Capturing a Shiny Part in 3 Seconds Without Stitching
Large automotive parts carry shiny surfaces, curved form and deep pockets. Manual inspection and 2D vision cannot evaluate 3D quality completely, and an offline CMM cannot keep up with the line. This project uses a blue-light full-field snapshot 3D camera based on structured light, capturing the whole field in a single exposure with a large field of view at a 3-second cycle, subject to site acceptance results.
Why high reflectivity makes structured light blind
A mirrored metal surface reflects the projected fringe straight away, so the camera either overexposes or receives no pattern at all, and the cloud has holes in those regions. A hole is not noise; it is missing data. Where data is missing, neither dimension nor form can be judged, and coverage is not coverage.
Blue light is not an aesthetic choice
The blue band reflects more usefully off metal, and ambient light carries little blue, so the system resists ambient interference. Choosing blue scores on both counts at once, getting the pattern into the surface and keeping outside light out. That is an imaging decision, not a picture-quality preference.
Full-field snapshot against line scanning
A full-field snapshot captures the whole view's 3D data in one exposure: no motion error and no stitching. Line scanning needs repeated acquisitions and stitching, which takes time and suffers from vibration during travel. For inline inspection, one-shot capture is the precondition of the cycle, and it also removes stitching error from the result. The camera sits above the station with no mechanical scanning mechanism, which simplifies maintenance as well.
Deep pockets are a physical limit, not an algorithm problem
A deep pocket creates projection occlusion: the fringe cannot get in, so the camera cannot see. The scheme uses a wide projection and multi-angle fringe sequences to cover as much of the pocket as possible, but a very deep and narrow cavity can only be assessed by projection angle when coverage is quoted. Describing that physical limit as something the algorithm will solve is the most common misunderstanding at selection stage.
What happens in 3 seconds, and why attitude wobble is tolerable
Fringes are projected onto the part and deform with the surface; the camera captures the deformed patterns and phase unwrapping recovers the 3D form, producing a point cloud; the cloud is then analysed for features and state, and anomalies are marked and archived. The 3 seconds covers projection through verdict.
Parts wobble slightly on a conveyor. The large field of view of a full-field snapshot covers the wobble range, which lowers the demand for precise location. That is an additional benefit of the wide field, beyond simply capturing everything.
Limits: coatings and oil film must be validated on real parts
Coatings and oil film change reflective behaviour, and the same parameters can behave differently across incoming states. Imaging should be validated on the actual incoming condition. Where a part exceeds one camera's field, several units are laid out in zones with unified projection and exposure settings so cloud quality stays consistent across views.
Questions before changing methods
Why can ordinary structured light not scan a shiny surface?
Specular reflection overexposes or erases the fringe pattern and leaves holes in the cloud. A blue band with exposure control suppresses the reflection enough to measure.
Why blue rather than white?
Blue reflects more usefully from metal and is scarce in ambient light, so it handles high reflectivity better with less interference.
Will deep pockets be missed?
There is occlusion risk. Wide projection and multi-angle fringe sequences extend coverage, but very deep narrow cavities need assessment by projection angle and full coverage cannot be guaranteed.