A Comprehensive Survey of Visual SLAM Algorithms

Several works have focused on Simultaneous Localization and Mapping (SLAM), which is a topic that has been studied for more than a decade to meet the needs of robots to navigate in an unknown environment. SLAM is an essential perception functionality in several applications, especially in robotics and autonomous vehicles. RGB-D cameras are among the sensors commonly used with recent SLAM algorithms. They provide an RGB image and the associated depth map, making it possible to solve scale drift with less complexity and create a dense 3D environment representation. Many RGB-D SLAM algorithms have been studied and evaluated on publicly available datasets without considering sensor specifications or image acquisition modes that could improve or decrease localization accuracy. In this work, we deal with indoor localization, taking into account the sensor specifications. In this context, our contribution is a deep experimental study to highlight the impact of the sensor acquisition modes on the localization accuracy, and a parametric optimization protocol for a precise localization in a given environment. Furthermore, we apply the proposed protocol to optimize a depth-related parameter of the SLAM algorithm. The study is based on a publicly available dataset in an indoor environment with a depth sensor. The reconstruction results’ analysis is founded on the study of different metrics involving translational and rotational errors. These metrics errors are compared with those obtained with a state-of-the-art stereo vision-based SLAM algorithm.

In indoor dynamic scenes, traditional visual simultaneous localization and mapping (SLAM) algorithms based on RGB-D cameras incorrectly use dynamic features to estimate the poses of the cameras, and do not fully utilize the geometric information in the scenes, resulting in the low positioning accuracy and robustness of SLAM systems. To solve this problem, this paper proposes an RGB-D SLAM algorithm based on multiple geometric features and semantic segmentation. The core of our SLAM system is the proposed robust exclusion method of dynamic point and line features. The method consists of the following three steps: (a) identify potential dynamic point features using motion consistency checks; (b) obtain potential motion regions via semantic segmentation, and then determine dynamic regions by combining these with dynamic point features; (c) remove point and line features in dynamic regions. The exclusion method of dynamic point and line features can be easily integrated into RGB-D SLAM systems for improving the accuracy and robustness of SLAM systems in dynamic scenes. Experimental results on the Technische Universität München (TUM) dataset demonstrate that the proposed algorithm has better positioning accuracy and stability than the original dynamic semantic (DS-SLAM) algorithm in dynamic environments. The effectiveness of the proposed algorithm is verified by comparison with other classical visual SLAM algorithms. Better mapping performance is achieved by this proposed algorithm in actual indoor scenes.

Simultaneous Localization and Mapping (SLAM) technology can help mobile intelligent robots to understand and perceive scenes in unknown environments, so it is playing an increasingly important role in the fields of intelligent robots, intelligent cars, and so on. And because camera sensors have wide applicability, visual SLAM in dynamic scenes has become a relatively popular research direction in recent years. And the SLAM algorithm needs to be tested and validated, which requires choosing appropriate datasets based on different application scenarios. Therefore, this paper comprehensively introduces the excellent open-source data sets commonly used in the research of visual SLAM. Since deep learning networks are often used to improve the performance of the SLAM system, this paper summarizes the advanced techniques in recent years for solving the common problems of visual SLAM in dynamic scenes based on deep learning networks.<br />

Localization and mapping in a dynamic scene is a crucial problem for the indoor visual simultaneous localization and mapping (SLAM) system. Most existed visual odometry (VO) or SLAM systems are based on the assumption that the environment is static. The performance of a SLAM system may degenerate when it is operated in a severely dynamic environment. The assumption limits the applications of RGB-D SLAM in the dynamic environment. In this paper, we propose a workflow to segment the objects accurately, which will be marked as the potentially dynamic-object area based on the semantic information. A novel approach for motion detection and removal from the moving camera is introduced. We integrate the semantics-based motion detection and the segmentation approach with an RGB-D SLAM system. To evaluate the effectiveness of the proposed approach, we conduct the experiments on the challenging dynamic sequences of TUM-RGBD datasets. The experimental results suggest that our approach improves the accuracy of localization and outperforms the state-of-the-art dynamic-removal-based SLAM system in both severely dynamic and slightly dynamic scenes.

