# US 2025/0157053 A1 · Object tracking using predicted positions

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

- Number: US 2025/0157053 A1
- Document ID: US20250157053A1
- Status: Published application
- Kind: A1
- Application number: 18/505,839
- Filed: 2023-11-09
- Published: 2025-05-15
- Earliest claimed: 2023-11-09
- Assignees: QUALCOMM Incorporated
- Applicants: QUALCOMM Incorporated
- Inventors: Per CRONVALL; Per Albert SIDEN
- Web page: https://patentgrep.com/patent/US20250157053A1

## CPC

- G06T 7/248: {involving reference images or patches}
- G06F 16/5854: {using shape and object relationship}
- G06T 7/246: using feature-based methods, e.g. the tracking of corners or segments
- G06T 7/74: {involving reference images or patches}
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30196: Human being; Person
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Systems and techniques are described herein for tracking objects. For instance, a method for tracking objects is provided. The method may include generating an output embedding based on an object in a first image; obtaining a predicted position of the object in a second image; modifying the output embedding based on the predicted position to generate a modified output embedding; and detecting the object in the second image based on the modified output embedding.

## Claims

### Claim 1

An apparatus for tracking objects, the apparatus comprising:
at least one memory; and
at least one processor coupled to the at least one memory and configured to:
generate an output embedding based on an object in a first image;
obtain a predicted position of the object in a second image;
modify the output embedding based on the predicted position to generate a modified output embedding; and
detect the object in the second image based on the modified output embedding.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein:
the output embedding comprises a first output embedding, and
to detect the object in the second image, the at least one processor is configured to:
provide the second image and the modified output embedding to an object-detection machine-learning model; and
receive, from the object-detection machine-learning model, a second output embedding based on the second image and the modified output embedding.

### Claim 3 (depends on 2)

The apparatus of claim 2, wherein the at least one processor is further configured to:
provide the second output embedding to a decoder; and
receive, from the decoder, image coordinates corresponding to the object in the second image based on the second output embedding.

### Claim 4 (depends on 3)

The apparatus of claim 3, wherein the image coordinates are indicative of a bounding box associated with the object.

### Claim 5 (depends on 2)

The apparatus of claim 2, wherein the modified output embedding is provided to the object-detection machine-learning model as a query.

### Claim 6 (depends on 2)

The apparatus of claim 2, wherein the object-detection machine-learning model comprises a detection transformer.

### Claim 7 (depends on 2)

The apparatus of claim 2, wherein the object-detection machine-learning model comprises:
a convolutional neural network (CNN) to generate features based on images;
a transformer encoder to generate image features based on the features; and
a transformer decoder to generate output embeddings based on the image features and queries.

### Claim 8 (depends on 1)

The apparatus of claim 1, wherein the at least one processor is further configured to decode, using a decoder, the output embedding to generate first image coordinates corresponding to the object in the first image, wherein the predicted position of the object in the second image comprises second image coordinates corresponding to the predicted position of the object in the second image; and
wherein, to modify the output embedding, the at least one processor is configured to determine a modification to the output embedding that results in the modified output embedding being decodable by the decoder to generate the second image coordinates.

### Claim 9 (depends on 8)

The apparatus of claim 8, wherein the decoder comprises a linear predictor configured to generate image coordinates based on output embeddings.

### Claim 10 (depends on 8)

The apparatus of claim 8, wherein the modification to the output embedding is determined based on a least-norm solution to an undetermined linear equation system, and wherein the undetermined linear equation system is based on the decoder.

### Claim 11 (depends on 8)

The apparatus of claim 8, wherein the modification to the output embedding is determined using a gradient-descent technique.

### Claim 12 (depends on 8)

The apparatus of claim 8, wherein the first image coordinates are indicative of a first bounding box associated with the object and wherein the second image coordinates are indicative of a second bounding box associated with the object.

