QSQinxuan Sun

Qinxuan Sun

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Qinxuan Sun has a total of 19 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Adaptive Soft Encoding: A General Unsupervised Feature Aggregation Method for Place Recognition

    by Yansong Gong, Jing Yuan, Fengchi Sun, Qinxuan Sun, Wenbin Zhu, Xuebo Zhang

    Place recognition is an important problem in environment perception, which can help simultaneous localization and mapping (SLAM), as well as scene understanding of mobile robots. For place recognition, feature aggregation plays an important role in the representation of environments. In this paper, an adaptive soft-encoding (ASE) method is proposed for aggregating numerous features into low-dimensional feature vectors, which includes a training phase and an encoding phase. For the training phase, the information along each dimension of the features is evaluated which is further employed to assign all dimensions into different subdimensional intervals. After that, subdimensional Gaussian mixture models (SD-GMMs) are fitted by features in the subdimensional intervals and organized into an ASE tree, of which the tree nodes hierarchically store the parameters of the SD-GMMs to reflect the distribution of the features. For the encoding phase, features are fed into the root node of the ASE tree. The weight of each node is calculated from top to bottom to generate multiple aggregated feature vectors with different description capabilities. In addition, an ASE-based framework for place recognition is proposed in which local descriptors are extracted from sensor data (e.g., images or LiDAR data), and then aggregated into global descriptors by ASE. Finally, owing to the hierarchical structure of the ASE tree, a hierarchical matching strategy of the global descriptors is designed to recognize places efficiently. Experimental results demonstrate that feature vectors aggregated by ASE have strong distinguishability, and the ASE-based place recognition method achieves good accuracy and efficiency. The code of ASE can be accessed at https://github.com/wdyiwdwd/ASE-Encoder.


  • VIDO: A Robust and Consistent Monocular Visual-Inertial-Depth Odometry

    by Yuanxi Gao, Jing Yuan, Jingqi Jiang, Qinxuan Sun, Xuebo Zhang

    Multi-sensor fusion is a mainstream method for localization of unmanned systems. How to achieve 6-degrees of freedom (DOF) pose estimation of the system is challenging in GPS-denied environments. Although map-aided localization methods normally perform well on intelligent transportation systems, prior maps are unavailable in some GPS-denied scenes (e.g., dense forests, tunnels, and underground parking lots). In this paper, we present a robust and consistent monocular visual-inertial-depth odometry (VIDO) to perform 6-DOF pose estimation without the need of prior information. The system contains a visual-inertial subsystem (VIS) based on tightly coupled optimization in a sliding window and a depth subsystem (DS) based on the iterative closest point (ICP) estimation using 3D point clouds obtained by a LiDAR or depth camera. The uncertainties of the estimation results in VIS and DS are rigorously calculated to consider measurement noises of the sensors. The obtained uncertainty estimates are fed into a covariance intersection (CI) filter for pose fusion, and the fused pose is further refined in the mapping process. We perform experiments on public datasets, as well as in various real-world outdoor and indoor scenes to verify the performance on localization and mapping in urban areas with buildings and cars, off-road environments with rugged terrains, as well as indoor structured environments. The results show that the proposed method can provide both a robust 6-DOF pose estimate and a precise 3D map for fully autonomous navigation in different scenes without a prior map, which presents an attractive complement to map-aided automated driving.


  • ORB-TEDM: An RGB-D SLAM Approach Fusing ORB Triangulation Estimates and Depth Measurements

