Search

When using deep learning networks for dynamic feature rejection in SLAM systems, problems such as a priori static object motion leading to disturbed build quality and accuracy and slow system runtime are prone to occur. In this paper, based on the ORB-SLAM2 system, we propose a method based on improved YOLOv5 networks combined with geometric constraint methods for SLAM map building in dynamic environments. First, this paper uses ShuffleNetV2 to lighten the YOLOv5 network, which increases the improved network's operation speed without reducing the accuracy. At the same time, a pyramidal scene parsing network segmentation head is added to the head part of the YOLOv5 network to achieve semantic extraction in the environment, so that the improved YOLOv5 network has both target detection and semantic segmentation functions. In order to eliminate the objects with low dynamic features in the environment, this paper adopts the method of geometric constraints to extract and eliminate the dynamic features of the low dynamic objects. By combining the improved YOLOv5 network with the geometric constraint method, the robustness of the system is improved and the interference of dynamic targets in the construction of the SLAM system map is eliminated. The test results on the TUM dataset show that, when constructing a map in a dynamic environment, compared with the traditional ORB-SLAM2 algorithm, the accuracy of map construction in a dynamic environment is significantly improved. The absolute trajectory error is reduced by 97.7% compared with ORB-SLAM2, and the relative position error is reduced by 59.7% compared with ORB-SLAM2. Compared with DynaSLAM for dynamic scenes of the same type, the accuracy of map construction is slightly improved, but the maximum increase in keyframe processing time is 94.7%. [ABSTRACT FROM AUTHOR]

Visual Simultaneous Localization and Mapping (VSLAM) is a critical foundation in mobile robotics and augmented reality (AR). However, VSLAM faces challenges in dynamic environments since both the camera and the object are in motion, which contradicts the classical static scene assumption. Generally, multi-view geometry is employed for static features to estimate camera pose and reconstruct environment maps. Hence, dynamic feature detection and data association become key issues in dynamic VSLAM. To solve these problems, we propose an innovative probability-based approach that combines instance segmentation and nonparametric Kolmogorov–Smirnov test approaches to detect the distribution of features on an object. Furthermore, we propose a data association algorithm based on the Bayesian model that comprehensively utilizes the descriptors of feature points and their spatial information. Experiments on the KITTI public dataset and the Oxford Multi-motion dataset validate the effectiveness of our method. • The dynamic features introduce errors to visual SLAM. • Dynamic features can be detected via a nonparametric Kolmogorov-Smirnov test method. • A Bayesian model-based data association algorithm enhance the camera localization. [ABSTRACT FROM AUTHOR]

Summary: Most conventional simultaneous localization and mapping (SLAM) approaches assume the working environment to be static. In a highly dynamic environment, this assumption divulges the impediments of a SLAM algorithm that lack modules that distinctively attend to dynamic objects despite the inclusion of optimization techniques. This work exploits such environments and reduces the effects of dynamic objects in a SLAM algorithm by separating features belonging to dynamic objects and static background using a generated binary mask image. While the features belonging to the static region are used for performing SLAM, the features belonging to non-static segments are reused instead of being eliminated. The approach employs deep neural network or DNN-based object detection module to obtain bounding boxes and then generates a lower resolution binary mask image using depth-first search algorithm over the detected semantics, characterizing the segmentation of the foreground from the static background. In addition, the features belonging to dynamic objects are tracked into consecutive frames to obtain better masking consistency. The proposed approach is tested on both publicly available dataset as well as self-collected dataset, which includes both indoor and outdoor environments. The experimental results show that the removal of features belonging to dynamic objects for a SLAM algorithm can significantly improve the overall output in a dynamic scene. [ABSTRACT FROM AUTHOR]

This paper presents an innovative method for motion segmentation in RGB-D dynamic videos with multiple moving objects. The focus is on finding static, small or slow moving objects (often overlooked by other methods) that their inclusion can improve the motion segmentation results. In our approach, semantic object based segmentation and motion cues are combined to estimate the number of moving objects, their motion parameters and perform segmentation. Selective object-based sampling and correspondence matching are used to estimate object specific motion parameters. The main issue with such an approach is the over segmentation of moving parts due to the fact that different objects can have the same motion (e.g. background objects). To resolve this issue, we propose to identify objects with similar motions by characterizing each motion by a distribution of a simple metric and using a statistical inference theory to assess their similarities. To demonstrate the significance of the proposed statistical inference, we present an ablation study, with and without static objects inclusion, on SLAM accuracy using the TUM-RGBD dataset. To test the effectiveness of the proposed method for finding small or slow moving objects, we applied the method to RGB-D MultiBody and SBM-RGBD motion segmentation datasets. The results showed that we can improve the accuracy of motion segmentation for small objects while remaining competitive on overall measures. [ABSTRACT FROM AUTHOR]

