[Paper Review] FastORB-SLAM: Fast ORB-SLAM Method with Descriptor Independent Keypoint Matching

Qiang Fu, Hongshan Yu+5 more|arXiv (Cornell University)|Aug 22, 2020

Robotics and Sensor-Based Localization69 references4 citations

TL;DR

FastORB-SLAM proposes a descriptor-independent keypoint matching method for RGB-D SLAM that accelerates ORB-SLAM2 by eliminating descriptor computation for non-keyframes. Using a two-stage coarse-to-fine approach based on sparse optical flow and motion/geometry constraints, it achieves state-of-the-art accuracy while running approximately twice as fast as ORB-SLAM2 on TUM and ICL-NUIM datasets.

ABSTRACT

Indirect methods for visual SLAM are gaining popularity due to their robustness to environmental variations. ORB-SLAM2 \cite{orbslam2} is a benchmark method in this domain, however, it consumes significant time for computing descriptors that never get reused unless a frame is selected as a this http URL overcome these problems, we present FastORB-SLAM which is lightweight and efficient as it tracks keypoints between adjacent frames without computing descriptors. To achieve this, a two stage coarse-to-fine descriptor independent keypoint matching method is proposed based on sparse optical flow. In the first stage, we predict initial keypoint correspondences via a simple but effective motion model and then robustly establish the correspondences via pyramid-based sparse optical flow tracking. In the second stage, we leverage the constraints of the motion smoothness and epipolar geometry to refine the correspondences. In particular, our method computes descriptors only for keyframes. We test FastORB-SLAM on extit{TUM} and extit{ICL-NUIM} RGB-D datasets and compare its accuracy and efficiency to nine existing RGB-D SLAM methods. Qualitative and quantitative results show that our method achieves state-of-the-art accuracy and is about twice as fast as the ORB-SLAM2.

Motivation & Objective

• To reduce computational overhead in ORB-SLAM2 by eliminating redundant descriptor computation for non-keyframes.
• To maintain high tracking accuracy despite descriptor-free matching in consecutive frames.
• To develop a lightweight, efficient visual SLAM system suitable for real-time deployment.
• To leverage motion smoothness and epipolar geometry to refine keypoint correspondences without descriptors.
• To achieve state-of-the-art performance in both accuracy and speed on standard RGB-D SLAM benchmarks.

Proposed method

• A two-stage coarse-to-fine keypoint matching pipeline is proposed: first, initial correspondences are predicted using a motion model.
• Initial matches are refined via pyramid-based sparse optical flow tracking for robustness to motion and illumination changes.
• Motion smoothness and epipolar geometry constraints are applied in the second stage to further refine correspondences.
• Descriptors are computed only for keyframes, significantly reducing computational load during tracking.
• The method relies on sparse optical flow and geometric constraints to maintain accuracy without descriptor computation in non-keyframes.
• The system is evaluated on TUM and ICL-NUIM RGB-D datasets using standard metrics for accuracy and efficiency.

Experimental results

Research questions

• RQ1 Can descriptor computation be avoided in non-keyframes without degrading tracking accuracy in visual SLAM?
• RQ2 How effective is a motion model combined with sparse optical flow for initial keypoint correspondence prediction?
• RQ3 To what extent do motion smoothness and epipolar geometry constraints improve matching robustness in descriptor-free tracking?
• RQ4 Can a descriptor-independent approach achieve state-of-the-art accuracy while significantly improving speed over ORB-SLAM2?
• RQ5 How does the proposed method compare in accuracy and efficiency to nine existing RGB-D SLAM methods?

Key findings

• FastORB-SLAM achieves state-of-the-art accuracy on the TUM and ICL-NUIM RGB-D datasets, outperforming nine existing RGB-D SLAM methods.
• The method runs approximately twice as fast as ORB-SLAM2 due to elimination of descriptor computation in non-keyframes.
• Descriptor computation is performed only for keyframes, reducing overall computational cost while preserving tracking stability.
• The two-stage matching pipeline—using motion model and pyramid-based optical flow—provides robust initial correspondences.
• Geometric constraints from motion smoothness and epipolar geometry significantly improve the accuracy of the final keypoint matches.
• Qualitative and quantitative results confirm that the descriptor-independent approach maintains high accuracy across diverse environments.