# US 12,602,807 B2 · Method for subpixel disparity calculation > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,602,807 B2 - Document ID: US12602807B2 - Status: Granted - Kind: B2 - Application number: 18/012,861 - Filed: 2021-06-24 - Published: 2023-09-21 - Granted: 2026-04-14 - Earliest claimed: 2020-06-24 - Assignees: Gideon Brothers d.o.o. - Applicants: Gideon Brothers d.o.o. - Inventors: Nikola Banić; Kruno Lenac; Dario Ljubić; Luka Pevec - Web page: https://patentgrep.com/patent/US12602807B2 ## CPC - G06T 7/593: from stereo images - G06T 7/97: {Determining parameters from multiple pictures (depth or shape recovery from multiple images G06T7/55; stereo camera calibration G06T7/85)} - G06T 2200/28: involving image processing hardware - G06T 2207/10012: Stereo images - G06T 2207/20081: Training; Learning - G06T 2207/20084: Artificial neural networks [ANN] - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL ## Abstract In a method for subpixel disparity calculation, image data for various images each representing a field of view of an input device is received by a processor, and the image data is applied to a machine learning model. The machine learning module uses the image data to compute an output representing calculated subpixel disparity between the various images. In an example of the method, the machine learning model is a neural network that produces accurate and reliable subpixel disparity estimation in real-time using synthetically generated data. ## Claims ### Claim 1 A computer implemented method for processing image data, the method comprising: providing one or more processors; configuring the one or more processors to receive first image data for a first image of a field of view and second image data for a second image of the field of view; configuring the one or more processors to apply the first image data and the second image data to a machine learning model, wherein the machine learning model is a neural network; configuring the one or more processors to compute, by the machine learning model, an output number that represents a calculated subpixel disparity between the first image and the second image; and configuring the one or more processors to train the neural network with synthetic data, wherein the synthetic data comprises one or more synthetically generated pairs of the first image and the second image, each pair of said one or more synthetically generated pairs comprising a predetermined subpixel disparity between the first image and the second image. ### Claim 2 (depends on 1) The method of claim 1, further comprising a first input device configured to obtain the first image data, and a second input device configured to obtain the second image data. ### Claim 3 (depends on 1) The method of claim 1, wherein the first input device and the second input device each include one or more digital cameras, computer readable medium, or both. ### Claim 4 (depends on 1) The method of claim 1, further comprising configuring the one or more processors to train the neural network in a supervised training mode, wherein the supervised training mode comprises: receiving a first training raw image data representative of a first training raw image; simulating a subpixel movement of the first training raw image by an input number to obtain second training raw image data representative of a second training raw image; converting the first training raw image data representative of the first training raw image into first training image data representative of a first training image in a prespecified image format by applying a set of non-linear operations, wherein the prespecified image format corresponds to an image format obtained by a digital camera; converting the second training raw image data representative of the second training raw image into second training image data representative of a second training image in the prespecified image format by applying the set of non-linear operations; and obtaining the first training image data and the second training image data. ### Claim 5 (depends on 4) The method of claim 4, further comprising: computing, by the neural network, a training output number based on the first training image data and the second training image data; computing, by the neural network, a loss, at least in part on comparing the input number to the training output number; and updating of one or more parameters of the neural network based at least in part on the loss. ### Claim 6 (depends on 5) The method of claim 5, further comprising: using backpropagation for updating one or more parameters of the neural network. ### Claim 7 (depends on 1) The method of claim 1, further comprising: providing a computer program directly loadable into an internal memory of the one or more processors, the computer program comprising software capable of controlling the steps of claim 1 when the program is run on the one or more processors. ### Claim 8 (depends on 1) The method of claim 1, further comprising: providing a computer readable medium having a computer program recorded thereon, wherein the computer program is configured to permit the one or more processors to control the steps of claim 1. ### Claim 9 (depends on 1) The method of claim 1, wherein the output number is a single scalar value. ### Claim 10 (depends on 1) The method of claim 1, wherein the synthetic data further comprises a training set of pairs of training raw image data. ### Claim 11 (depends on 10) The method of claim 10, wherein the training set of pairs of training raw image data represents pairs of training raw images having an input number, wherein the input number represents a ground truth subpixel disparity between each pair of training raw images. ### Claim 12 A system for processing image data, the system comprising: one or more processors; a non-transitory, computer readable medium, configured with instructions that, when executed by the one or more processors, cause the system to implement a neural network comprising an input layer configured to receive first image data for a first image of a field of view and second image data for a second image of the field of view; wherein the neural network is configured to provide an output number that represents a calculated subpixel disparity between the first image and the second image; and wherein the one or more processors are configured to train the neural network with synthetically generated data comprising one or more synthetically generated pairs of the first image and the second image, each pair of said one or more synthetically generated pairs comprising a predetermined subpixel disparity between the first image and the second image. ### Claim 13 (depends on 12) The system of claim 12, wherein the output number is a single scalar value. ### Claim 14 (depends on 12) The system of claim 12, further comprising a dense layer configured to process the first image data and the second image data, and a parametric rectified linear layer capable of performing non-linear processing of the first image data and the second image data. ### Claim 15 (depends on 14) The system of claim 14, further comprising a dropout layer, and a batch normalization layer. ### Claim 16 (depends on 12) The system of claim 12, wherein the one or more synthetically generated pairs comprise one or more pairs of training raw images with an input number representing a ground truth, said training raw images formed by simulating a predetermined sub-pixel shift between the images.