局部不变特征综述

局部不变特征是近年来计算机视觉领域的研究热点。局部不变特征在宽基线匹配、特定目标识别、目标类别识别、图像及视频检索、机器人导航、场景分类、纹理识别和数据挖掘等多个领域得到了广泛的应用。本文基于局部不变特征检测、局部不变特征描述和局部不变特征匹配3个基本问题,综述了文献中现有的局部不变特征研究方法,并比较了各类方法的优缺点。根据特征层次的不同,局部不变特征检测方法可以分为角点不变特征、blob不变特征和区域不变特征检测方法3类。局部不变特征的描述方法可以分为基于分布的描述方法、基于滤波的描述方法、基于矩的描述方法和其他描述方法。局部不变特征匹配的研究主要集中在相似性度量、匹配策略和匹配验证3个方面。最后在分析各类研究方法的基础上,总结了局部不变特征研究目前存在的一些问题及可能的发展方向。     

β-SLAM: Simultaneous localization and grid mapping with beta distributions

Simultaneous localization and mapping (SLAM) is one of the most frequently studied problems in mobile robotics. Different map representations have been proposed in the past and a popular one are occupancy grid maps, which are particularly well suited for navigation tasks. The uncertainty in these maps is usually modeled as a single Bernoulli distribution per grid cell. This has the disadvantage that one cannot distinguish between uncertainty caused by different phenomena like missing or conflicting information. In this paper, we overcome this limitation by modeling the occupancy probabilities as random variables. Those are assumed to be beta-distributed and account for the different causes of uncertainty. Based on this map representation, we derive a SLAM algorithm, including all necessary sensor models, for building maps composed of beta-distributed random variables and using these maps for localization. Furthermore, we propose measures for quantifying uncertainty in the resulting maps and for solving navigation tasks. We evaluate our approach using real-world as well as simulation-based datasets and we compare it to a state-of-the-art SLAM algorithm for building classical grid maps.     

An Efficient Direct Approach to Visual SLAM

The majority of visual simultaneous localization and mapping (SLAM) approaches consider feature correspondences as an input to the joint process of estimating the camera pose and the scene structure. In this paper, we propose a new approach for simultaneously obtaining the correspondences, the camera pose, the scene structure, and the illumination changes, all directly using image intensities as observations. Exploitation of all possible image information leads to more accurate estimates and avoids the inherent difficulties of reliably associating features. We also show here that, in this case, structural constraints can be enforced within the procedure as well (instead of a posteriori), namely the cheirality, the rigidity, and those related to the lighting variations. We formulate the visual SLAM problem as a nonlinear image alignment task. The proposed parameters to perform this task are optimally computed by an efficient second-order approximation method for fast processing and avoidance of irrelevant minima. Furthermore, a new solution to the visual SLAM initialization problem is described whereby no assumptions are made about either the scene or the camera motion. Experimental results are provided for a variety of scenes, including urban and outdoor ones, under general camera motion and different types of perturbations.      

基于SuperPoint的轻量级特征点及描述子提取网络

图像特征点及描述子提取是SLAM、SFM和3D重建等任务的基础,较好的图像特征点及描述子提取算法会对这些任务的进步产生十分重要的作用.本文聚焦于提取特征点和描述子算法中鲁棒性较高、性能较好的SuperPoint网络,对该网络进行了一定程度的改进.针对其计算量和参数较大的问题,首先将普通卷积改成深度可分离卷积,改变卷积层数和下采样方式,之后改进通道剪枝算法,使其可以应用于深度可分离卷积,对网络进行剪枝.实验结果显示,在轻微损失特征点检测和匹配效果的情况下,将网络参数量压缩为原来网络的15％,运算量压缩为原来网络的5％,FPS提升6.64倍,取得了较好的实验效果.     

Evaluation of Interest Point Detectors

Many different low-level feature detectors exist and it is widely agreed that the evaluation of detectors is important. In this paper we introduce two evaluation criteria for interest points' repeatability rate and information content. Repeatability rate evaluates the geometric stability under different transformations. Information content measures the distinctiveness of features. Different interest point detectors are compared using these two criteria. We determine which detector gives the best results and show that it satisfies the criteria well.     

