Accurate initialization is essential for reliable visual-inertial odometry (VIO), but it is often ill-conditioned under degenerate motions. Existing methods typically require restrictive excitation motions to ensure sufficient observability or rely on computationally expensive 3D structure reconstruction, limiting efficient and practical deployment. To address these limitations, we propose SLIM-init, a structureless monocular VIO initializer that directly exploits geometric constraints from tracked 2D line features without explicit 3D landmark reconstruction. Specifically, SLIM-init leverages line-derived vanishing points (VPs) as translation-invariant orientation cues to provide robust rotation-only constraints under degenerate scenarios such as low-parallax or translation-dominant motions. It further incorporates a line epipolar residual to constrain translation and a line-normal projection residual to improve the conditioning of linear alignment, enhancing the accuracy and robustness of initial state estimation. Extensive experiments on a public benchmark and challenging custom degenerate-motion sequences demonstrate improved accuracy and robustness over state-of-the-art initialization methods. The source code is available at: https://github.com/cjunwan/SLIM-init.
Figures & tables
Figure 1: Overall pipeline of SLIM-init. Colored blocks highlight our augmentations: a VP-based consistency residual for robust gyroscope bias estimation (Step 1 ), followed by line epipolar and line-normal projection residuals for translation-based linear alignment (Step 2 ). The entire pipeline is strictly structureless, avoiding depth estimation or any explicit 3D reconstruction.
Figure 2: Line epipolar geometry between two camera views observing a common 3D line Lℓ . The two interpretation planes intersect at the underlying 3D line, and their structural coplanarity with the camera baseline constrains the plane normals as in ( 9 ).
Figure 3: Custom degenerate dataset acquisition setup. (a) Representative indoor environments in the KAIST E3-2 building, including corridors and a lobby. The collected sequences feature textureless walls, repetitive patterns, low parallax, and translation-dominant motions. (b) Sensor module used for data collection.
Metric
Method
MH_01
MH_02
MH_03
MH_04
MH_05
V1_01
V1_02
V1_03
V2_01
V2_02
V2_03
Avg.
Scale RMSE ↓
VINS-Mono [ 28 ]
0.151
0.153
0.187
0.182
0.225
0.222
0.182
0.260
0.170
0.161
0.255
0.195
OpenVINS [ 11 ]
0.137
0.121
0.111
0.151
0.161
0.125
0.101
0.151
0.149
0.117
0.118
0.131
DRT-l [ 14 ]
0.120
0.124
0.119
0.157
0.178
0.127
0.091
0.187
0.137
0.112
0.142
0.136
Ours
0.092
0.100
0.116
0.144
0.161
0.100
0.085
0.189
0.104
0.083
0.149
0.120
Gravity dir. RMSE (°) ↓
VINS-Mono [ 28 ]
1.409
1.130
2.261
1.940
1.629
3.417
1.007
1.378
1.363
1.615
1.505
1.696
OpenVINS [ 11 ]
1.931
1.558
2.239
1.758
2.326
3.332
3.255
3.946
1.835
3.251
4.760
2.745
Table 1: Quantitative comparison of exhaustive initialization performance on the EuRoC dataset. Bold and underlined values denote the best and second-best results, respectively.
Figure 4: Qualitative comparison of initialization performance on the V1_01 sequence of the EuRoC dataset. Trajectory segments are colored by the pose RMSE (m) computed for each initialization window, where lighter colors indicate lower errors. Black segments denote failed initializations. Overall, SLIM-init achieves the fewest failures and the lowest pose errors among the evaluated methods.
Sequence
Success rate (%) ↑
VINS-Mono [ 28 ]
OpenVINS [ 11 ]
DRT-l [ 14 ]
Ours
Machine Hall
63.84
60.22
91.77
94.01
Vicon Room 1
22.97
62.90
72.08
73.14
Vicon Room 2
28.78
59.35
71.51
72.11
Table 2: Initialization success rates on the EuRoC dataset. Best results are marked in bold .
Metric
Method
C_01
C_02
C_03
Avg.
