CN-121982054-A - Physical model driven method and system for segmenting candy atomizing video

CN121982054ACN 121982054 ACN121982054 ACN 121982054ACN-121982054-A

Abstract

The invention discloses a physical model driven method and a physical model driven system for segmenting a candy material atomized video, and relates to the technical field of intersection of computer vision and digital image processing. The method comprises the steps of obtaining a video sequence of a candy material atomization process, and performing preliminary frame-by-frame foreground segmentation on the video sequence by adopting a mixed Gaussian model to obtain an initial foreground mask corresponding to each frame. For the initial foreground mask of each frame, at least one foreground region is extracted, and feature information of each foreground region is calculated. Based on the atomized fluid mechanics model, and by utilizing the matching relation of the foreground regions among the continuous frames, the physical constraint condition for verifying the foreground regions is constructed. And verifying the successfully matched foreground region by using a physical constraint condition, screening out the region which does not accord with the physical rule based on the verification result, and reserving the region which accords with the physical rule as a real foreground region. Based on all the real foreground regions, a final foreground segmentation result for each frame is generated.

Inventors

XU BINGYANG
LIU CIDE
LIANG JIAXIN
LU DI
HU RONGHUI

Assignees

江西中烟工业有限责任公司

Dates

Publication Date: 20260505
Application Date: 20251223

Claims (10)

1. A physical model driven method for segmenting a video of atomizing a candy material is characterized by comprising the steps of obtaining a video sequence of the candy material in an atomizing process, and carrying out preliminary foreground segmentation on the video sequence frame by adopting a mixed Gaussian model to obtain an initial foreground mask corresponding to each frame; extracting at least one foreground region from the initial foreground mask of each frame, and calculating characteristic information of each foreground region; based on an atomization hydrodynamic model, constructing a physical constraint condition for verifying a foreground region by utilizing a matching relation of the foreground region between continuous frames; verifying the successfully matched foreground region by using a physical constraint condition, screening out a region which does not accord with a physical rule based on a verification result, and reserving the region which accords with the physical rule as a real foreground region; Based on all the real foreground regions, a final foreground segmentation result for each frame is generated.
2. The method for segmenting a physical model driven candy material atomized video according to claim 1, wherein the characteristic information comprises centroid coordinates and areas of a foreground region; the physical constraints include motion consistency constraints, size distribution constraints, and spatial distribution constraints.
3. The method for partitioning a physical model driven candy material atomized video according to claim 1, wherein said matching relationship using successive inter-frame foreground regions comprises: for each frame in the video sequence except for the first frame, each foreground region of the frame is matched to the foreground region in the previous frame.
4. A physical model driven method of video segmentation of a binder composition according to claim 3, wherein the motion consistency constraint requires that the motion speed of successfully matched foreground regions between successive frames satisfies the following condition: In the formula, A minimum speed threshold value preset according to the characteristic parameters of the nozzle and the air flow speed; a maximum speed threshold value preset according to the characteristic parameters of the nozzle and the air flow speed; Calculating a velocity vector based on the change of the centroid position of the same foreground region between the continuous frames by matching the region; is a model of the velocity vector; the module value calculation formula of the speed vector is as follows: In the formula, For the moment of time Centroid coordinates of the matching regions in the corresponding frames; For the moment of time Centroid coordinates of the matching regions in the corresponding frames; is the time interval between adjacent frames.
5. The method of claim 4, wherein the motion consistency constraint further comprises a direction consistency constraint that requires a motion direction angle between successive frames for a successfully matched foreground region With a preset expected movement direction angle Not exceeding a preset angular tolerance threshold : The movement direction angle The calculation formula of (2) is as follows: In the formula, As a four-quadrant arctangent function.
6. The method for physical model driven video segmentation of candy atomizing according to claim 2, wherein the size distribution constraint requires a predicted atomizing diameter of a foreground region The method meets the following conditions: In the formula, A minimum atomization diameter threshold preset according to the characteristic parameters of the nozzle; a maximum atomization diameter threshold value preset according to the characteristic parameters of the nozzle; The estimated atomization diameter The calculation formula of (2) is as follows: In the formula, An apparent pixel radius for the foreground region; Distance from foreground region to camera; Is the equivalent focal length of the camera.
7. The method of claim 6, wherein the size distribution constraint further requires that the estimated size distribution of all foreground regions matches a theoretical atomized size distribution model; the theoretical atomization size distribution model adopts Rosin-Rammler distribution, and the probability density function of the distribution is as follows: In the formula, A theoretical reference diameter predetermined according to the characteristic parameters of the nozzle; Is a distribution index predetermined according to the nozzle characteristic parameters.
8. The method for physically model driven video segmentation of sweetened material according to claim 2, wherein the spatial distribution constraint requires foreground region density within any preset rectangular local window in the image Not exceeding the maximum allowable density The maximum allowable density Determining according to the nozzle position and the airflow diffusion model; The foreground region density The calculation formula of (2) is as follows: In the formula, In terms of coordinates The number of foreground areas in a preset rectangular local window serving as the center; in terms of coordinates The area of the preset rectangular local window is the center.
9. A physical model driven method of video segmentation for atomizing a sugar material according to claim 3, wherein the verifying the successfully matched foreground region using physical constraints comprises: For each physical constraint, calculating a single coincidence score of successfully matched foreground regions ; Scoring based on single item compliance Calculating a total coincidence score of successfully matched foreground regions The calculation formula is as follows: In the formula, Is the total number of physical constraints; Is the first Weight coefficients corresponding to the term physical constraint conditions; if the total coincidence degree is scored And if the foreground region exceeds the preset threshold value, judging the foreground region as a real foreground region.
10. A physical model driven candy atomizing video segmentation system, based on the method of any one of claims 1 to 9, comprising: the video acquisition and primary segmentation module is used for acquiring a video sequence of the candy material atomization process, and carrying out primary foreground segmentation on the video sequence frame by adopting a mixed Gaussian model to obtain an initial foreground mask corresponding to each frame; The foreground extraction and feature calculation module is used for extracting at least one foreground region for the initial foreground mask of each frame and calculating feature information of each foreground region; The physical constraint construction module is used for constructing physical constraint conditions for verifying the foreground region based on the atomization hydrodynamic model and by utilizing the matching relation of the foreground region between the continuous frames; The foreground verification and screening module is used for verifying successfully matched foreground areas by utilizing physical constraint conditions, screening out areas which do not accord with the physical rules based on verification results, and reserving the areas which accord with the physical rules as real foreground areas; And the segmentation result generation module is used for generating a final foreground segmentation result of each frame based on all the real foreground areas.

