# US 12,183,048 B2 · Image stitching method, apparatus and device based on reinforcement learning and storage medium

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

- Number: US 12,183,048 B2
- Document ID: US12183048B2
- Status: Granted
- Kind: B2
- Application number: 18/767,924
- Filed: 2024-07-09
- Published: 2024-11-14
- Granted: 2024-12-31
- Earliest claimed: 2023-05-10
- Assignees: Guangdong University of Technology
- Applicants: Guangdong University of Technology
- Inventors: Jian Gao; Junlang Liang; Lanyu Zhang; Yuheng Luo; Zhuojun Zheng; Xin Chen
- Web page: https://patentgrep.com/patent/US12183048B2

## CPC

- G06V 10/16: {using multiple overlapping images; Image stitching}
- G06T 7/33: using feature-based methods
- G06T 7/70: Determining position or orientation of objects or cameras (camera calibration G06T7/80)
- G06T 7/80: Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
- G06V 10/761: {Proximity, similarity or dissimilarity measures}
- G06V 10/85: {Markov-related models; Markov random fields}
- G06T 2200/32: involving image mosaicing
- G06T 2207/10016: Video; Image sequence
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30208: Marker matrix
- Y02T 10/40: Engine management systems
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- Y02T: CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION

## Abstract

The present application provides an image stitching method, apparatus and device based on reinforcement learning and a storage medium. The method includes: acquiring initial calibration parameters, collecting a sample image and position information of a motion platform; setting a negative reward function; acquiring a state set and a negative reward value set according to a randomly generated action set, the initial calibration parameters, the position information of the motion platform and the negative reward function to construct a probability kinematics model; constructing a state value function based on an occurrence probability of the state, and acquiring an optimal action by optimizing the state value function; and acquiring optimized calibration parameters through the optimal action and the initial calibration parameters, and carrying out image stitching on corresponding sample images through the optimized calibration parameters. The application solves the technical problem of low image stitching quality in the prior art.

## Claims

### Claim 1

An image stitching method based on reinforcement learning, comprising the following steps of:
acquiring initial calibration parameters through a calibration board arranged on a motion platform, and collecting a sample image of a detected sample on the motion platform at each moment and position information of the motion platform at each moment in a movement process of the motion platform;
setting a negative reward function based on image stitching quality at each moment by taking the image stitching quality at each moment and the position information of the motion platform at each moment as a state at each moment and a calibration parameter adjustment amount at each moment as an action at each moment;
randomly generating an action set, and acquiring a state set and a negative reward value set according to the action set, the initial calibration parameters, the position information of the motion platform at each moment and the negative reward function;
constructing a Markov experience sequence according to the action set, the state set and the negative reward value set, and constructing a probability kinematics model through the Markov experience sequence, wherein the probability kinematics model is used for predicting an occurrence probability of a state at the next moment according to a state and action at a current moment;
constructing a state value function based on an occurrence probability of the state at each moment and a negative reward value at each moment, and acquiring an optimal action at each moment by optimizing the state value function; and
acquiring optimized calibration parameters at each moment through the optimal action and the initial calibration parameters, and carrying out image stitching on corresponding sample images through the optimized calibration parameters at each moment.

### Claim 2 (depends on 1)

The image stitching method based on reinforcement learning according to claim 1, wherein a calculation process of the image stitching quality comprises:
after carrying out image stitching on sample images at two adjacent moments, capturing an overlapping region of a stitched image to obtain a first overlapping image and a second overlapping image; and
calculating a similarity degree between the first overlapping image and the second overlapping image to obtain the image stitching quality.

### Claim 3 (depends on 1)

The image stitching method based on reinforcement learning according to claim 1, wherein the state value function is:
V
π
φ
(
s
0
)
=
∑
t
=
0
T
∫
c
⁡
(
s
t
)
⁢
p
⁡
(
s
t
)
⁢
ds
t
;
wherein, Vπφ(s0) is the state value function, s0 is a state at an initial moment, πφ is an action strategy, st is a state at a moment t, c(st) is a negative reward value corresponding to the state at the moment t, p(st) is an occurrence probability of the state at the moment t, and T is a final moment.

### Claim 4 (depends on 1)

The image stitching method based on reinforcement learning according to claim 1, wherein the carrying out image stitching on the corresponding sample images through the optimized calibration parameters at each moment, comprises:
calculating a platform movement distance of the motion platform at each two adjacent moments according to the position information of the motion platform at each moment;
calculating an image translation distance of sample images at each two adjacent moments according to the optimized calibration parameters at each moment, the platform movement distance at each two adjacent moments and the position information of the motion platform at each moment; and
carrying out image stitching on the sample images at each two adjacent moments based on the image translation distance of the sample images at each two adjacent moments.

### Claim 5 (depends on 1)

An image stitching device based on reinforcement learning, wherein the device comprises a processor and a storage;
the storage is used for storing a program code and transmitting the program code to the processor; and
the processor is used for executing the image stitching method based on reinforcement learning according to claim 1 based on an instruction in the program code.

### Claim 6 (depends on 1)

A non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used for storing a program code, and the program code, when executed by a processor, realizes the image stitching method based on reinforcement learning according to claim 1.