3D sensing depth camera
AssigneeDuke University
InventorsJoseph Izatt, Ruobing Qian, Kevin Zhou
Life of this patent
Estimating the term…
Abstract
A 3D sensing depth camera involves a time-frequency multiplexed frequency-modulated continuous wave (FMCW) LiDAR technique. A sample arm of such a system for 3D depth sensing includes a scanner for beam scanning in a first axis and a diffractive optical element for spectrally encoded scanning along a second axis. The sample arm can further include beam shaping optics such as a collimator and a lens. Processing of an interferogram comprising depth information at a different position along the second axis for each frequency sweep of the light source involves applying a windowed spectral estimator at a particular spectral window size with zero-padding and according to a specified lateral sampling approach.
Claims18
A system for 3D depth sensing, comprising:
- a light source that produces a light across frequencies from a first frequency to a last frequency;
- a beam splitter system that is disposed to split the light from the light source into at least a reference beam and a sample beam;
- a sample arm structured to receive the sample beam from the beam splitter system and return a return sample beam, the sample arm comprising a scanner for beam scanning in a first axis and a diffractive optical element for spectrally encoded scanning along a second axis;
- a detector that is disposed to collect an interference signal between the reference beam and the return sample beam returning from the sample arm, the interference signal forming an interferogram, wherein for each frequency sweep across the frequencies from the first frequency to the last frequency, the interferogram comprises depth information at a different position along the second axis;
- and
- a processing system coupled to the detector, the processing system comprising a hardware processor and a memory storing instructions that when executed by the hardware processor direct the processing system to:
- receive the interferogram of each frequency sweep across the frequencies from the first frequency to the last frequency from the detector;
- for each interferogram, apply a windowed spectral estimator at a particular spectral window size with zero-padding and according to a specified lateral sampling approach, wherein the spectral window size is based on a lateral resolution of the system such that if a point spread function of two adjacent wavelengths are unresolvable, detected signals from the two adjacent wavelengths are treated as arising from a same position and are analyzed within a same spectral window, by:
- sampling the interferogram according to the specified lateral sampling approach;
- and
- for each spectral window sample of the interferogram, applying zero-padding to a particular number of samples and applying a spectral estimator to the zero-padded spectral window sample to generate a signal for a pixel;
- performing peak localization analysis on the signal to identify a dominant peak;
- and if the dominant peak is identified, assigning a depth value according to a corresponding depth position of the dominant peak for the pixel, otherwise, assigning the pixel as a background;
- and
- provide a depth map of the pixels comprising the assigned depth values.
A sample arm for 3D depth sensing, comprising:
- a collimator for collimating a received sample beam from a light source,
- a scanner for beam scanning of the sample beam in a first axis,
- a lens for collimating or focusing the sample beam, and
- a diffractive optical element for spectrally encoded scanning along a second axis, wherein the collimator, lens, and diffractive optical element are configured to direct light comprising each spectral window of the light source on or within a diameter of single Airy disk.
A method of generating a depth map, comprising:
- receiving an interferogram;
- sampling the interferogram according to a specified lateral sampling approach to obtain a spectral window sample;
- for each spectral window sample of the interferogram:
- applying zero-padding to a particular number of samples, and
- applying a spectral estimator to the zero-padded spectral window sample to generate a signal for a pixel for a 2D scene;
- identifying a dominant peak for at least one spectral window sample by performing peak localization analysis on the signal for the pixel of that at least one spectral window sample, and assigning a depth value to the pixel of that at least one spectral window sample according to a corresponding depth position of the dominant peak for the pixel of that at least one spectral window sample;
- assigning a background depth value to the pixel of at least one other spectral window sample, the at least one other spectral window sample having no identified dominant peak;
- providing a depth map of the pixels, the depth map comprising the assigned depth values and the assigned background depth values for each pixel in the 2D scene;
- and
- applying a gradient-based background noise removal algorithm and median filter to the depth map.
Dependencies are inferred from the claim text.