DGM-VINS: Visual–Inertial SLAM for Complex Dynamic Environments With Joint Geometry Feature Extraction and Multiple Object Tracking

Song, Boyi; Yuan, Xianfeng; Ying, Zhiyuan; Yang, Biao; Song, Yun Heub; Zhou, Fengyu

doi:10.1109/tim.2023.3280533

Cited by 15 publications

(1 citation statement)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Literature [39] made improvements to YOLOv3 to make it lighter, and used geometric constraints and random RANSAC to distinguish dynamic features. Literature [40] uses epipolar geometry and optical flow joint constraints, applies spatial clustering method to distinguish the status of feature points, and proposes a time series instance segmentation algorithm, establishing the order to solve the instance segmentation problem in complex scenes. In literature [41], object detection technology is used to process images, following which epipolar geometry constraints are applied to calculate the distance of each feature point from the baseline.…”

Section: Related Workmentioning

confidence: 99%

OMS-SLAM: dynamic scene visual SLAM based on object detection with multiple geometric feature constraints and statistical threshold segmentation

Tang,

Feng,

Liao

et al. 2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

Simultaneous localization and mapping (SLAM) technology is crucial to robot navigation. Despite the good performance of traditional SLAM algorithms in static environments, dynamic objects typically exist in realistic operating environments. These objects can lead to misassociated features, which in turn considerably impact the system’s localization accuracy and robustness. To better address this challenge, we have proposed the OMS-SLAM. In OMS-SLAM, we adopted the YOLOv8 target detection network to extract object information from environment and designed a dynamic probability propagation model that is coupled with target detection and multiple geometric constrains to determine the dynamic objects in the environment. For the identified dynamic objects, we have designed a foreground image segmentation algorithm based on depth image histogram statistics to extract the object contours and eliminate the feature points within these contours. We then use the grid-based motion statistics matching pair as the filtering strategy to enhance the quality of the feature points and use the enhanced feature points for tracking. This combined method can accurately identify dynamic objects and extract related feature points, significantly reducing its interference and consequently enhancing the system’s robustness and localization accuracy. We also built static dense point cloud maps to support advanced tasks of robots. Finally, through testing on the high-speed dataset of TUM RGB-D, it was found that the root mean square error of the absolute trajectory error in this study decreased by an average of 97.10%, compared to ORB-SLAM2. Moreover, tests in real-world scenarios also confirmed the effectiveness of the OMS-SLAM algorithm in dynamic environments.

show abstract

Section: Related Workmentioning

confidence: 99%

OMS-SLAM: dynamic scene visual SLAM based on object detection with multiple geometric feature constraints and statistical threshold segmentation

Tang,

Feng,

Liao

et al. 2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

show abstract

Visual Inertial SLAM Based on Spatiotemporal Consistency Optimization in Diverse Environments

Pu,

Luo,

Wang

et al. 2024

Journal of Field Robotics

View full text Add to dashboard Cite

Currently, the majority of robots equipped with visual‐based simultaneous mapping and localization (SLAM) systems exhibit good performance in static environments. However, practical scenarios often present dynamic objects, rendering the environment less than entirely “static.” Diverse dynamic objects within the environment pose substantial challenges to the precision of visual SLAM system. To address this challenge, we propose a real‐time visual inertial SLAM system that extensively leverages objects within the environment. First, we reject regions corresponding to dynamic objects. Following this, geometric constraints are applied within the stationary object regions to elaborate the mask of static areas, thereby facilitating the extraction of more stable feature points. Second, static landmarks are constructed based on the static regions. A spatiotemporal factor graph is then created by combining the temporal information from the Inertial Measurement Unit (IMU) with the semantic information from the static landmarks. Finally, we perform a diverse set of validation experiments on the proposed system, encompassing challenging scenarios from publicly available benchmarks and the real world. Within these experimental scenarios, we compare with state‐of‐the‐art approaches. More specifically, our system achieved a more than 40% accuracy improvement over baseline method in these data sets. The results demonstrate that our proposed method exhibits outstanding robustness and accuracy not only in complex dynamic environments but also in static environments.

show abstract

Strong-SLAM: real-time RGB-D visual SLAM in dynamic environments based on StrongSORT

Huang,

Zou,

Yun

et al. 2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

The assumptions of a static environment and scene rigidity are important theoretical underpinnings of traditional visual Simultaneous Localization and Mapping (SLAM) algorithms. However, these assumptions are difficult to work in dynamic environments containing non-rigid objects, and cannot effectively handle the characteristics of local areas of non-rigid moving objects, seriously affecting the robustness and accuracy of the SLAM system in localization and mapping. To address these problems, we improved ORB-SLAM3 and proposed a real-time RGB-D visual SLAM framework for dynamic environments based on StrongSORT - Strong-SLAM. First, we combine YOLOv7-tiny with StrongSORT to match the semantic information of dynamic targets. Optical flow and epipolar constraints are then used to initially extract geometric and motion information between adjacent frames. Subsequently, based on an improved adaptive threshold segmentation algorithm and geometric residuals, a background model and a Gaussian residual model are constructed to further extract the geometric information of dynamic targets. Finally, semantic and geometric information are integrated to perform global feature motion level classification, and motion probabilities and optimization weights are defined to participate in global pose estimation and optimization. Experimental results on the publicly available TUM RGB-D dataset show that Strong-SLAM reduces the Absolute Trajectory Error (ATE) and Relative Pose Error (RPE) by at least 90% compared to ORB-SLAM3, achieving performance comparable to the most advanced dynamic SLAM solutions.

show abstract

DGM-VINS: Visual–Inertial SLAM for Complex Dynamic Environments With Joint Geometry Feature Extraction and Multiple Object Tracking

Cited by 15 publications

References 36 publications

OMS-SLAM: dynamic scene visual SLAM based on object detection with multiple geometric feature constraints and statistical threshold segmentation

OMS-SLAM: dynamic scene visual SLAM based on object detection with multiple geometric feature constraints and statistical threshold segmentation

Visual Inertial SLAM Based on Spatiotemporal Consistency Optimization in Diverse Environments

Strong-SLAM: real-time RGB-D visual SLAM in dynamic environments based on StrongSORT

Contact Info

Product

Resources

About