Does Dynamic-Point Filtering Help When Texture Is Scarce? A Controlled Study of ORB-SLAM2 Front-Ends in Synthetic Indoor Scenes

Summary

Abstract Summary

Dynamic-point filters are routinely added to feature-based visual SLAM, and several recent systems argue that removing dynamic features can leave too few static features in low-texture regions. So far, these systems have been evaluated only on texture-rich benchmark sequences. We present a controlled study that isolates this interaction. We render synthetic indoor sequences in which surface texture (four levels, quantified by FAST-corner density and image-gradient entropy) and scene dynamics (three levels) are varied factorially along identical camera trajectories, with stereo, RGB-D, ground-truth poses and dynamic masks. On this grid we compare ORB-SLAM2 without filtering, with an optical-flow and epipolar-residual filter (FLOW), and with a multi-view depth-consistency filter (GEOM), and report trajectory error, tracking completeness and surviving static features over five runs. Because the masks give per-keypoint ground truth, we also measure each filter's dynamic-point precision and recall and its static-feature false-removal rate, so that mechanistic explanations can be tested directly. We do not propose a new filter. On 720 runs over 24 sequences, filtering helped mainly in the most dynamic cells; the benefit did not decline monotonically with texture, but at the lowest level filtering reduced tracking completeness, and ORB-SLAM2 never initialised in static L3 scenes. Contrary to our hypothesis, GEOM discarded more static keypoints than FLOW (median FRR 6.8% vs. 1.5% for RGB-D, 19.6% vs. 1.5% for stereo); its RGB-D advantage tracked dynamic-point recall and vanished in stereo mode. Data and code are available at https://github.com/felixxxue/texture-dynamics-slam.

Key Points

Dynamic-point filters are routinely added to feature-based visual SLAM, and several recent systems argue that removing dynamic features can leave too few static features in low-texture regions.
So far, these systems have been evaluated only on texture-rich benchmark sequences
We present a controlled study that isolates this interaction
We render synthetic indoor sequences in which surface texture (four levels, quantified by FAST-corner density and image-gradient entropy) and scene dynamics (three levels) are varied factorially along identical camera trajectories, with stereo, RGB-D, ground-truth poses and dynamic masks.
On this grid we compare ORB-SLAM2 without filtering, with an optical-flow and epipolar-residual filter (FLOW), and with a multi-view depth-consistency filter (GEOM), and report trajectory error, tracking completeness and surviving static features over five runs.

Study Design

We use the public ORB-SLAM2 implementation [ 2 ] in its stereo and RGB-D modes with the default tracking, local-mapping and loop-closing threads and with $N_{f}=1000$ ORB features [ 33 ] per frame (scale factor 1.2, 8 levels, FAST thresholds 20/7, close/far depth threshold 40 baselines); all other parameters are left at their defaults.

Experiments

Hypotheses. The design is built to test three pre-stated hypotheses; each may be refuted, and refutations will be reported as such. The accuracy benefit of dynamic-point filtering ($-\Delta\mathrm{ATE}_{f}$) decreases as texture decreases, and may become negative at the lowest texture level, because filtering removes a larger share of the scarce static keypoints.

Geom has a lower static-feature false-removal rate than Flow , and the gap widens as texture decreases, because $\mathbf{F}_{t}$ in Eq.

Experiments — Does Dynamic-Point Filtering Help When Texture Is Scarce? A Controlled Study of ORB-SLAM2 Front-Ends in Synthetic Indoor Scenes — Fig. 3: Effect of filtering, $\Delta\mathrm{ATE}_{f}=\mathrm{med}\,\mathrm{ATE}(f)-\mathrm{med}\,\mathrm{ATE}(\textsc{None})$ in cm (medians over 2 scenes $\times$ 5 runs, failed runs excluded; negative = filter helps; symmetric-log colour scale centred at 0), per texture level (rows, L0 richest) and dynamic level (columns). ∗ : Holm-corrected Wilcoxon rank-sum $p None .

Sources and Demonstrations

Figures are reproduced from Xue See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.