Depth Maps from Simulated Stereo Images using Kernel Matching

PDF

Authors
  1. Barton, S.A.
  2. Mackay, D.J.
Corporate Authors
Defence R&D Canada - Suffield, Ralston ALTA (CAN)
Abstract
Simulated stereo images and depth maps were generated using the open-source ray tracing program POV-Ray, operating in an environment containing obstacles. The image pairs were used to generate depth maps by finding the relative positions of matching groups (kernels) of image pixels. Kernel matching generated an average depth accuracy of better than 90% over all images and pixels tested. When neighbourhood average and median filter smoothing techniques were applied to the resulting depth maps, the average accuracy exceeded 95%. Occlusion and lighting differences between the images were attenuated by smoothing and kernel normalization. Computation times for the simulated depth maps indicate that kernel matching could be used in real video imagery on a vehicle moving at 30 km/hr.

Il y a un résumé en français ici.

Report Number
DRDC-SUFFIELD-TM-2006-198 — Technical Memorandum
Date of publication
01 Nov 2006
Number of Pages
28
DSTKIM No
CA028840
CANDIS No
527017
Format(s):
CD ROM

Permanent link

Document 1 of 1

Date modified: