# US 12,002,253 B2 · Feature point integration positioning system, feature point integration positioning method and non-transitory computer-readable memory

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

- Number: US 12,002,253 B2
- Document ID: US12002253B2
- Status: Granted
- Kind: B2
- Application number: 17/456,646
- Filed: 2021-11-29
- Published: 2023-06-01
- Granted: 2024-06-04
- Earliest claimed: 2021-11-29
- Assignees: Automotive Research & Testing Center
- Applicants: Automotive Research & Testing Center
- Inventors: Yu-Fang Wang; Yi-Jie Lin; Cheng-Kai Wang
- Web page: https://patentgrep.com/patent/US12002253B2

## CPC

- G06V 10/443: {by matching or filtering}
- G06V 10/82: using neural networks
- G06V 10/462: {Salient features, e.g. scale invariant feature transforms [SIFT]}
- G06V 20/64: Three-dimensional [3D] objects
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A feature point integration positioning system includes a moving object, an image input source, an analyzing module and a positioning module. The image input source is disposed at the moving object to shoot an environment for obtaining a sequential image dataset. The analyzing module includes a machine vision detecting unit configured to generate a plurality of first feature points in each of the images based on each of the images, a deep learning detecting unit configured to generate a plurality of second feature points in each of the images based on each of the images, and an integrating unit configured to integrate the first feature points and the second feature points in each of the images into a plurality of integrated feature points in each of the images. The positioning module confirms a position of the moving object relative to the environment at each of the time points.

## Claims

### Claim 1

A non-transitory computer-readable memory, storing a computer program which is executed by a processor for:
obtaining a sequential image dataset from an image input source shooting an environment, wherein the sequential image dataset comprises a first to a last images, and the first to the last images respectively correspond to a first to a last time points; and
generating a plurality of first feature points in the first image by a machine vision detecting method, generating a plurality of second feature points in the first image by a deep learning detecting method, integrating the first feature points and the second feature points in the first image into a plurality of integrated feature points in the first image, and receiving the integrated feature points in the first image to confirm a position of a moving object relative to the environment at the first time point, wherein each of the first feature points and the second feature points represents a point in the first image, and at least one of the second feature points is different from any one of the first feature points, wherein the first feature points and the second feature points are the points in the first image where gray values vary in high light contrast.

### Claim 2 (depends on 1)

The non-transitory computer-readable memory of claim 1, wherein a spatial geometry model is used to obtain a three-dimensional point group dataset of the integrated feature points in the first image.

### Claim 3 (depends on 1)

The non-transitory computer-readable memory of claim 1, wherein a map is constructed by the integrated feature points in the first image.

### Claim 4 (depends on 1)

The non-transitory computer-readable memory of claim 1, wherein a plurality of environment variation images are previously used to construct a deep learning model, and the deep learning model is used to identify the second feature points.

### Claim 5 (depends on 4)

The non-transitory computer-readable memory of claim 4, wherein a plurality of items in each of the environment variation images have light variations or position variations.

### Claim 6 (depends on 1)

The non-transitory computer-readable memory of claim 1, wherein an oriented fast and rotated brief algorithm or a scale-invariant feature transform algorithm is used to obtain the first feature points in the first image.