后的特征点和分割后的图像信息来二次检查动态点，随后位姿估计，而语义建图线程主要完成语义八叉树地图的构建。在TUM RGB-D数据集上进行了广泛实验，在walking系列数据中的旋转漂移误差达到1.19m、平移漂移误差达到0.046m,满足实时性要求，所提方法有效提高了SLAM的精度和鲁棒性。     

基于改进PFPN的语义地图构建方法研究

针对园区等环境结构性强而全球导航卫星系统(GNSS)信号不稳定的应用场景及无人车应用激光雷达同时定位与建图(SLAM)技术缺乏对于场景的语义理解能力而造成的定位误差问题,提出了一种单目相机与激光雷达融合构建三维语义地图的方法.该方法以园区环境结构性强、车辆行人动态变化高的特征为依据,通过改进的全景特征金字塔网络(PFP...     

基于YOLOv4和HOG的闭环检测算法

传统的视觉SLAM闭环检测算法大多采用手工设计的图像特征,适用于室内静态场景但在复杂场景下的准确性不高.为此,在卷积自编码器网络模型基础之上设计了一种新颖的闭环检测算法.首先针对场景中可能会出现的动态物体干扰,使用重训练的YOLOv4目标检测算法对原始图像进行动态物体去除.此外,利用定向梯度直方图(HOG)提供的几何信...     

基于虚拟现实的移动机器人真实环境三维建模系统

传统的虚拟现实（VR）技术通过人为建模的方式生成室内三维地图模型，存在速度慢、模型与现实物体尺度之间存在偏差的问题。鉴于此，提出基于VR的移动机器人的真实环境三维建模系统。首先通过视觉同时定位与建图（SLAM）技术快速地获取室内的高精度稠密三维点云地图；其次将三维点云通过曲面重建算法重建为室内三维模型并导入到unity 3D中；然后借助VR设备将室内三维模型置于三维立体的虚拟环境中；最后通过视觉SLAM技术实现移动机器人在室内环境的重定位，实时映射机器人在模型中的位姿，完成交互。利用视觉SLAM技术构建三维地图模型不仅快速，解决了场景尺度偏差的问题，且实现地图的重复使用。同时VR技术也使操作人员可以获得强烈的沉浸感，从而更好地理解机器人的工作环境。     

Encode–decode network with fully connected CRF for dynamic objects detection and static maps reconstruction

The key of robots operating autonomously in dynamic environments is understanding the dynamic characteristics of objects. This paper aims to detect dynamic objects and reconstruct 3D static maps from consecutive scans of scenes. Our work starts from an encode–decode network, which receives two range maps provided by a Velodyne HDL-64 laser scanner and outputs dynamic probability of each point. Since the soft segmentation produced by the network tends to be smooth, a 3D fully connected CRF (Conditional Random Field) is proposed to improve the segmentation performance. Experiments on both the public datasets and real-word platform demonstrate the effectiveness of our method.     

Calibrated depth and color cameras for accurate 3D interaction in a stereoscopic augmented reality environment

A Human–machine interaction system requires precise information about the user’s body position, in order to allow a natural 3D interaction in stereoscopic augmented reality environments, where real and virtual objects should coherently coexist. The diffusion of RGB-D sensors seems to provide an effective solution to such a problem. Nevertheless, the interaction with stereoscopic 3D environments, in particular in peripersonal space, requires a higher degree of precision. To this end, a reliable calibration of such sensors and an accurate estimation of the relative pose of different RGB-D and visualization devices are crucial. Here, robust and straightforward procedures to calibrate a RGB-D camera, to improve the accuracy of its 3D measurements, and to co-register different calibrated devices are proposed. Quantitative measures validate the proposed approach. Moreover, calibrated devices have been used in an augmented reality system, based on a dynamic stereoscopic rendering technique that needs accurate information about the observer’s eyes position.     

基于混合高斯模型的运动目标检测改进算法

针对传统混合高斯模型法的不足,提出一种基于混合高斯模型的运动目标检测改进算法。首先对模型的参数更新机制进行了改进,不同阶段采用不同的更新率,并选择性地更新背景模型;其次,将改进后的混合高斯模型法与和帧差法结合,进行两次与运算和一次形态学膨胀处理,得到最后的运动目标。实验结果表明,该方法能够有效地消除复杂环境中的噪声,并对阴影有一定的抑制作用,提高了运动目标检测的准确性。     

Detecting moving object from dynamic background video sequences via simulating heat conduction

Moving object detection is one of the essential tasks for surveillance video analysis. The dynamic background often composed by waving trees, rippling water or fountains, etc. in nature scene greatly interferes with the detection of moving objects in the form of noise. In this paper, a method simulating heat conduction is proposed to extract moving objects from dynamic background video sequences. Based on the visual background extractor (ViBe) with an adaptable distance threshold, we design a temperature field relying on the generated mask image to distinguish between the moving objects and the noise caused by dynamic background. In temperature field, a brighter pixel is associated with more energy. It will transfer a certain amount of energy to its neighboring darker pixels. Through multiple steps of energy transfer the noise regions loss more energy so that they become darker than the detected moving objects. After heat conduction, K-Means algorithm with the customized initial clustering centers is utilized to separate the moving objects from background. We test our method on many videos with dynamic background from public datasets. The results show that the proposed method is feasible and effective for moving object detection from dynamic background sequences.     

