# US 2026/0073531 A1 · Particle tracking method for broken coral soil based on multimodal model

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

- Number: US 2026/0073531 A1
- Document ID: US20260073531A1
- Status: Published application
- Kind: A1
- Application number: 19/393,546
- Filed: 2025-11-18
- Published: 2026-03-12
- Earliest claimed: 2025-09-25
- Assignees: Zhejiang University of Technology
- Applicants: Zhejiang University of Technology
- Inventors: Zili HE; Honglei SUN; Mengfen SHEN; Xiaodong PAN; Kun PAN; Junda ZHOU; Yuanqiang CAI
- Web page: https://patentgrep.com/patent/US20260073531A1

## CPC

- G06T 7/20: Analysis of motion (motion estimation for coding, decoding, compressing or decompressing digital video signals H04N19/43, H04N19/51)
- G06T 3/4007: based on interpolation, e.g. bilinear interpolation (image demosaicing G06T3/4015; edge-driven or edge-based scaling G06T3/403)
- G06T 7/10: Segmentation; Edge detection (motion-based segmentation G06T7/215)
- G06T 7/246: using feature-based methods, e.g. the tracking of corners or segments
- G06T 7/277: involving stochastic approaches, e.g. using Kalman filters
- G06T 7/80: Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
- G06T 15/00: Three-dimensional [3D] image rendering
- G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods}
- G06V 10/82: using neural networks
- G06T 2207/10028: Range image; Depth image; 3D point clouds
- G06T 2207/20016: Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
- G06T 2207/30241: Trajectory
- G06T 2207/30242: Counting objects in image
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

The present disclosure belongs to the field of particle tracking technology of coral particles in soil, and specifically discloses a particle tracking method for coral particles in soil based on a multimodal model. The method comprises a particle tracking method that enables counting the coral soil particles in an image by using an improved small sample counting model. Coral soil particles in the figures are segmented, and contour coordinates and morphological characteristics of coral soil particles are obtained. Multiple images are put into an optimized BoT-SORT algorithm, and a three-dimensional motion trajectory of the surface coral soil particles is obtained during the test. This method allows for comprehensive observation throughout a triaxial test, and it not only increases the speed and accuracy of coral soil particle segmentation and reduces the tracking error rate, but also improves overall precision.

## Claims

### Claim 1

A particle tracking method for coral particles in soil based on a multimodal model, comprising the following steps:
S1, conducting a triaxial test using a transparent rubber triaxial membrane;
S2, shooting layers of coral soil particles in real time by two cameras and saving the images taken;
S3, obtaining a three-dimensional cloud image of surface particles by three-dimensional reconstruction of the images taken in S2 through a binocular vision algorithm;
S4, expanding a side of the triaxial specimen taken in S2 into a plane by a bilinear interpolation method;
S5, improving a small sample counting model;
S6, counting the coral soil particles in the image by using an improved small sample counting model, and outputting the position of each target;
S7, inputting the image in S6 into a segmentation model, segmenting the coral soil particles, and obtaining a segmentation result of the coral soil particles;
S8, according to the segmentation results of coral soil particles in S7, obtaining contour coordinates of the coral soil particles, and calculating an aspect ratio, convexity, sphericity, roundness, and Feret diameter of coral soil particles according to the contour coordinates;
S9, inputting multiple images processed in S4 into an optimized BoT-SORT algorithm, and then performing a multi-target tracking, and adding a crushing matching mechanism in a multi-target tracking process;
S10, inputting the tracking results in S9 into the three-dimensional cloud map of S3, and obtaining a three-dimensional motion trajectory of the surface coral soil particles during the test.

### Claim 2 (depends on 1)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 1, wherein in S3, the binocular vision algorithm is divided into four steps: camera calibration, image correction, feature matching, disparity smoothing, and post-processing.

### Claim 3 (depends on 2)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 2, wherein the camera calibration obtains internal parameters of each camera and external parameters between the cameras through Zhang's calibration method, and converts image coordinates to a same scale and reference frame to ensure an accuracy of subsequent matching calculations; wherein image correction is to project two images onto a plane so that the corresponding points of each scene point overlap on a horizontal line; feature matching uses a global matching algorithm to match corresponding pixels of the physical scene points in left and right images, and performs a disparity estimation; and disparity smoothing and post-processing uses filtering and smoothing algorithms for post-processing to solve a problem of noise and discontinuity in disparity maps and improve a continuity and edge retention of disparity maps.

### Claim 4 (depends on 1)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 1, wherein in S4, the bilinear interpolation method is based on information of four nearest pixels to be estimated, wherein the four nearest pixels are in the upper left, the upper right, the lower left, and the lower right positions; and wherein a weighted average is performed according to relative distances between pixels and the target point, thereby achieving a smooth transition that converts the captured surface information into planar data.

### Claim 5 (depends on 1)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 1, wherein in S5, the improvement of the small sample counting model is to introduce a self-attention mechanism and a feature enhancement module into the model to enhance a mutual relationship between different spatial locations.

### Claim 6 (depends on 1)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 1, wherein in S6, a position of each target refers to a center point coordinate of each coral soil particle.

### Claim 7 (depends on 1)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 1, wherein in S7, the segmentation model uses Segment Anything Model; the center point coordinates of the coral soil particles are used as point prompt, and the “coral soil particles” are input into the segmentation model as text prompt to segment the image, and wherein the segmentation result of the coral soil particles refers to the contour coordinates of the coral soil particles.

### Claim 8 (depends on 1)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 1, wherein in S9, BoT-SORT optimization refers to adding convexity, sphericity, roundness, and Feret diameter to the Kalman filter of BoT-SORT.

### Claim 9 (depends on 1)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 1, wherein in S9, the crushing matching mechanism refers to using Harris corner detection to extract the feature points in the image, and then matching the feature points through Fast Library for Approximate Nearest Neighbors to realize a crushing matching between the original particles and the broken particles; wherein, after the matching between the original particles and the broken particles, the broken particles will inherit the identity of the original particles, and a dimension based on an original identity is added, wherein the identity of the broken particles will increase by one dimension, thus achieving the purpose of traceability.

### Claim 10 (depends on 1)

The particle tracking method for coral particles in soil based on the multimodal model according to claim 1, wherein in S10, displacement and breakage of soil particles in coral are obtained by combining the two-dimensional plane tracking information with the particle information in the three-dimensional cloud image.