CN-121982619-A - Video intelligent acquisition and processing method for mine monitoring

CN121982619ACN 121982619 ACN121982619 ACN 121982619ACN-121982619-A

Abstract

The invention relates to the technical field of image communication, in particular to a video intelligent acquisition and processing method for mine monitoring, which is characterized by extracting a current video frame at a current moment from a monitoring video stream of a target mine area, acquiring semantic areas corresponding to each preset semantic category in the current video frame by utilizing a semantic segmentation model, evaluating the analysis value of each semantic area at the current moment in real time according to the optical flow and dynamic characteristics of each semantic area and the priori importance weight of each semantic area corresponding to the preset semantic category to obtain comprehensive dynamic value scores, and differentially processing the monitoring video of the target mine area in a future period according to the comprehensive dynamic value scores of each semantic area, so that the intelligent processing of the monitoring video of the target mine area is realized under the condition of limited total calculation resources, and the recognition and early warning capability of key risks are ensured and enhanced while the processing efficiency of the system is remarkably improved.

Inventors

LI ZHENGZHONG
LI BOGUANG
LV FUXIANG
WANG QISHUAI
ZHANG ZHAO

Assignees

济南福深兴安科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260403

Claims (7)

1. The intelligent video acquisition and processing method for mine monitoring is characterized by comprising the following steps of: extracting video frames at each moment in a preset time interval up to the current moment from a monitoring video stream of a target mine area, recording the video frames at the current moment as current video frames, and acquiring at least one semantic area corresponding to at least one preset semantic category in the current video frames by utilizing a trained semantic segmentation model; For any semantic region, according to the position of the any semantic region in a current video, acquiring a region with the same position in a video frame at the moment previous to the current moment, marking the region as a reference region of the any semantic region, and according to the optical flow of the any semantic region and the gray level difference between the any semantic region and the reference region, acquiring a dynamic liveness score of the any semantic region; Acquiring dynamic liveness scores of reference areas of any semantic area in video frames at each moment before the current moment, and acquiring comprehensive dynamic value scores of any semantic area according to the persistence and stability of all dynamic liveness scores and prior importance weights of the corresponding preset semantic categories of the any semantic area; and acquiring the comprehensive dynamic value score of each semantic region in the current video frame, and differentially processing the monitoring video of the target mine region in a future period according to the comprehensive dynamic value score of each semantic region to realize intelligent processing of the monitoring video of the target mine region.
2. The method for intelligent video acquisition and processing for mine monitoring according to claim 1, wherein the step of obtaining the dynamic liveness score of any semantic region according to the optical flow of the any semantic region and the gray level difference between the any semantic region and the reference region comprises the following steps: Acquiring the optical flow amplitude of each pixel point in any semantic region by using Farneback optical flow method, and accumulating all the optical flow amplitudes to obtain a motion intensity characteristic value of any semantic region; According to the position of each pixel point in any semantic region, acquiring pixel points with the same position in the reference region, recording the pixel points as reference pixel points of each pixel point in any semantic region, respectively carrying out gray processing on a current video frame and a video frame at the moment of the current moment to obtain a gray value of each pixel point in any semantic region in the current video and a gray value of each reference pixel point in the video frame at the moment of the current moment, and respectively calculating a difference absolute value between the gray value of each pixel point in any semantic region and the gray value of the reference pixel point to obtain a gray value difference; Calculating the average value and standard deviation of all gray value differences, marking the sum between the standard deviation and the average value which are preset multiples as a gray difference threshold value, marking the pixel points corresponding to the gray value differences which are larger than or equal to the gray difference threshold value as changed pixel points, calculating the duty ratio of the number of all changed pixel points in the number of all pixel points in any semantic region, and taking the product between a preset scaling factor and the duty ratio as a pixel change ratio characteristic value of the any semantic region; And taking the number of all pixel points in any semantic region as a denominator, and taking the sum of the motion intensity characteristic value and the pixel change ratio characteristic value as a molecule to obtain the dynamic liveness score of any semantic region.
3. The method for intelligent video acquisition and processing for mine monitoring according to claim 1, wherein the obtaining the comprehensive dynamic value score of any semantic region according to the persistence and stability of all dynamic liveness scores and the priori importance weight of the corresponding preset semantic category of any semantic region comprises: Obtaining standard deviation of all dynamic liveness scores to obtain a continuous liveness factor of any semantic region; Acquiring a state transition factor of any semantic region according to the difference between the dynamic liveness score of the any semantic region and the dynamic liveness score of each reference region of the any semantic region; And acquiring the comprehensive dynamic value score of any semantic region according to the dynamic liveness score, the continuous liveness factor and the state transition factor of the any semantic region and the priori importance weight of the corresponding preset semantic category of the any semantic region.
