A Mono SLAM Method Based on Depth Estimation by DenseNet-CNN

Jin, Yifan; Yu, Lei; Chen, Zhong; Fei, Shumin

doi:10.1109/jsen.2021.3134014

Cited by 10 publications

(3 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…SLAM based on the direct method mainly includes intensive tracking and mapping (DTAM) and large-scale direct monocular SLAM (LSD-SLAM) [23]. For the MonoSLAM method proposed in literature [24] for the indirect SLAM system, the extended Kalman filter (EKF) method is used in the back-end optimization scheme of the algorithm. This method uses the method of minimizing reprojection error to estimate.…”

Section: State Of the Artmentioning

confidence: 99%

Research on Visual SLAM Navigation Techniques for Dynamic Environments

Wang,

Zhao

2023

International Journal of Distributed Sensor Networks

View full text Add to dashboard Cite

Synchronous positioning and mapping mainly realize the functions of self-positioning and environment map construction for intelligent navigation technology. In order to solve the problems of low positioning accuracy and poor mapping effect of existing SLAM (simultaneous localization and mapping) systems in indoor dynamic environments and to improve the positioning accuracy, timeliness, and robustness of visual SLAM systems in dynamic environments, an improved visual SLAM method is proposed. Aiming at the inconsistency between the direction of dynamic objects and static background optical flow, this method adopts a high-real-time dynamic region mask detection algorithm to eliminate the feature points in the dynamic region mask, remove the camera motion optical flow according to the original feature information, and then cluster the optical flow amplitude of dynamic objects so as to realize the dynamic region mask detection and eliminate the dynamic signpost points combined with the polar geometric constraints. In order to verify the effectiveness of the improved algorithm, the three evaluation indexes of system accuracy, real-time performance, and the amount of drift are analyzed and verified, respectively, on the TUM dataset. The results show that the proposed algorithm not only has good real-time performance but also improves the accuracy of the system and reduces the amount of drift.

show abstract

Section: State Of the Artmentioning

confidence: 99%

Research on Visual SLAM Navigation Techniques for Dynamic Environments

Wang,

Zhao

2023

International Journal of Distributed Sensor Networks

View full text Add to dashboard Cite

show abstract

“…Among them, depth estimation has received much attention to acquire 3D information from a monocular camera. The objective of depth estimation is to predict pixel-level distances from the camera coordinate, and it has been applied to augmented reality 1 , robot navigation 2 , and autonomous driving 3 . However, due to the scale-ambiguity in the inverse projection problem, it has many additional challenges compared to other computer vision tasks which infer 2D semantic information.…”

Section: Introductionmentioning

confidence: 99%

Knowledge Distillation of Multi-scale Dense Prediction Transformer for Self-supervised Depth Estimation

Song

Lee

2023

Preprint

View full text Add to dashboard Cite

Depth estimation is an inverse projection problem that estimates pixel-level distances from a single image. Although, supervised methods have shown promising results, it has intrinsic limitations in requiring ground truth depth from an external sensor. On the other hand, self-supervised depth estimation relieves the burden for collecting calibrated training data, while there is still a large performance gap between supervised and self-supervised methods. The objective of this study is to reduce the performance gap in supervised and self-supervised depth estimation. The loss function of previous self-supervised methods is mainly based on a photometric error, which is indirectly computed from synthesized images using depth and pose estimates. In this paper, we argue that direct depth cue is more effective to train a depth estimation network. To obtain the direct depth cue, we employed a knowledge distillation technique, which is a teacher-student learning framework. The teacher network was trained in a self-supervised manner based on a photometric error, and its predictions were utilized to train a student network. We constructed a multi-scale dense prediction transformer with Monte Carlo dropout, and multi-scale distillation loss was proposed to train the student network based on the ensemble of stochastic estimates. Experiments were conducted on the KITTI and Make3D datasets, and our proposed method achieved the state-of-the-art accuracy in self-supervised depth estimation. Our code is publicly available at https://github.com/ji-min-song/KD-of-MS-DPT.

show abstract

“…At present, there are two main methods based on vision and lidar. For example, the mono slam, the first monocular slam system proposed by Smith [ 2 ] and others, uses a Kalman filter as the backend to track the sparse feature points at the frontend. Another example is the orb designed by Mur et al The SLAM algorithm [ 3 ] solves the problem of cumulative error.…”

Section: Introductionmentioning

confidence: 99%

The Navigation System of a Logistics Inspection Robot Based on Multi-Sensor Fusion in a Complex Storage Environment

Zhang

Zhou

et al. 2022

Sensors

View full text Add to dashboard Cite

To reliably realize the functions of autonomous navigation and cruise of logistics robots in a complex logistics storage environment, this paper proposes a new robot navigation system based on vision and multiline lidar information fusion, which can not only ensure rich information and accurate map edges, but also meet the real-time and accurate positioning and navigation in complex logistics storage scenarios. Simulation and practical verification showed that the robot navigation system is feasible and robust, and overcomes the problems of low precision, poor robustness, weak portability, and difficult expansion of the mobile robot system in a complex environment. It provides a new idea for inspection in an actual logistics storage scenario and has a good prospective application.

show abstract

A Mono SLAM Method Based on Depth Estimation by DenseNet-CNN

Cited by 10 publications

References 33 publications

Research on Visual SLAM Navigation Techniques for Dynamic Environments

Research on Visual SLAM Navigation Techniques for Dynamic Environments

Knowledge Distillation of Multi-scale Dense Prediction Transformer for Self-supervised Depth Estimation

The Navigation System of a Logistics Inspection Robot Based on Multi-Sensor Fusion in a Complex Storage Environment

Contact Info

Product

Resources

About