    by Jing Yuan, Shuhao Zhu, Kaitao Tang, Qinxuan Sun

    3-D position estimates of feature points in traditional RGB-D simultaneous localization and mapping (SLAM) systems are directly obtained by depth measurements. However, the available information provided by the triangulation of feature points has not been involved. In this article, a novel RGB-D SLAM approach is proposed based on the ORB-SLAM2 system, by fusing the triangulation estimates and depth measurements of ORB features, termed ORB-TEDM. Specifically, uncertainties on both the pose estimate of the camera and triangulation of ORB features are rigorously computed in a closed form for the ORB-SLAM2 system using covariance propagation and implicit differentiation. On the other hand, the uncertainty is evaluated for the 3-D position estimate of each map point obtained by its depth measurement from the RGB-D camera. Then, the triangulation results and position estimates from depth measurements are further fused with the covariance intersection (CI) filter to generate a more precise and consistent map. In addition, a more flexible selection policy of map points and keyframes is designed based on the uncertainty evaluation results. As a consequence, a more accurate and robust RGB-D SLAM system can be achieved. It is worthwhile to point out that the obtained closed-form solution to uncertainties on estimates of the camera pose and feature points can facilitate further development of the ORB-SLAM framework, such as active SLAM and multisensor SLAM. The experimental results on public datasets and in real-world environments are presented to show the effectiveness of the proposed approach.


  • IT-HYFAO-VO: Interpretation Tree-Based VO With Hybrid Feature Association and Optimization

    by Qinxuan Sun, Jing Yuan, Xuebo Zhang

    For the visual odometry (VO) and simultaneous localization and mapping (SLAM) in indoor environments, high-level geometric features have attracted more and more attention in recent years. Unlike the generally used point features, the geometric features, such as planes and lines, encode more higher-level semantic information of the scene which is beneficial for various tasks of mobile robots. In this article, an RGB-D VO system with hybrid high-level geometric features is developed. An interpretation tree (IT)-based hybrid feature association framework is proposed, which turns the feature association into a multiple-hypothesis decision problem. The IT expansion method is elaborately designed. Specifically, an internode consistency is proposed for generation of hypotheses and a consistent transformation model (CTM) for each hypothesis is explicitly expressed and incrementally updated. When the IT is constructed, a closed-form solution to the feature association and the camera transformation can be obtained. Then, a hybrid feature joint optimization method is introduced to further refine the pose estimate and parameters of geometric features. During optimization, the planes and lines are appropriately parameterized and the uncertainties arising from feature extraction are derived and used to balance the contributions of two types of features in the cost function. Extensive experiments are executed on public datasets and the results demonstrate that the proposed method can achieve higher accuracy and stronger robustness.


  • Open Access

    Moving Target Tracking with a Mobile Robot based on Modified Social Force Model

    by Qinxuan Sun, Shengming Zhang, Jing Yuan, Xuebo Zhang, Shuhao Zhu

    A target tracking approach is proposed for mobile robots in this paper to address the human-robot coexistence and collaboration problem. The improved social force model (SFM) is applied to improve the tracking performance of the robot in crowded environments. When the robot approaches the pedestrians or obstacles, the tracking strategy is adaptively adjusted to avoid collision. The inverse reinforcement learning (IRL) is used to learn the parameters of the improved SFM, where the training data for the IRL is collected in real-world scenes. An effective criterion is designed to evaluate the tracking performance, which fully considers the relationship between the robot and surrounding environments. The experimental results demonstrate the effectiveness of the proposed target tracking method.


  • Plane-Edge-SLAM: Seamless Fusion of Planes and Edges for SLAM in Indoor Environments

    by Qinxuan Sun, Jing Yuan, Xuebo Zhang, Feng Duan

    Planes and edges are attractive features for simultaneous localization and mapping (SLAM) in indoor environments because they can be reliably extracted and are robust to illumination changes. However, it remains a challenging problem to seamlessly fuse two different kinds of features to avoid degeneracy and accurately estimate the camera motion. In this article, a plane-edge-SLAM system using an RGB-D sensor is developed to address the seamless fusion of planes and edges. Constraint analysis is first performed to obtain a quantitative measure of how the planes constrain the camera motion estimation. Then, using the results of the constraint analysis, an adaptive weighting algorithm is elaborately designed to achieve seamless fusion. Through the fusion of planes and edges, the solution to motion estimation is fully constrained, and the problem remains well-posed in all circumstances. In addition, a probabilistic plane fitting algorithm is proposed to fit a plane model to the noisy 3-D points. By exploiting the error model of the depth sensor, the proposed plane fitting is adaptive to various measurement noises corresponding to different depth measurements. As a result, the estimated plane parameters are more accurate and robust to the points with large uncertainties. Compared with the existing plane fitting methods, the proposed method definitely benefits the performance of motion estimation. The results of extensive experiments on public data sets and in real-world indoor scenes demonstrate that the plane-edge-SLAM system can achieve high accuracy and robustness.