EMOS: Enhanced moving object detection and classification via sensor fusion and noise filtering

Dynamic object detection is essential for ensuring safe and reliable autonomous driving. Recently, light detection and ranging (LiDAR)-based object detection has been introduced and shown excellent performance on various benchmarks. Although LiDAR sensors have excellent accuracy in estimating distance, they lack texture or color information and have a lower resolution than conventional cameras. In addition, performance degradation occurs when a LiDAR-based object detection model is applied to different driving environments or when sensors from different LiDAR manufacturers are utilized owing to the domain gap phenomenon. To address these issues, a sensor-fusion-based object detection and classification method is proposed. The proposed method operates in real time, making it suitable for integration into autonomous vehicles. It performs well on our custom dataset and on publicly available datasets, demonstrating its effectiveness in real-world road environments. In addition, we will make available a novel three-dimensional moving object detection dataset called ETRI 3D MOD.     

Efficient visual object detection with spatially global Gaussian mixture models and uncertainties

In this paper, we deal with the problem of visual detection of moving objects using innovative Gaussian mixture models (GMM). The proposed method, the Spatially Global Gaussian Mixture Model (SGGMM) uses RGB and pixel uncertainty for background modelling. The SGGMM with colours only is used for scenes with moderate illumination changes. When combined with pixel uncertainty statistics, the method can deal efficiently with dynamic backgrounds and scene backgrounds with fast change in luminosity. Experimental evaluation in indoor and outdoor environments shows the performance of the foreground segmentation with the proposed SGGMM model using solely RGB colour or in combination with pixel uncertainties. These experimental scenarios take into account changes in the background within the scene. They are also used to compare the proposed technique with other state-of-the-art segmentation approaches in terms of accuracy and execution time performance. Further, our solution is implemented and tested in embedded camera sensor network nodes.     

基于背景差分和混合帧差的运动目标检测

针对利用核密度估计建立背景模型时计算量大,运动目标和外界环境容易发生变化,提出一种基于改进的核密度估计背景差分法和改进的混合帧差法相结合的运动目标检测方法。该方法在背景建模时,先对背景差分后的图像进行分块和分类,并简化了核密度估计的核函数,对前景块中的像素进行核密度估计,减少了计算量。在混合帧差法中增加了动态阈值,提高了对光线变化的适应性。实验结果表明该方法能够完整地提取出运动目标,提高了目标检测的准确率。     

基于五帧差分和模板匹配的运动目标检测识别

针对帧差法和背景差分法检测运动目标准确率低,自适应能力弱等缺陷,提出了一种改进五帧差分法与背景差分法和模板匹配相结合的运动目标检测和识别算法;通过改进的五帧差分和背景差分法融合的算法从视频图像序列中检测出运动目标;利用形态学方法去除噪声,改善运动目标提取效果;在Harris算法提取图像匹配特征值的基础上角点配准,提高图像识别的准确率,通过提取目标特征与自适应模板图像进行特征匹配的方法实现了目标检测识别和跟踪。仿真结果和实验表明该方法有噪声和部分遮挡的运动目标有良好的检测识别效果,识别率达到了95%。     

基于深度帧差卷积神经网络的运动目标检测方法研究

复杂场景中的运动目标检测是计算机视觉领域的重要问题,其检测准确度仍然是一大挑战.本文提出并设计了一种用于复杂场景中运动目标检测的深度帧差卷积神经网络（Deep Difference Convolutional Neural Network,DFDCNN）.DFDCNN由DifferenceNet和AppearanceNet组成,不需要后处理就可以预测分割前景像素.DifferenceNet具有孪生Encoder-Decoder结构,用于学习两个连续帧之间的变化,从输入（t帧和t+1帧）中获取时序信息;AppearanceNet用于从输入（t帧）中提取空间信息,并与时序信息融合;同时,通过多尺度特征图融合和逐步上采样来保留多尺度空间信息,以提高网络对小目标的敏感性.在公开标准数据集CDnet2014和I2R上的实验结果表明：DFDCNN不仅在动态背景、光照变化和阴影存在的复杂场景中具有更好的检测性能,而且在小目标存在的场景中也具有较好的检测效果.