Depth Completion from Sparse 3D Maps for Automated Defect Quantification
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
For lifeline systems, routine visual inspections play a vital role in gathering structure-level data to inform asset maintenance strategies. Current research has shown that data repeatability and objectivity may be enhanced with computer vision and robotics, whereby manual collection, cataloguing, and quantification of defects is automated. For instance, low-cost lidar sensors integrated with a robotic platform can collect 3D geometric information of a structure via simultaneous localization and mapping (SLAM), with image data fusion allowing for defect measurement. While low-cost lidars provide a fast and inexpensive way of obtaining defect measurements, maps they produce often lack the density required for accurate quantification. To remedy this, we combine depth completion enhanced point cloud data from a lidar with labelled image data to extract a more complete surface estimate in physical scale. Given a 3D map, image data with known camera poses, and camera intrinsic calibrations, our approach first transforms the 3D map to a depth map (within camera frame) through ray casting. Depth maps are then densified using a depth completion algorithm that employs image processing techniques tailored to suit sparse lidar map data. Lastly, the combination of labeled defect images and dense depth maps is exploited to extract high-resolution area measurements for defects such as spalls and delaminations. The accuracy of our method has been assessed by comparing results to those obtained from non-depth completed data and to ground truth measurements obtained using a high-resolution monocular camera with manual scale input to each image. Using a data set containing six concrete area defects, our method was shown to reduce error, on average, by 14.2%, with depth completed results having an average area error of 5.3% from the ground truth. Our results show that depth completion on sparse 3D maps can be effective in fine-scale defect quantification, providing a more complete assessment of lifeline system condition using affordable sensors.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it