  • Fusing Skeleton Recognition With Face-TLD for Human Following of Mobile Service Robots

    by Jing Yuan, Jingxin Cai, Xuebo Zhang, Qinxuan Sun, Fengchi Sun, Wenbin Zhu

    Target recognition is a challenging task for human following of mobile service robots. In this paper, we combine the principal-component-analysis (PCA)-based face recognition with the tracking-learning-detection applied to the human face (Face-TLD) to obtain an improvement, named as IFace-TLD. The proposed IFace-TLD can significantly improve the discrimination ability of the Face-TLD for ambiguous facial appearances. To further deal with motion uncertainties of the human head, especially the sudden motion change, which makes face-based target recognition methods unstable or even loses the target, a skeleton-based model is introduced to improve the accuracy and robustness of the target recognition. Specifically, within a walk half-cycle, the skeleton features are extracted from the upper-body three-dimensional skeleton coordinates. Then, the extracted skeleton features are fed into the support vector data description (SVDD) to identify the target person when the IFace-TLD becomes invalid. The seamless fusion of the skeleton recognition and the IFace-TLD, named as the SIFace-TLD, significantly enhances the robustness in complex scenarios, especially for people tracking from both front and behind. To achieve a complete human following system, the particle filter (PF) is adopted for estimating the state of the human motion. And then, a controller is designed to maintain the relative position between the robot and the target. Experimental results demonstrate that the proposed IFace-TLD is more accurate and flexible than the original Face-TLD. And the SIFace-TLD shows a robust performance to human motion uncertainties. Moreover, the developed controller can achieve a satisfactory human following performance.


  • Laser-Based Intersection-Aware Human Following With a Mobile Robot in Indoor Environments

    by Jing Yuan, Shengming Zhang, Qinxuan Sun, Gangdun Liu, Jingxin Cai

    Human following in structured indoor environments has to face the challenge of full occlusion caused by the walls when the target person makes a turn at the corridor intersections. This may result in short-term, and even permanent loss of the target from the field of view of the robot. In this paper, human following with a mobile robot in presence of potential occlusions occurring at corridor intersections is addressed. The robot detects four different types of corridor intersections using the on-board laser scanner. Then, a potential-field-based human tracker is designed by integrating the intersection information into the potential function, in order to increase the visibility of the target, while maintaining the relative distance and orientation between the target and the robot. Simultaneously, the robot builds a multihypothesis topological map of the environment based on an improved generalized Voronoi graph, where the edges represent the corridors and the vertices are the virtual meet points extracted from the intersections. In such a way, simultaneous human following and topological mapping are achieved. Simulation and experimental results show that the proposed method can largely avoid occlusion of the target and obtain good performance of human following and topological mapping.


  • RGB-D SLAM in Indoor Environments With STING-Based Plane Feature Extraction

    by Qinxuan Sun, Jing Yuan, Xuebo Zhang, Fengchi Sun

    In this paper, the RGB-D camera-based simultaneous localization and mapping (SLAM) of indoor environments is addressed using plane features. The plane parameter space (PPS) is defined for a compact representation of planes in the Cartesian space. The statistical information grid (STING) structure is constructed in the PPS to extract plane features. The plane association graph (PAG) is developed to determine the correspondences between the plane features from two successive scans. The RGB-D camera pose is directly calculated using the matched plane features if they can provide sufficient constraints for the pose estimation. Otherwise, a novel STING-based scan matching method is developed in the PPS to achieve a full six Degrees of Freedom (6-DoF) camera pose estimation. The proposed method uses only the plane features independent of any other features to estimate the RGB-D camera poses and can thus be suitable for some challenging scenes. The experimental results demonstrate that the proposed plane feature-based RGB-D SLAM can achieve high accuracy and robustness in both on-board and hand-held applications.