Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SharpSLAM: 3D Object-Oriented Visual SLAM with Deblurring for Agile Drones

Denis Davletshin, Iana Zhura, Vladislav Cheremnykh, Mikhail Rybiyanov, Aleksey Fedoseev, Dzmitry Tsetserukou

TL;DR

SharpSLAM algorithm has the potential to highly improve the quality of 3D reconstruction and segmentation in DSP-SLAM and to impact a wide range of fields, including robotics, autonomous vehicles, and augmented reality.

Abstract

The paper focuses on the algorithm for improving the quality of 3D reconstruction and segmentation in DSP-SLAM by enhancing the RGB image quality. SharpSLAM algorithm developed by us aims to decrease the influence of high dynamic motion on visual object-oriented SLAM through image deblurring, improving all aspects of object-oriented SLAM, including localization, mapping, and object reconstruction. The experimental results revealed noticeable improvement in object detection quality, with F-score increased from 82.9% to 86.2% due to the higher number of features and corresponding map points. The RMSE of signed distance function has also decreased from 17.2 to 15.4 cm. Furthermore, our solution has enhanced object positioning, with an increase in the IoU from 74.5% to 75.7%. SharpSLAM algorithm has the potential to highly improve the quality of 3D reconstruction and segmentation in DSP-SLAM and to impact a wide range of fields, including robotics, autonomous vehicles, and augmented reality.

SharpSLAM: 3D Object-Oriented Visual SLAM with Deblurring for Agile Drones

TL;DR

SharpSLAM algorithm has the potential to highly improve the quality of 3D reconstruction and segmentation in DSP-SLAM and to impact a wide range of fields, including robotics, autonomous vehicles, and augmented reality.

Abstract

The paper focuses on the algorithm for improving the quality of 3D reconstruction and segmentation in DSP-SLAM by enhancing the RGB image quality. SharpSLAM algorithm developed by us aims to decrease the influence of high dynamic motion on visual object-oriented SLAM through image deblurring, improving all aspects of object-oriented SLAM, including localization, mapping, and object reconstruction. The experimental results revealed noticeable improvement in object detection quality, with F-score increased from 82.9% to 86.2% due to the higher number of features and corresponding map points. The RMSE of signed distance function has also decreased from 17.2 to 15.4 cm. Furthermore, our solution has enhanced object positioning, with an increase in the IoU from 74.5% to 75.7%. SharpSLAM algorithm has the potential to highly improve the quality of 3D reconstruction and segmentation in DSP-SLAM and to impact a wide range of fields, including robotics, autonomous vehicles, and augmented reality.
Paper Structure (15 sections, 2 equations, 6 figures, 1 table)

This paper contains 15 sections, 2 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Visual SLAM
  4. Object-oriented SLAM
  5. 3D Reconstruction for UAV
  6. Deblurring techniques in SLAM and object detection
  7. System Overview: Hardware and Software
  8. Object-oriented SLAM
  9. Deblurring with Neural Network
  10. Experiments
  11. Dataset preparation
  12. DSP-SLAM parameters
  13. Monocular SLAM position and scale calibration
  14. Shape reconstruction validation
  15. Conclusion and Future Work

Figures (6)

  • Figure 1: Improvement in visual SLAM performance under the SharpSLAM with image deblurring algorithm.
  • Figure 2: The overview of SharpSLAM hardware architecture used for the experimental evaluation of the proposed algorithm. Arrows show direction of data exchange between modules.
  • Figure 3: Visual SLAM pipeline with SharpSLAM approach.
  • Figure 4: Results of DSP-SLAM reconstruction for different number of keyframes waited before start reconstruction. (a) Initial camera position. (b) No. keyframes = 15, reconstruction was too early. (c) No. keyframes = 50, improved reconstruction after observing all sides of the car
  • Figure 5: Schematic explanation of SLAM scale and position calibration.
  • ...and 1 more figures