4. The method for intelligent video acquisition and processing for mine monitoring according to claim 3, wherein the step of obtaining the state transition factor of any semantic region according to the difference between the dynamic liveness score of the any semantic region and the dynamic liveness score of each reference region of the any semantic region comprises the steps of: Acquiring the mean value and standard deviation of the dynamic liveness scores of all the reference areas of any semantic area, and respectively marking the mean value and standard deviation as the mean liveness and liveness fluctuation degree; Taking the sum of the activity fluctuation degree and a preset constant as a denominator, taking the absolute value of the difference between the dynamic activity score of any semantic region and the average activity as a molecule to obtain the activity deviation degree of any semantic region, and taking the activity deviation degree as an independent variable of a hyperbolic tangent function to obtain the state transition factor of any semantic region.
5. The method for intelligent video acquisition and processing for mine monitoring according to claim 3, wherein the obtaining the comprehensive dynamic value score of any semantic region according to the dynamic liveness score, continuous liveness factor, state transition factor of the any semantic region and the prior importance weight of the any semantic region corresponding to the preset semantic category comprises the following steps: respectively carrying out normalization processing on the dynamic liveness score and the continuous liveness factor of any semantic region to obtain a dynamic liveness score normalization value and a continuous liveness factor normalization value; Weighting the dynamic liveness score normalization value, the continuous liveness factor normalization value, the state transition factor and the prior importance weight of any semantic region corresponding to the preset semantic category according to the dynamic liveness score, the continuous liveness factor, the state transition factor and the preset weight coefficient corresponding to the prior importance weight of any semantic region corresponding to the preset semantic category respectively to obtain a corresponding weighted dynamic liveness score normalization value, a weighted continuous liveness factor normalization value, a weighted state transition factor and a weighted prior importance weight; And calculating the sum among the weighted dynamic liveness score normalization value, the weighted priori importance weight and the weighted state transition factor to obtain a sum result, and subtracting the weighted continuous liveness factor normalization value from the sum result to obtain the comprehensive dynamic value score of any semantic region.
6. The intelligent video acquisition and processing method for mine monitoring according to claim 1, wherein the differential processing of the monitoring video of the target mine area in the future period according to the comprehensive dynamic value score of each semantic area, to realize the intelligent processing of the monitoring video of the target mine area, comprises the following steps: for any semantic region, if the comprehensive dynamic value score of any semantic region is greater than or equal to a preset high-value threshold, determining the value grade of each pixel point in the any semantic region as high value; If the comprehensive dynamic value score of any semantic region is smaller than a preset high value threshold and larger than a preset low value region, determining the value grade of each pixel point in any semantic region as a medium value; If the comprehensive dynamic value score of any semantic region is smaller than or equal to a preset low value threshold, determining that the value grade of each pixel point in any semantic region is low; Acquiring the value grade of each pixel point in each semantic region in the current video frame, constructing an image mask of the current video frame according to the value grade of each pixel point in each semantic region in the current video frame, and marking the image mask as a self-adaptive frame acquisition decision graph; And according to the self-adaptive frame acquisition decision diagram, the monitoring video of the target mine area in the future period is differentially processed, so that the intelligent processing of the monitoring video of the target mine area is realized.
7. The intelligent video acquisition and processing method for mine monitoring according to claim 6, wherein the step of differentially processing the monitoring video of the target mine area in the future period according to the adaptive frame acquisition decision diagram to realize intelligent processing of the monitoring video of the target mine area comprises the following steps: Each value grade in the self-adaptive frame acquisition decision diagram corresponds to a preset processing decision, any value grade in the self-adaptive frame acquisition decision diagram is aimed at, a video frame at a future processing moment is acquired in a monitoring video stream of a target mine area in a future period according to the preset processing decision corresponding to the any value grade, pixel points with the same position are acquired in the video frame at the future processing moment according to the position of the pixel point corresponding to the any value grade in the self-adaptive frame acquisition decision diagram, an ROI image of the any value grade in the video frame at the future processing moment is formed, the ROI image is input into a preset model corresponding to the any value grade, and analysis processing is carried out on the ROI image.

