Depth from spectrally-varying blurring detected by a snapshot narrow band multispectral imaging sensor

摘要

There exists few reports in the literature aimed at estimating depth from the spectrally-varying-amount-of-defocus contained in a single image generally captured by a commercial colour camera. However, the colour camera can adversely affect the depth measurement accuracy, due to the cross-talk between its wide and overlapping red/green/blue channels. In this study, the authors propose an improved method to estimate depth from non-overlapping and narrow-band spectral images captured by a snapshot narrow-band multispectral imaging (SNB-MSI) sensor. An optical lens with visible longitudinal chromatic aberration is placed in front of the SNB-MSI sensor to enlarge the spectrally-varying-focal-depth so that spectral images corresponding to different narrow bands are focused on apparently different planes. This setting not only differentiates depth ambiguity on both sides of the focal plane, that is caused by two different depths lead to the same blurred image, but also prevents cross talks between different spectral channels. Preliminary experimental data show that the central wavelength of best focused spectral band shifts with increasing depth of the target. This suggests that it is feasible to dynamically measure the depth of the scene using the combination of a longitudinal chromatic aberration optical lens with a SNB-MSI sensor.