ClusterVO: Clustering Moving Instances and Estimating Visual Odometry for Self and Surroundings

Huang, Jiahui; Yang, Songfan; Mu, Tai‐Jiang; Hu, Shi‐Min

doi:10.1109/cvpr42600.2020.00224

Cited by 84 publications

(67 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, some inevitable changes, lights, people moving, and object moving can cause the loss of position and posture of mobile robots [5] and, in turn, lead to inaccurate mapping. In order to deal with some dynamic objects that may exist in large scenes, the current algorithm usually eliminates dynamic objects by adding object detection or image segmentation algorithm based on deep learning in the system, such as Dynamic SLAM [6], Cluster VO [7], and Cluster SLAM [8] algorithm [9]. However, this will inevitably lead to a large consumption of computing resources, and some microdynamic objects are usually distributed near the camera.…”

Section: Introductionmentioning

confidence: 99%

A 3D Reconstruction Method Using Multisensor Fusion in Large-Scale Indoor Scenes

Zhou

Yu³

et al. 2020

Complexity

View full text Add to dashboard Cite

RGBD camera-based VSLAM (Visual Simultaneous Localization and Mapping) algorithm is usually applied to assist robots with real-time mapping. However, due to the limited measuring principle, accuracy, and distance of the equipped camera, this algorithm has typical disadvantages in the large and dynamic scenes with complex lightings, such as poor mapping accuracy, easy loss of robot position, and much cost on computing resources. Regarding these issues, this paper proposes a new method of 3D interior construction, which combines laser radar and an RGBD camera. Meanwhile, it is developed based on the Cartographer laser SLAM algorithm. The proposed method mainly takes two steps. The first step is to do the 3D reconstruction using the Cartographer algorithm and RGBD camera. It firstly applies the Cartographer algorithm to calculate the pose of the RGBD camera and to generate a submap. Then, a real-time 3D point cloud generated by using the RGBD camera is inserted into the submap, and the real-time interior construction is finished. The second step is to improve Cartographer loop-closure quality by the visual loop-closure for the sake of correcting the generated map. Compared with traditional methods in large-scale indoor scenes, the proposed algorithm in this paper shows higher precision, faster speed, and stronger robustness in such contexts, especially with complex light and dynamic objects, respectively.

show abstract

Section: Introductionmentioning

confidence: 99%

A 3D Reconstruction Method Using Multisensor Fusion in Large-Scale Indoor Scenes

Zhou

Yu³

et al. 2020

Complexity

View full text Add to dashboard Cite

show abstract

“…Key design choices of VO schemes can be classified based on the used visual sensor(s) and the selected processing modules, into geometric and nongeometric approaches. In the first VO method, camera geometrical relations are identified to estimate the ego-motion such as the intensity value of image pixels (appearance-based VO [36], [37], [43], [44], [48]- [50], [128]) and the image texture (feature-based VO [64], [65], [104]- [106], [108], [109], [111], [126], [137]). This method could provide precise state estimation only if enough features within the environment are observed in good lighting conditions.…”

Section: Discussion and Future Research Directionsmentioning

confidence: 99%

State of the Art in Vision-Based Localization Techniques for Autonomous Navigation Systems

2021

View full text Add to dashboard Cite

Vision-based localization systems, namely visual odometry (VO) and visual inertial odometry (VIO), have attracted great attention recently. They are regarded as critical modules for building fully autonomous systems. The simplicity of visual and inertial state estimators, along with their applicability in resource-constrained platforms motivated robotic community to research and develop novel approaches that maximize their robustness and reliability. In this paper, we surveyed state-of-the-art VO and VIO approaches. In addition, studies related to localization in visually degraded environments are also reviewed. The reviewed VO techniques and related studies have been analyzed in terms of key design aspects including appearance, feature, and learning based approaches. On the other hand, research studies related to VIO have been categorized based on the degree and type of fusion process into loosely-coupled, semi-tightly coupled, or tightly-coupled approaches and filtering or optimization-based paradigms. This paper provides an overview of the main components of visual localization, key design aspects highlighting the pros and cons of each approach, and compares the latest research works in this field. Finally, a detailed discussion of the challenges associated with the reviewed approaches and future research considerations are formulated.

show abstract

“…A general solution is to leverage historical information and establish associations between this information and current observations. Huang et al [ 47 ] predicted cluster motion based on historical information during occlusion and associated it with re-detected observations. They then recovered the motion based on the information before and after occlusion.…”

Section: Low-level-feature-based Dynamic Slammentioning

confidence: 99%

“…Therefore, Yang and Scherer [ 14 ] employed sparse optical flows to handle dynamic associations without using point positions. Huang et al [ 47 ] elaborately established a probabilistic model to explore enhanced point-object associations for fast-moving objects. They proposed a heterogeneous CRF combining semantic, spatial, and motion information to associate features with landmarks and bounding boxes with clusters jointly, and then implemented the Kuhn-Munkres algorithm to match current clusters with previous clusters.…”

Section: Using High-level Features In Dynamic Slammentioning

confidence: 99%

A survey: which features are required for dynamic visual simultaneous localization and mapping?

Zheng

2021

Vis. Comput. Ind. Biomed. Art

View full text Add to dashboard Cite

In recent years, simultaneous localization and mapping in dynamic environments (dynamic SLAM) has attracted significant attention from both academia and industry. Some pioneering work on this technique has expanded the potential of robotic applications. Compared to standard SLAM under the static world assumption, dynamic SLAM divides features into static and dynamic categories and leverages each type of feature properly. Therefore, dynamic SLAM can provide more robust localization for intelligent robots that operate in complex dynamic environments. Additionally, to meet the demands of some high-level tasks, dynamic SLAM can be integrated with multiple object tracking. This article presents a survey on dynamic SLAM from the perspective of feature choices. A discussion of the advantages and disadvantages of different visual features is provided in this article.

show abstract

ClusterVO: Clustering Moving Instances and Estimating Visual Odometry for Self and Surroundings

Cited by 84 publications

References 39 publications

A 3D Reconstruction Method Using Multisensor Fusion in Large-Scale Indoor Scenes

A 3D Reconstruction Method Using Multisensor Fusion in Large-Scale Indoor Scenes

State of the Art in Vision-Based Localization Techniques for Autonomous Navigation Systems

A survey: which features are required for dynamic visual simultaneous localization and mapping?

Contact Info

Product

Resources

About