Scale RMSE ↓
DRT-l [ 14 ]
0.223
0.132
0.144
0.166
Ours
0.208
0.101
0.144
0.151
Gravity dir. RMSE (°) ↓
DRT-l [ 14 ]
5.269
2.550
4.817
4.212
Ours
5.154
2.507
4.751
4.137
Velocity RMSE (m/s) ↓
DRT-l [ 14 ]
0.346
0.176
0.239
0.254
Ours
0.335
0.152
0.233
0.240
Table 3: Quantitative comparison of exhaustive initialization performance on custom degenerate sequences. Best results are marked in bold .
Figure 5: Comparison of initialization solve time on the EuRoC dataset. The central line denotes the median, and diamond markers denote the mean solve time.
Method
Sca. ↓
Grav. ↓
Vel. ↓
Pos. ↓
Ours w/o both line residuals
0.166
4.212
0.254
0.216
Ours w/o line epipolar residual
0.162
4.142
0.245
0.215
Ours w/o line normal residual
0.158
4.196
0.245
0.212
Ours
0.151
4.137
0.240
0.206
Table 4: Ablation study on custom degenerate sequences, reported as averages across all degenerate sequences. (Sca.: scale RMSE, Grav.: gravity direction RMSE ( ∘ ), Vel.: velocity RMSE (m/s), Pos.: pose RMSE (m)). Best results are marked in bold .
Fast and reliable initialization is critical for monocular visual-inertial navigation systems (VINS), as it establishes the starting conditions for subsequent state estimation. Despite steady progress, most existing methods heavily rely on visual feature correspondences and require 3-4 seconds of sensory data for successful initialization, which limits their applicability and efficiency. With the advent of feed-forward 3D models that can directly predict point clouds from images, we revisit the visual-inertial initialization problem from a concise perspective. In this work, we propose a feature-free initialization framework that leverages up-to-scale point clouds predicted by a feed-forward 3D model, thereby obviating the need for visual feature tracking and estimation. This design substantially reduces system complexity and improves the reliability of initialization. Experiments on public datasets demonstrate that the proposed feature-free initialization method achieves the highest success rate, exceeding 90%, and significantly reduces the data duration required for successful initialization, typically to under 1.2 s. We further validate our method on a self-collected dataset covering various indoor and outdoor scenarios, demonstrating robust performance, particularly in visually degraded environments where existing methods often fail. The code and dataset are available at https://github.com/Yuantai-Z/FF-VIO-Init.
SLAM (Simultaneous Localization and Mapping) and Odometry are important systems for estimating the position of mobile devices, such as robots and cars, utilizing one or more sensors. Particularly in camera-based SLAM or Odometry, effectively tracking visual features is important as it significantly impacts system performance. In this paper, we propose a method that leverages deep learning to robustly track visual features in monocular camera images. This method operates reliably even in textureless environments and situations with rapid lighting changes. Additionally, we evaluate the performance of our proposed method by integrating it into VINS-Fusion (Monocular-Inertial), a commonly used Visual-Inertial Odometry (VIO) system.
Sanghyun Park, Soohee Han
Department of Convergence IT Engineering, POSTECH, Pohang, 37673, Korea · Department of Electrical Engineering and of Convergence IT Engineering, POSTECH, Pohang, 37673, Korea
Traditional monocular Visual-Inertial Odometry (VIO) systems struggle in low-texture environments where sparse visual features are insufficient for accurate pose estimation. To address this, dense Monocular Depth Estimation (MDE) has been widely explored as a complementary information source. While recent Vision Transformer (ViT) based complex foundational models offer dense, geometrically consistent depth, their computational demands typically preclude them from real-time edge deployment. Our work bridges this gap by integrating learned depth priors directly into the VINS-Mono optimization backend. We propose a novel framework that enforces affine-invariant depth consistency and pairwise ordinal constraints, explicitly filtering unstable artifacts via variance-based gating. This approach strictly adheres to the computational limits of edge devices while robustly recovering metric scale. Extensive experiments on the TartanGround and M3ED datasets demonstrate that our method prevents divergence in challenging scenarios and delivers significant accuracy gains, reducing Absolute Trajectory Error (ATE) by up to 28.3%. Code will be made available.
Arda Alniak, Sinan Kalkan, Mustafa Mert Ankarali +2
Dept. of Electrical & Electronics Engineering · Center for Image Analysis (OGAM), METU, Ankara, Türkiye · Dept. of Computer Engineering