### Claim 13 (depends on 1)

The apparatus of claim 1, wherein:
the output embedding comprises a first output embedding;
the predicted position comprises a first predicted position;
the modified output embedding comprises a first modified output embedding; and
to detect the object in the second image, the at least one processor is configured to generate a second output embedding based on the object in the second image;
the at least one processor is further configured to:
obtain a second predicted position of the object in a third image;
modify the second output embedding based on the second predicted position to generate a second modified output embedding; and
detect the object in a third image based on the second modified output embedding.

### Claim 14 (depends on 1)

The apparatus of claim 1, wherein the output embedding is associated with detection of the object in the first image.

### Claim 15 (depends on 1)

The apparatus of claim 1, wherein, to generate the output embedding, the at least one processor is configured to:
provide the first image to an object-detection machine-learning model; and
receive, from the object-detection machine-learning model, the output embedding based on the first image.

### Claim 16 (depends on 15)

The apparatus of claim 15, wherein the object-detection machine-learning model comprises a detection transformer.

### Claim 17 (depends on 15)

The apparatus of claim 15, wherein the object-detection machine-learning model comprises:
a convolutional neural network (CNN) to generate features based on images;
a transformer encoder to generate image features based on the features; and
a transformer decoder to generate output embeddings based on the image features and queries.

### Claim 18 (depends on 1)

The apparatus of claim 1, wherein the predicted position of the object in the second image is based on relative motion data.

### Claim 19 (depends on 18)

The apparatus of claim 18, wherein the relative motion data is based on at least one of ego-motion data that is indicative of motion of a camera associated with the first image and the second image or object-motion data that is indicative of motion of the object.

### Claim 20 (depends on 1)

The apparatus of claim 1, wherein the at least one processor is further configured to determine a position of the object in the first image, wherein the predicted position of the object in the second image is based on the position of the object in the first image.

### Claim 21 (depends on 20)

The apparatus of claim 20, wherein the at least one processor is further configured to:
determine a relative position of the object relative to a camera associated with the first image based on the position of the object in the first image; and
generate the predicted position of the object in the second image based on the relative position of the object.

### Claim 22 (depends on 21)

The apparatus of claim 21, wherein the predicted position of the object in the second image is further based on at least one of ego-motion data that is indicative of motion of a camera associated with the first image and the second image or object-motion data that is indicative of motion of the object.

### Claim 23 (depends on 1)

The apparatus of claim 1, wherein the predicted position of the object in the second image is based on at least one of:
a position of the object in the first image;
a relative position of the object relative to a camera associated with the first image and the second image;
ego-motion data indicative of motion of the camera associated with the first image and the second image; or
object-motion data indicative of motion of the object.

### Claim 24

A method for tracking objects, the method comprising:
generating an output embedding based on an object in a first image;
obtaining a predicted position of the object in a second image;
modifying the output embedding based on the predicted position to generate a modified output embedding; and
detecting the object in the second image based on the modified output embedding.

### Claim 25 (depends on 24)

The method of claim 24, wherein:
the output embedding comprises a first output embedding, and
detecting the object in the second image comprises:
providing the second image and the modified output embedding to an object-detection machine-learning model; and
receiving, from the object-detection machine-learning model, a second output embedding based on the second image and the modified output embedding.

### Claim 26 (depends on 25)

The method of claim 25, further comprising:
providing the second output embedding to a decoder; and
receiving, from the decoder, image coordinates corresponding to the object in the second image based on the second output embedding.

### Claim 27 (depends on 26)

The method of claim 26, wherein the image coordinates are indicative of a bounding box associated with the object.

### Claim 28 (depends on 25)

The method of claim 25, wherein the modified output embedding is provided to the object-detection machine-learning model as a query.

### Claim 29 (depends on 25)

The method of claim 25, wherein the object-detection machine-learning model comprises a detection transformer.

### Claim 30 (depends on 25)

The method of claim 25, wherein the object-detection machine-learning model comprises:
a convolutional neural network (CNN) to generate features based on images;
a transformer encoder to generate image features based on the features; and
a transformer decoder to generate output embeddings based on the image features and queries.
