MiCaDe: Online LiDAR-Camera Miscalibration Change Detection for Autonomous Driving
Organizations: Politecnico di Milano · Niulinx
Abstract
Reliable multi-sensor perception in autonomous vehicles requires accurate calibration between cameras and LiDAR. However, calibration drift is often unavoidable in long-term deployments: vibrations, thermal effects, and minor mechanical shocks may alter the relative pose between sensors, leading to systematic errors in downstream fusion tasks. We propose MiCaDe (Miscalibration Change Detection), an online solution detecting when the extrinsic LiDAR-camera calibration parameters are no longer accurate by means of sequential change detection. MiCaDe requires no calibration targets, or modification of the perception stack, providing low false-alarm rates with negligible computational overhead. Our key insight is that 3D and 2D bounding boxes produced by standard object detectors should remain geometrically consistent under correct calibration. By matching cross-modal detections, we derive geometric consistency metrics that serve as sensitive indicators of extrinsic drift. These metrics are aggregated into fixed-size batches and scored against a reference distribution, producing a likelihood stream that is monitored by a Change Point Model (CPM). Crucially, the CPM accumulates statistical evidence over time, enabling detection of subtle drifts that would otherwise go unnoticed. Experiments on multiple long-sequence datasets and on a real-world dataset with physically induced miscalibration confirm reliable detection under diverse conditions. Code available at: https://github.com/CarloSgaravatti/MiCaDe.
Figures & tables
| Metric | Definition | Calibration cue |
|---|---|---|
| -displacement | Horizontal projection offset | |
| -displacement | Vertical projection offset | |
| Magnitude | Projection error magnitude | |
| Direction | Direction of systematic drift | |
| IoU | Overall 2D box overlap | |
| Depth error | Coarse 3D consistency |
| Dataset | Method | Threshold | Clean sequences | Abrupt changes | Gradual changes | Speed [ms] | AUC | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ARL 0 [s] | ARL 0 [b] | FP/h | FNR | FPR | Delay [s] | FNR | FPR | Delay [s] | Angle [deg] | |||||
| KITTI | MiCaDe | 617.4 | 398.5 | 5.83 | 0.088 | 0.001 | 8.78 | 0.044 | 0.043 | 8.47 | 1.865 | 2 | 0.9976 | |
| MiCaDe | - | - | - | 0.161 | 0.000 | 8.92 | 0.044 | 0.000 | 8.77 | 1.943 | ||||
| Tahiraj et al. | .5 | 6.00 | - | 597.6 | 0.000 | 0.651 | 0.40 | 0.000 | 0.594 | 3.20 | 0.992 | 12* | 0.9969 | |
| Tahiraj et al. (adj.) | .999998 | 595.2 | - | 6.48 | 0.024 | 0.012 | 0.60 | 0.031 | 0.031 | 4.40 | 1.364 | |||
| ZOD | MiCaDe | 941.8 | 1041.4 | 3.82 | 0.000 | 0.000 | 4.66 | 0.000 | 0.000 | 21.43 | 1.148 | 2 | 0.9965 | |
| DOF | FNR | FPR | Delay [s] | AUC |
|---|---|---|---|---|
| Roll | 0.000 | 0.000 | 8.3 | 0.991 |
| Yaw | 0.000 | 0.000 | 7.3 | 0.962 |
| Pitch | 0.067 | 0.000 | 21.9 | 0.749 |
| 0.100 | 0.000 | 45.3 | 0.617 | |
| 0.267 | 0.000 | 91.2 | 0.482 | |
| 0.133 | 0.000 | 36.5 | 0.692 |
| Configuration | Metrics used | FNR | FPR | Delay [s] | AUC | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| only | ✓ | 0.063 | 0.080 | 5.32 | 0.972 | ||||||||
| only | ✓ | ✓ | 0.051 | 0.040 | 8.55 | 0.976 | |||||||
| only | ✓ | ✓ | 0.222 | 0.114 | 17.09 | 0.909 | |||||||
| Direction only | ✓ | ✓ | 0.165 | 0.000 | 7.52 | 0.949 | |||||||
| Variance only | ✓ | ✓ | 0.290 | 0.108 | 10.09 | 0.845 | |||||||
Appendix figures & tables19 assets
Supplementary material from the paper’s appendix.
Appendix
| DOF | Magnitude | FNR | FPR | Delay [s] | AUC |
|---|---|---|---|---|---|
| Roll | 0.0125 rad | 0.000 | 0.000 | 10.43 | 0.974 |
| 0.0250 rad | 0.000 | 0.000 | 6.31 | 0.999 | |
| 0.0500 rad | 0.000 | 0.000 | 8.28 | 1.000 | |
| Yaw | 0.0125 rad | 0.000 | 0.000 | 11.72 | 0.920 |
| 0.0250 rad | 0.000 | 0.000 | 5.66 | 0.972 | |
| 0.0500 rad | 0.000 | 0.000 | 4.58 | 0.994 |
| Dataset | Heterog. | Control | ||
|---|---|---|---|---|
| Pairs | Junct.-FPR | Pairs | Junct.-FPR | |
| KITTI | 24 | 0.0 | 18 | 0.0 |
| ZOD | 30 | 0.0 | 30 | 0.0 |
| ONCE | 40 | 0.0 | 32 | 0.0 |
| Dataset | FP/h | Global FPR | Emp. [s] | |||
|---|---|---|---|---|---|---|
| Het. | Ctrl. | Het. | Ctrl. | Het. | Ctrl. | |
| KITTI | 3.92 | 3.14 | .250 | .111 | 919 | 1,146 |
| ZOD | 2.79 | 3.78 | .233 | .400 | 1,292 | 952 |
| ONCE | 0.49 | 0.56 | .275 | .312 | 7,292 | 6,454 |
| Variant | Heterog. | Control | ||
|---|---|---|---|---|
| Pairs | Junct.-FPR | Pairs | Junct.-FPR | |
| ZOD country/daypart | 30 | 0.0 | 30 | 0.0 |
| ONCE weather/period | 4 | 0.0 | 2 | 0.0 |
| Metric | Before miscalibration | After miscalibration |
|---|---|---|
| Matching accuracy | 92% | 89% |
| Matches per frame | 5 | 3 |
| Degradation | FP/h | Onset ratio |
|---|---|---|
| 64 32 rings | 1.28 | 0.00 |
| 64 16 rings | 3.87 | 0.00 |
| 64 8 rings | 5.52 | 0.00 |
| sector missing | 0.00 | n/a |
| sector missing | 0.00 | n/a |
| Degradation | FNR | FPR | Delay |
|---|---|---|---|
| 64 32 rings | .207 | .000 | 12.3 s |
| 64 16 rings | .379 | .000 | 13.9 s |
| 64 8 rings | .600 | .040 | 18.2 s |
| sector missing | .207 | .000 | 12.7 s |
| sector missing | .207 | .000 | 13.4 s |