Simultaneous Localization and Mapping (SLAM) is a fundamental problem in mobile robotics. However, the majority of Visual SLAM algorithms assume a static scenario, limiting their applicability in real-world environments. Dealing with dynamic content in Visual SLAM is still an open problem, with solutions usually relying on direct or feature-based methods. Deep learning techniques can improve the SLAM solution in environments with a priori dynamic objects, providing high-level information of the scene. This paper presents a new approach to SLAM in human populated environments using deep learning-based techniques. The system is built on ORB-SLAM2, a state-of-the-art SLAM system. The proposed methodology is evaluated using a benchmark dataset, outperforming other Visual SLAM methods in highly dynamic scenarios.

Right invariant extended Kalman filter (RIEKF) based simultaneous localization and mapping (SLAM) proposed recently has shown to be able to produce more consistent SLAM estimates as compared with traditional EKF based SLAM methods, including some improved EKF SLAM methods such as observability constrained-EKF (OC-EKF) SLAM. Latest results have demonstrated that its performance is very close to optimization based SLAM algorithms such as iSAM. In this paper, we propose to use RIEKF SLAM algorithm in active SLAM where both the predicted SLAM results for choosing control actions and the actual estimated SLAM results applying the selected control actions are computed using RIEKF algorithms. The advantages over traditional EKF based active SLAM are the more accurate and consistent predicted uncertainty estimates which result in robustness of the active SLAM algorithm. The advantages over optimization based active SLAM is the reduced computational cost. Simulation results are presented to validate the advantages of the proposed algorithm³.

Current visual simultaneous localization and mapping (SLAM) systems have demonstrated commendable efficacy in static environments. However, the presence of dynamic objects in real-world settings frequently leads to system discrepancies, significantly impairing the accuracy and robustness of SLAM systems. Conventional visual SLAM approaches typically utilize epipolar constraints to mitigate the impact of outliers; nevertheless, they encounter limitations when confronted with a substantial number of dynamic or planar moving objects. To tackle these issues, this paper introduces a novel dynamic visual SLAM system, termed DEG-SLAM. Initially, the system employs the YOLOv5 object detection network to identify dynamic objects, subsequently relaying the semantic information to the tracking module. During the tracking phase, both semantic information and epipolar constraints are leveraged to filter out dynamic feature points. To address the challenges posed by the malfunctioning of epipolar constraints in degenerate scenes, DEG-SLAM incorporates a degenerate constraint mechanism aimed at further eliminating dynamic feature points. Furthermore, a reprojection constraint has been introduced to enhance the filtering of absent dynamic feature points that lie outside the detection boxes. Experimental findings reveal that DEG-SLAM significantly improves accuracy and robustness when compared to traditional ORB-SLAM3 in dynamic environments. The performance benefits of DEG-SLAM are particularly evident in degenerate scenarios, thereby affirming its practicality and reliability in complex settings.

• Design a self-attention aggregation module using a pre-trained diffusion model. • Integrate high-precision masks into RGB-D SLAM for robust dynamic tracking. • Validate superior accuracy and efficiency on dynamic simulation datasets. In dynamic environments, performing RGB-D SLAM (Simultaneous Localization and Mapping) faces significant challenges primarily due to the presence of moving objects. The motion of these objects can introduce tracking errors and inaccuracies in map construction, thereby compromising the stability and overall performance of the system. To maintain high-precision localization and mapping under such conditions, a SLAM system must effectively detect and handle dynamic objects. To address these challenges, this paper presents a novel RGB-D SLAM method, referred to as DMSAA-SLAM (Dynamic Scene SLAM Based on Diffusion Model Self-Attention Aggregation). The core idea is to leverage a pre-trained stable diffusion model, particularly its self-attention layers, to handle the complexity of dynamic scenes. By employing a multi-resolution aggregation approach, combined with iterative merging and nonmaximum suppression, the proposed method generates high-precision segmentation masks. These masks enable fine-grained segmentation of moving objects and effectively eliminate dynamic feature points, thereby mitigating the impact of dynamic elements on the SLAM process and ensuring efficient and accurate tracking and mapping.

