JLJiansheng Li

Jiansheng Li

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Jiansheng Li has a total of 8 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A Resilient Method for Visual–Inertial Fusion Based on Covariance Tuning

    by Kailin Li, Jiansheng Li, Ancheng Wang, Haolong Luo, Xueqiang Li, Zidi Yang

    To improve localization and pose precision of visual–inertial simultaneous localization and mapping (viSLAM) in complex scenarios, it is necessary to tune the weights of the visual and inertial inputs during sensor fusion. To this end, we propose a resilient viSLAM algorithm based on covariance tuning. During back-end optimization of the viSLAM process, the unit-weight root-mean-square error (RMSE) of the visual reprojection and IMU preintegration in each optimization is computed to construct a covariance tuning function, producing a new covariance matrix. This is used to perform another round of nonlinear optimization, effectively improving pose and localization precision without closed-loop detection. In the validation experiment, our algorithm outperformed the OKVIS, R-VIO, and VINS-Mono open-source viSLAM frameworks in pose and localization precision on the EuRoc dataset, at all difficulty levels.


  • Open Access

    A MAPPING METHOD OF SLAM BASED ON LOOK UP TABLE

    by Z. Wang, J. Li, A. Wang, J. Wang

    Keywords: V-SLAM, 3D modelling, Indoor Mapping, Scene Reconstruction, LUT, Eight-Neighbourhood Direction In the last years several V-SLAM(Visual Simultaneous Localization and Mapping) approaches have appeared showing impressive reconstructions of the world. However these maps are built with far more than the required information. This limitation comes from the whole process of each key-frame. In this paper we present for the first time a mapping method based on the LOOK UP TABLE(LUT) for visual SLAM that can improve the mapping effectively. As this method relies on extracting features in each cell divided from image, it can get the pose of camera that is more representative of the whole key-frame. The tracking direction of key-frames is obtained by counting the number of parallax directions of feature points. LUT stored all mapping needs the number of cell corresponding to the tracking direction which can reduce the redundant information in the key-frame, and is more efficient to mapping. The result shows that a better map with less noise is build using less than one-third of the time. We believe that the capacity of LUT efficiently building maps makes it a good choice for the community to investigate in the scene reconstruction problems.