Recent advances in simultaneous localization and map-building using computer vision

Simultaneous localization and map-building (SLAM) continues to draw considerable attention in the robotics community due to the advantages it can offer in building autonomous robots. It examines the ability of an autonomous robot starting in an unknown environment to incrementally build an environment map and simultaneously localize itself within this map. Recent advances in computer vision have contributed a whole class of solutions for the challenge of SLAM. This paper surveys contemporary progress in SLAM algorithms, especially those using computer vision as main sensing means, i.e., visual SLAM. We categorize and introduce these visual SLAM techniques with four main frameworks: Kalman filter (KF)-based, particle filter (PF)-based, expectation-maximization (EM)-based and set membership-based schemes. Important topics of SLAM involving different frameworks are also presented. This article complements other surveys in this field by being current as well as reviewing a large body of research in the area of vision-based SLAM, which has not been covered. It clearly identifies the inherent relationship between the state estimation via the KF versus PF and EM techniques, all of which are derivations of Bayes rule. In addition to the probabilistic methods in other surveys, non-probabilistic approaches are also covered.     

Simultaneous localization and mapping for aerial vehicles: a 3-D sensor-based GAS filter

This paper presents the design, analysis, and experimental validation of a globally asymptotically stable (GAS) filter for simultaneous localization and mapping (SLAM) with application to unmanned aerial vehicles. The main contributions of this paper are the results of global convergence and stability for SLAM in tridimensional (3-D) environments. The SLAM problem is formulated in a sensor-based framework and modified in such a way that the structure may be regarded as linear time-varying for observability purposes, from which a Kalman filter with GAS error dynamics follows naturally. The proposed solution includes the estimation of both body-fixed linear velocity and rate gyro measurement biases. Experimental results from several runs, using an instrumented quadrotor equipped with a RGB-D camera, are included in the paper to illustrate the performance of the algorithm under realistic conditions.     

Fusing Monocular Information in Multicamera SLAM

This paper explores the possibilities of using monocular simultaneous localization and mapping (SLAM) algorithms in systems with more than one camera. The idea is to combine in a single system the advantages of both monocular vision (bearings-only, infinite range observations but no 3-D instantaneous information) and stereovision (3-D information up to a limited range). Such a system should be able to instantaneously map nearby objects while still considering the bearing information provided by the observation of remote ones. We do this by considering each camera as an independent sensor rather than the entire set as a monolithic supersensor. The visual data are treated by monocular methods and fused by the SLAM filter. Several advantages naturally arise as interesting possibilities, such as the desynchronization of the firing of the sensors, the use of several unequal cameras, self-calibration, and cooperative SLAM with several independently moving cameras. We validate the approach with two different applications: a stereovision SLAM system with automatic self-calibration of the rig's main extrinsic parameters and a cooperative SLAM system with two independent free-moving cameras in an outdoor setting.      

Simulation study of a constant time hybrid approach for large scale terrain mapping using satellite stereo imagery

This work addresses the problem of performing large scale SLAM (Simultaneous Localization And Mapping) with satellite stereo imagery for terrain mapping, using a constant time estimation approach. The approach adopts the relative bundle adjustment approach (RBA) and integrates with it a particle-based framework to obtain a constant time probabilistic pose estimation model. The approach further uses a concept of fuzzy landmark-based similarity between poses to make common landmark identification across poses easier, especially when landmarks are sparsely encountered. In order to achieve robustness under varying environmental conditions, we use Speeded Up Robust Features (SURF) for computing spatial and temporal landmark correspondences across time steps. Finally, we use a fast loop closure approach to reduce drifts and obtain global pose estimates. For simulation study, the robot images are cropped from stereo-pair satellite images at different time steps incorporating errors in the robot’s control information. Extensive experimentation has been carried out to study the robot trajectories and the determination of Digital Elevation Model (DEM), with encouraging findings. We have also compared our work with 6D FastSLAM 2.0 (Thrun et al. (2005)) as well as Relative SLAM (RSLAM) due to Mei et al. (2010).