Rotating Scans for Systematic Error Removal
Fatemeh Abbasinejad, Yong Joo Kil, Andrei Sharf, Nina Amenta

Eurographics Symposium on Geometry Processing 2009 (SGP09)
Awarded 2nd Best Paper
 

Abstract:

Optical triangulation laser scanners produce errors at surface discontinuities and sharp features. These systematic errors are anisotropic. We examine the causes of these errors theoretically, and we study the correlation of systematic error with edge size and orientation experimentally. We then present a novel processing method for removing systematic errors, by combining scans taken at several different orientations. We apply an anisotropic filter to the separate scans, and use it to weight the data in a final combination step. Unlike previous approaches, our method does not require access to the scanner's internal data or firmware. We demonstrate the technique on data from laser range scanners by two different manufacturers.

 
Coming Soon!
PDF SGP 09 Slides Download Data
 
Results:
 
Systematic error removal from scans of a cast of a Mayan hieroglyphic. The output depth map is in the center. On either side, close-ups of selected regions, with cross sections highlighted and shown beneath. (a) shows the super-resolved depth map from scans taken at zero degrees, with systematic errors near the deep horizontal grooves. (b) is the super-resolved depth map taken at ninety degrees, with errors near vertical edges. The depth map created by equally averaging all four input orientations with equal weight shows in (c). Finally (d) shows our result. While the averaging method in (c) does reduce the systematic errors, it clearly fails to eliminate them, indicating that our weighting scheme is necessary. [Mayan Hieroglyphic model data]
 
An input scan of a gear, taken with the NextEngine scanner (left). Averaging ten scans (mid-left) removes general noise, but very large systematic errors become evident. Another de-noised depth map (mid-right), scanned in a different orientation shows equally large systematic errors, but in different regions. Finally, combining the depth maps captured at the four different orientations (right) produces a high-quality output depth map, dramatically better than any of the input scans. [Gear model data]
 
Removing systematic error from a scanned relief of a wall in Persepolis. Left, we see an extra ridge to the right of the spear. Errors at depth discontinuity edges perpendicular to the triangulation baseline are typical of optical triangulation laser scanners. Right, capturing four depth maps at different orientations and combining them using our novel anisotropic filter removes these systematic errors. [Persepolis Relief model data]
 
Fine feather features of the parrot model: The super-resolution scan (a) taken at zero degrees. There are systematic errors in the vertical direction but the mostly vertical fine details in the feathers are captured correctly. In the scan taken at ninety degrees (b), the systematic errors, in particular the overshoot, serves to emphasize the texture. The combined final output (c) is similar to a very high-resolution scan (d), taken by focusing the laser scanner at the region shown. [Parrot model data]
 

Left, a close-up of a depth map captured from our customized bulls-eye calibration object, produced with a high-precision laser range scanner (Minolta Vivid 910). The corners should all be 90 degrees. A side view of a strip cut along the scanner’s sensor-emitter axis shows the systematic error: a stair-step error on each rising edge, and an overshoot before each falling edge. Right, combining four such depth maps successfully removes the systematic error.

 
[Back to Research page]

This page was last updated: Tuesday, November 24, 2009