In this paper, we present a method for single image three-dimensional (3-D) cuboid object detection and multiview object simultaneous localization and mapping in both static and dynamic environments, and demonstrate that the two parts can improve each other. First, for single image object detection, we generate high-quality cuboid proposals from two-dimensional (2-D) bounding boxes and vanishing points sampling. The proposals are further scored and selected based on the alignment with image edges. Second, multiview bundle adjustment with new object measurements is proposed to jointly optimize poses of cameras, objects, and points. Objects can provide long-range geometric and scale constraints to improve camera pose estimation and reduce monocular drift. Instead of treating dynamic regions as outliers, we utilize object representation and motion model constraints to improve the camera pose estimation. The 3-D detection experiments on SUN RGBD and KITTI show better accuracy and robustness over existing approaches. On the public TUM, KITTI odometry and our own collected datasets, our SLAM method achieves the state-of-the-art monocular camera pose estimation and at the same time, improves the 3-D object detection accuracy. [ABSTRACT FROM AUTHOR]

Motion segmentation is important for autonomous robots' activities in dynamic scenes, which has been a challenging problem due to the dual motion caused by the camera and moving objects. The existing methods based on the optical or scene flow are sensitive to the environment, whereas deep learning-based methods commonly do not consider geometric constraints. Aiming at a vehicle, as the most common object in outdoor urban dynamic scenes because other objects, such as people, can be almost directly determined as a moving object, this paper proposes an object-level motion segmentation method for multiple moving objects combining geometric methods and deep learning. First, we choose three kinds of learned information, constructing object-level three-dimensional (3D) scene structure. And then, to infer motion segmentation results from learned information, we propose an instance cross-matching method to obtain instance correspondence and a motion segmentation extraction method based on instance reprojection residual to determine whether the objects are moving or not. The proposed approach is simple and direct, and it can obtain vehicle motion segmentation results from the original RGB images without using the optical or scene flow commonly used to extract motion information. Experiments on public motion segmentation datasets demonstrate that the proposed method can effectively improve the performance of vehicle motion segmentation and is superior to the most advanced methods in terms of accuracy. [ABSTRACT FROM AUTHOR]

This paper presents a survey of Simultaneous Localization And Mapping (SLAM) algorithms for unmanned ground robots. SLAM is the process of creating a map of the environment, sometimes unknown a priori, while at the same time localizing the robot in the same map. The map could be one of different types i.e. metrical, topological, hybrid or semantic. In this paper, the classification of algorithms is done in three classes: (i)Metric map generating approaches, (ii) Qualitative map generating approaches, and (iii) Hybrid map generating approaches. SLAM algorithms for both static and dynamic environments have been surveyed. The algorithms in each class are further divided based on the techniques used. The survey in this paper presents the current state-of-the-art methods, including important landmark works reported in the literature. [ABSTRACT FROM AUTHOR]

The idea of SLAM (Simultaneous Localization and Mapping) being a solved problem revolves around the static world assumption, even though autonomous systems are gaining environmental perception capabilities by exploiting the advances in computer vision and data-driven approaches. The computational demands and time complexities remain the main impediment in the effective fusion of the paradigms. In this paper, a framework to solve the dynamic SLAM problem is proposed. The dynamic regions of the scene are handled by making use of Visual-LiDAR based MODT (Multiple Object Detection and Tracking). Furthermore, minimal computational demands and real-time performance are ensured. The framework is tested on the KITTI Datasets and evaluated against the publicly available evaluation tools for a fair comparison with state-of-the-art SLAM algorithms. The results suggest that the proposed dynamic SLAM framework can perform in real-time with budgeted computational resources. In addition, the fused MODT provides rich semantic information that can be readily integrated into SLAM. [ABSTRACT FROM AUTHOR]