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

Qiang Fu, Hongshan Yu+4 more|arXiv (Cornell University)|Aug 22, 2020

Robotics and Sensor-Based Localization6 citations

TL;DR

FastORB-SLAM proposes a descriptor-independent, two-stage coarse-to-fine keypoint matching method using sparse optical flow to accelerate ORB-SLAM2, eliminating descriptor computation between non-keyframes. It achieves state-of-the-art accuracy and ~2x speedup on TUM and ICL-NUIM datasets by computing descriptors only for keyframes.

ABSTRACT

Indirect methods for visual SLAM are gaining popularity due to their robustness to varying environments. ORB-SLAM2 is a benchmark method in this domain, however, the computation of descriptors in ORB-SLAM2 is time-consuming and the descriptors cannot be reused unless a frame is selected as a keyframe. To overcome these problems, we present FastORB-SLAM which is light-weight 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 first predict initial keypoint correspondences via a uniform acceleration motion model and then robustly establish the correspondences via a pyramid-based sparse optical flow tracking method. In the second stage, we leverage motion smoothness and the epipolar constraint to refine the correspondences. In particular, our method computes descriptors only for keyframes. We test FastORB-SLAM with an RGBD camera on extit{TUM} and extit{ICL-NUIM} datasets and compare its accuracy and efficiency to nine existing RGBD SLAM methods. Qualitative and quantitative results show that our method achieves state-of-the-art performance in accuracy and is about twice as fast as the ORB-SLAM2

Motivation & Objective

• Address the high computational cost of descriptor computation in ORB-SLAM2 during tracking.
• Enable descriptor reuse across frames without keyframe selection, improving efficiency.
• Develop a fast, descriptor-independent keypoint matching method for real-time visual SLAM.
• Maintain high accuracy while reducing processing time in varying environments.
• Achieve state-of-the-art performance in both accuracy and speed on standard RGBD SLAM benchmarks.

Proposed method

• Use a uniform acceleration motion model to predict initial keypoint correspondences between adjacent frames.
• Apply a pyramid-based sparse optical flow method to robustly establish initial correspondences.
• Refine matches using motion smoothness and epipolar geometry constraints in a second refinement stage.
• Compute ORB descriptors only for keyframes, avoiding redundant computation in non-keyframes.
• Leverage the coarse-to-fine matching strategy to improve robustness and reduce computational load.
• Integrate the method into the ORB-SLAM2 framework with minimal architectural changes for compatibility and efficiency.

Experimental results

Research questions

• RQ1 Can descriptor computation be eliminated during tracking between adjacent frames without sacrificing accuracy?
• RQ2 How effective is a two-stage coarse-to-fine keypoint matching strategy based on optical flow in reducing SLAM computation time?
• RQ3 To what extent does computing descriptors only for keyframes maintain tracking robustness and accuracy?
• RQ4 How does the proposed method compare in speed and accuracy to existing RGBD SLAM systems on standard benchmarks?
• RQ5 Can motion smoothness and epipolar constraints significantly improve keypoint matching reliability in descriptor-free tracking?

Key findings

• FastORB-SLAM achieves approximately twice the speed of ORB-SLAM2 on the TUM and ICL-NUIM datasets.
• The method maintains state-of-the-art accuracy, outperforming nine existing RGBD SLAM methods in quantitative evaluation.
• Descriptor computation is reduced to only keyframes, significantly lowering computational overhead during tracking.
• The coarse-to-fine matching strategy based on sparse optical flow improves correspondence robustness and tracking stability.
• Qualitative results confirm consistent and accurate trajectory estimation across diverse environments.
• The method demonstrates strong performance on both TUM and ICL-NUIM datasets, validating its generalization and efficiency.