This thesis is about feature-based mobile robot navigation in unknown environments. The work is focusing on solving the problem of Simultaneous Localization and Mapping (SLAM) including feature extraction, estimation and complete map building solution. The robot platforms used in this work have the encoders for odometry and several laser scanners for exteroperceptive sensors. Simulations, real experiments and existing well known datasets are used in order to verify the performance of the developed algorithms. The algorithms developed in this thesis are features based meaning that instead of directly using raw sensory data geometric features are first extracted. This allows the filtering of noise from the sensors and the dynamics of the environment. More importantly, feature based approaches can reduce the amount of data needed to process, enabling the algorithms to scale in large environments. Another advantage of using features is the characteristic of better isolating informative patterns so that by exploiting specific knowledge of the environment, specialized algorithms can be easily adopted with different types of features. Several geometric features (points, lines, planes) are considered in this work. In this thesis, the relative mapping approach is adopted for solving the SLAM problem. This approach has many advantages. First, the mapping is decoupled from the estimation of robot pose so that the map estimation is independent from odometry errors. In other words, odometry errors do not affect and accumulate to the map estimate as in the absolute mapping approach. Second, the estimation does not involve the dynamics model of the robot which is identity in a relative mapping algorithm, meaning that only the observation model is considered in the estimation. A relative map has extra consistency problem compared to an absolute map. It is subject to the geometrical consistency problem which may generate an inconsistent absolute map after an transformation. However, having only observation model in the estimation gives an insight to separating nonlinearities from the iterative part of the estimator and thus minimize the effects of linearization errors. In this work, these subjects are investigated and reported. For some service robotic applications, especially for indoor office-like environments, we would like to have a fast, robust and lightweight approach which is capable to perform real-time SLAM on minimal hardware platforms. By exploiting some known knowledge about the structure of the environment, we claim to be able to achieve that goal. Particularly, the OrthoSLAM algorithm has been developed for building 2D and 3D maps. The algorithm adopts the Orthogonality assumption (or constraint) on the environment. First, the complexity of the SLAM algorithm is significantly reduced. Second, the linearization error from the robot heading uncertainty is removed and thus the map consistency is greatly improved. Implementations of the developed algorithms are tested on simulated data, on existing real world datasets and on robot platforms. Existing algorithms used in this work have been also reimplemented. The performance of the algorithms are analysed and compared on well chosen criteria. Particularly, the issues on computational complexity, linearization error, map consistency, map precision, realtime capability are investigated.

The mobile robot which accomplishes a work in explored region does not know location information of surroundings. Traditionally, simultaneous localization and mapping(SLAM) algorithms solve the localization and mapping problem in explored regions. Among the several SLAM algorithms, the EKF (Extended Kalman Filter) based SLAM is the scheme most widely used. The EKF is the optimal sensor fusion method which has been used for a long time. The odometric error caused by an encoder can be compensated by an EKF, which fuses different types of sensor data with weights proportional to the uncertainty of each sensor. In many cases the EKF based SLAM requires artificially installed features, which causes difficulty in actual implementation. Moreover, the computational complexity involved in an EKF increases as the number of features increases. And SLAM is a weak point of long operation time. Therefore, this paper presents a symmetrical model based SLAM algorithm (called M-SLAM).

In the automatic driving system, the external environment of the vehicle is perceived, in fact, there is also a perception sensor that has been silently dedicated in the system, that is, the positioning module. This paper explores the role of SLAM (Simultaneous Localization and Mapping) technology in autonomous vehicles, particularly in automatic lane change behavior prediction and environment perception. It emphasizes the limitations of traditional methods and the advantages of SLAM, especially visual SLAM, for accurate positioning and mapping. The discussion covers SLAM fundamentals, challenges, and the significance of visual SLAM's higher perception ability. Case studies on Waymo and Tesla illustrate the application of visual SLAM in achieving high-precision navigation and lane change prediction. The paper concludes by highlighting future research directions to enhance the intelligence and adaptability of automated lane change systems through advancements in AI and sensor technology, alongside optimizing SLAM algorithms for reliable driving in various scenarios.

