# US 2023/0053786 A1 · Enhanced object detection

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2023/0053786 A1
- Document ID: US20230053786A1
- Status: Published application
- Kind: A1
- Application number: 17/406,381
- Filed: 2021-08-19
- Published: 2023-02-23
- Earliest claimed: 2021-08-19
- Assignees: Ford Global Technologies, LLC
- Applicants: Ford Global Technologies, LLC
- Inventors: John Moore; Sahib Singh; Uttara Thakre
- Web page: https://patentgrep.com/patent/US20230053786A1

## CPC

- G06V 20/58: Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V 20/582: {of traffic signs}
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/22: Matching criteria, e.g. proximity measures
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06V 10/26: Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A plurality of subimages of an image are generated based on output from a random number generator. Each subimage is input to a machine learning program trained to output an object classified in the subimage. When a number of subimages with a same classified object exceeds a threshold, an object in the image is identified as the classified object.

## Claims

### Claim 1

A system, comprising a computer including a processor and a memory, the memory storing instructions executable by the processor to:
generate, based on output from a random number generator, a plurality of subimages, wherein each subimage is a set of at least two divisions of an image, each division of the image being a portion of the image different than each other division of the image;
input each subimage to a machine learning program trained to output an object classified in the subimage; and
when a number of subimages with a same classified object exceeds a threshold, identify an object in the image as the classified object.

### Claim 2 (depends on 1)

The system of claim 1, wherein the instructions further include instructions to divide the image into a plurality of divisions and to generate the output from the random number generator based on a total number of the plurality of divisions.

### Claim 3 (depends on 1)

The system of claim 1, wherein the instructions further include instructions to input an annotation identifying the object in the image to the machine learning program and to determine whether the annotation matches the classified object output from the machine learning program.

### Claim 4 (depends on 3)

The system of claim 3, wherein the instructions further include instructions to input the matching determination into a training dataset of the machine learning program.

### Claim 5 (depends on 1)

The system of claim 1, wherein the instructions further include instructions to, when no number of subimages with the same classified object exceeds the threshold, input the image to a second machine learning program trained to output the object identified in the image.

### Claim 6 (depends on 1)

The system of claim 1, wherein the instructions further include instructions to assign an identifying number to each of the divisions, the identifying numbers defining a set, and the output of the random number generator is a subset of the set of the identifying numbers.

### Claim 7 (depends on 1)

The system of claim 1, wherein the machine learning program is a multi-task neural network, wherein each task of the multi-task network is programmed to classify a specific type of object different from the other tasks.

### Claim 8 (depends on 1)

The system of claim 1, wherein the instructions further include instructions to identify a plurality of classification regions in the image, each classification region being a region in the image that the machine learning program outputs an identification of an object, and to determine a number of divisions of the image based on the identified plurality of classification regions.

### Claim 9 (depends on 8)

The system of claim 8, wherein the instructions further include instructions to generate at least one division that includes no classification regions.

### Claim 10 (depends on 1)

The system of claim 1, wherein the instructions further include instructions to, when the object is identified in the image, actuate a vehicle subsystem based on the identified object.

### Claim 11 (depends on 1)

The system of claim 1, wherein the instructions further include instructions to divide the image into a plurality of equally-sized divisions in a grid pattern.

### Claim 12 (depends on 11)

The system of claim 11, wherein the instructions further include instructions to generate the plurality of subimages based on the divisions in the grid pattern, to divide the image into a second plurality of equally-sized divisions in a second grid pattern, and to generate a second plurality of subimages based on the divisions in the second grid pattern.

### Claim 13

A method, comprising:
generating, based on output from a random number generator, a plurality of subimages, wherein each subimage is a set of at least two divisions of an image, each division of the image being a portion of the image different than each other division of the image;
inputting each subimage to a machine learning program trained to output an object classified in the subimage; and
when a number of subimages with a same classified object exceeds a threshold, identifying an object in the image as the classified object.

### Claim 14 (depends on 13)

The method of claim 13, further comprising dividing the image into a plurality of divisions and generating the output from the random number generator based on a total number of the plurality of divisions.

### Claim 15 (depends on 13)

The method of claim 13, further comprising inputting an annotation identifying the object in the image to the machine learning program and determining whether the annotation matches the classified object output from the machine learning program.

### Claim 16 (depends on 13)

The method of claim 13, further comprising, when no number of subimages with the same classified object exceeds the threshold, inputting the image to a second machine learning program trained to output the object identified in the image.

### Claim 17 (depends on 13)

The method of claim 13, further comprising assigning an identifying number to each of the divisions, the identifying numbers defining a set, and the output of the random number generator is a subset of the set of the identifying numbers.

### Claim 18 (depends on 13)

The method of claim 13, wherein the machine learning program is a multi-task neural network, wherein each task of the multi-task network is programmed to classify a specific type of object different from the other tasks.

### Claim 19 (depends on 13)

The method of claim 13, further comprising, when the object is identified in the image, actuating a vehicle subsystem based on the identified object.

### Claim 20 (depends on 13)

The method of claim 13, further comprising dividing the image into a plurality of equally-sized divisions in a grid pattern.
