IMAGING SYSTEMS AND IMAGING METHODS

Patent ID: 10392 | Status: Filed

Abstract

Normal backprojection is straightforward to implement but can be very slow due to the summing over a full 2-D aperture for every voxel which results in a computationally intensive O(N^5) process. However, in the case of scanning person or other object the radar image is sparse, or in other words the 3D volume only has meaningful data above a certain threshold, for example 30 dB below the maximum, in a small percentage of the voxels. If a-priori the voxels with data, or in other words the precise location of the target, were known, then the summation or focusing would only need to be done for those voxels resulting in a speed up of an order of magnitude or more. Ordinarily these voxels cannot be known without external sensors such as stereo cameras which results in an increase in complexity and a need for synchronization between the radar and the external sensor. However, the sparse nature of the radar image can still be exploited without any extra sensors by doing a multi-resolution reconstruction. The first step in the process is to define the final 3D image voxel grid with dimensions (nx, ny, nz) . Then to determine voxels of interest a 3D image of a lower resolution version of the final image voxel grid is generated based on a coarseness factor, cf, that occupies the same volume as the final grid with shape (nx /cf, ny / cf, nz /cf). The standard Range-Domain Multistatic Backprojection algorithm is used to focus the image over this new volume. The computation time for focusing over this coarse voxel grid is (cf)^3 lower than the final voxel grid. This lower resolution coarse image volume will determine the voxels in the higher resolution final 3D image that will be focused. To avoid missing scatters because of the coarseness of the image volume it is desirable to dilate the low-resolution image in some manner. One such method is binary dilation with a kernel of size (kx, ky, kz). This is a way to ensure that there is ample padding around detected scatters in the final volume to generate the complete image. The coarse image voxels are only dilated if they exceed a given threshold, for example 30 dB below the max. Finally, the final high-resolution image is focused over the selected voxels with the Range-Domain Multistatic Backprojection algorithm.

Application Number

18/133,749

Inventors

Clark,Richard
Sheen,David M

Market Sector

Sensors