Description

Physical model driven method and system for segmenting candy atomizing video Technical Field The invention relates to the technical field of intersection of computer vision and digital image processing, in particular to a physical model driven method and a physical model driven system for dividing a candy atomizing video. Background In the industrial production of cigarettes, flavoring and charging are core process links for determining the consistency of sensory quality and batches of cigarettes, and the core process links are that liquid feed liquid such as essence, sugar and the like is atomized into fine liquid drops through high-speed air flow or pressure so as to realize uniform and accurate application on the surfaces of tobacco shreds. In recent years, machine vision technology based on high-speed industrial cameras has become an important means for online monitoring of atomization processes. However, due to the dynamic complexity of the candy atomizing scene, the traditional image segmentation algorithm has significant limitation when being directly applied, and is difficult to meet the requirement of industrial-level accurate monitoring. The method has the advantages that firstly, atomized particles move at a high speed and are easy to generate motion blur, and meanwhile, the particles are frequently overlapped and shielded, so that boundary confusion is further caused, and the segmentation accuracy is affected. Secondly, most of the method only depends on apparent characteristics such as pixel colors and textures, the segmentation logic is seriously different from the inherent physical rule of the atomization process, and the true fog drops and various false prospects are difficult to distinguish, so that the physical reliability of analysis results is low, and the method cannot be directly used for accurate process regulation. Disclosure of Invention Aiming at the problems of low accuracy, insufficient result reliability and the like of the traditional candy atomizing video segmentation method, the invention provides a candy atomizing video segmentation method driven by a physical model and a system thereof. The method takes the atomized fluid mechanics model as a theoretical support, builds a complete technical system comprising visual perception, characteristic association, physical verification and accurate generation, improves the accuracy and the robustness of atomized particle segmentation in a complex dynamic scene, ensures that a segmentation result has definite physical interpretability and high credibility, and further provides a direct and reliable data basis for accurate quantification of key characteristics such as atomized particle size distribution, a speed field and the like and subsequent process optimization. The invention realizes the aim through the following technical scheme: A physical model driven method for segmenting a candy material atomized video comprises the following steps: Obtaining a video sequence of a candy material atomization process, and performing preliminary foreground segmentation on the video sequence frame by adopting a mixed Gaussian model to obtain an initial foreground mask corresponding to each frame; extracting at least one foreground region from the initial foreground mask of each frame, and calculating characteristic information of each foreground region; based on an atomization hydrodynamic model, constructing a physical constraint condition for verifying a foreground region by utilizing a matching relation of the foreground region between continuous frames; verifying the successfully matched foreground region by using a physical constraint condition, screening out a region which does not accord with a physical rule based on a verification result, and reserving the region which accords with the physical rule as a real foreground region; Based on all the real foreground regions, a final foreground segmentation result for each frame is generated. As a preferred scheme of the invention, the characteristic information comprises centroid coordinates and areas of the foreground region; the physical constraints include motion consistency constraints, size distribution constraints, and spatial distribution constraints. As a preferred embodiment of the present invention, the matching relationship using the foreground region between consecutive frames includes: for each frame in the video sequence except for the first frame, each foreground region of the frame is matched to the foreground region in the previous frame. As a preferred embodiment of the present invention, the motion consistency constraint requires that the motion speed of successfully matched foreground regions between consecutive frames satisfies the following condition: ; In the formula, A minimum speed threshold value preset according to the characteristic parameters of the nozzle and the air flow speed; a maximum speed threshold value preset according to the characteristic parameters of t