This paper describes an approach that combines the navigation data given by a Doppler Velocity Logs (DVL), the MTi Motion Reference Unit (MRU) and a Mechanically Scanned Imaging Sonar (MSIS) as a principal sensor to efficiently solve underwater Simultaneous Localization and Mapping (SLAM) problem in structured environments such as marine platforms, harbors, or dams, etc. The MSIS has been chosen of its capacity to produce a rich representation of the environment. In recent years, to solve the SLAM Autonomous Underwater Vehicle (AUV) problem, very few solutions have been proposed. Our contribution has introduced a method based on the Nonlinear H-infinity filter \((NH\infty )\) to solve the SLAM-AUV problem. In this work, the \(NH\infty \)-SLAM algorithm is implemented to construct a map in partially structured environments and localize the AUV within this map. The data-set used in this paper are taken from the experiments carried out in a marina located in the Costa Brava (Spain) with the Ictineu AUV which is necessary to test different SLAM algorithms. The validation of the proposed algorithm through simulation in offline is presented and compared to the EKF-SLAM algorithm. The \(NH\infty \)-SLAM algorithm provides an accurate estimate than EKF-SLAM and good results were obtained.

Simultaneous localization and mapping (SLAM) is a central and complex problem in robot research community. In SLAM, extended Kalman filter (EKF) implementation is widely used to localize the robot and build the environment map incrementally. In this paper, we propose a strong tracking filter (STF) SLAM algorithm. This algorithm applies STF to deal with the non-linear estimated problem in SLAM instead of EKF. It can make the performance of the nonlinear filter approximate to that of optimal linear Kalman Filter (KF), so it can construct high accuracy maps and locate the robot more accurately than EKF SLAM. Simulation experiments illustrate the superior performance of our approach compared to EKF SLAM algorithm.

The calibration problem of binocular stereo vision rig is critical for its practical application. However, most existing calibration methods are based on manual off-line algorithms for specific reference targets or patterns. In this paper, we propose a novel simultaneous localization and mapping (SLAM)-based self-calibration method designed to achieve real-time, automatic and accurate calibration of the binocular stereo vision (BSV) rig’s extrinsic parameters in a short period without auxiliary equipment and special calibration markers, assuming the intrinsic parameters of the left and right cameras are known in advance. The main contribution of this paper is to use the SLAM algorithm as our main tool for the calibration method. The method mainly consists of two parts: SLAM-based construction of 3D scene point map and extrinsic parameter calibration. In the first part, the SLAM mainly constructs a 3D feature point map of the natural environment, which is used as a calibration area map. To improve the efficiency of calibration, a lightweight, real-time visual SLAM is built. In the second part, extrinsic parameters are calibrated through the 3D scene point map created by the SLAM. Ultimately, field experiments are performed to evaluate the feasibility, repeatability, and efficiency of our self-calibration method. The experimental data shows that the average absolute error of the Euler angles and translation vectors obtained by our method relative to the reference values obtained by Zhang’s calibration method does not exceed 0.5˚ and 2 mm, respectively. The distribution range of the most widely spread parameter in Euler angles is less than 0.2˚ while that in translation vectors does not exceed 2.15 mm. Under the general texture scene and the normal driving speed of the mobile robot, the calibration time can be generally maintained within 10 s. The above results prove that our proposed method is reliable and has practical value.

Typical visual simultaneous localization and mapping (SLAM) systems rely on purely geometric elements. Recently semantic SLAM has become popular due to its capability to obtain high-level understanding of the environment. However, most semantic SLAM systems merely aim to produce semantic maps rather than exploiting semantic constraints to boost the accuracy performance. In this paper, we propose a new RGB-D SLAM system for the use in indoor environments. Given that a ground plane represents a significant part of most indoor scenes, our new system exploit a semantic ground plane constraint inferred from a deep neural network along with geometric constraints on the state estimation to improve the performance. More specifically, we introduce a semantically meaningful geometric primitive, namely, a global ground plane to the optimization process of visual SLAM. The global ground model is able to provide an extra constraint on the state estimation through which both frame-to-keyframe and frame-to-model tracking processes are combined, and low-texture ground plane is effectively utilized. Experimental results for the TUM dataset and a real scene have demonstrated the improved performance brought by the semantic constraint in our method.