Description

Video intelligent acquisition and processing method for mine monitoring Technical Field The invention relates to the technical field of image communication, in particular to an intelligent video acquisition and processing method for mine monitoring. Background In mine safety production and intelligent management, the video monitoring system plays a core role of an electronic sentry. The method is characterized in that mass video stream data are collected uninterruptedly through cameras deployed in key areas such as mining areas, transportation galleries, dumping grounds, dispatching centers and the like. The video data contains key information such as equipment running state, personnel operation behaviors, environment security situations (such as slope stability and dust concentration) and the like, and is an important basis for realizing risk early warning, violation identification, flow optimization and accident tracing. Through intelligent analysis to mine monitoring video, the enterprise can grasp production site dynamic in real time, and potential safety hazards (such as personnel intrude into dangerous areas, equipment abnormal operation, initial signs of slope sliding and the like) are automatically identified, so that safety management efficiency is improved, and accident rate is reduced. In the prior art, when mine monitoring videos are processed, an indiscriminate and fixed strategy frame acquisition and analysis mode is generally adopted, but because the monitoring videos have background areas (such as remote mountains and fixed buildings) which are stationary or change slowly for a long time, equipment areas (such as constant-speed running conveyor belts) which move regularly, key attention areas (such as dangerous side slopes and personnel activity areas) with strong randomness and burstiness and the like, the frame acquisition and analysis mode of the fixed strategy cannot fully consider the huge difference of information values and change rates of different areas and different targets in a monitoring scene in a space-time dimension, so that the waste of calculation resources and the timeliness of analysis are reduced. Therefore, how to ensure and enhance the identification and early warning capability of the critical risk while significantly improving the processing efficiency of the system is a problem to be solved. Disclosure of Invention In view of the above, the embodiment of the invention provides a video intelligent acquisition processing method for mine monitoring, which aims to solve the problem of ensuring and enhancing the recognition and early warning capability of key risks while obviously improving the processing efficiency of a system. The embodiment of the invention provides a video intelligent acquisition processing method for mine monitoring, which comprises the following steps: extracting video frames at each moment in a preset time interval up to the current moment from a monitoring video stream of a target mine area, recording the video frames at the current moment as current video frames, and acquiring at least one semantic area corresponding to at least one preset semantic category in the current video frames by utilizing a trained semantic segmentation model; For any semantic region, according to the position of the any semantic region in a current video, acquiring a region with the same position in a video frame at the moment previous to the current moment, marking the region as a reference region of the any semantic region, and according to the optical flow of the any semantic region and the gray level difference between the any semantic region and the reference region, acquiring a dynamic liveness score of the any semantic region; Acquiring dynamic liveness scores of reference areas of any semantic area in video frames at each moment before the current moment, and acquiring comprehensive dynamic value scores of any semantic area according to the persistence and stability of all dynamic liveness scores and prior importance weights of the corresponding preset semantic categories of the any semantic area; and acquiring the comprehensive dynamic value score of each semantic region in the current video frame, and differentially processing the monitoring video of the target mine region in a future period according to the comprehensive dynamic value score of each semantic region to realize intelligent processing of the monitoring video of the target mine region. Preferably, the obtaining the dynamic liveness score of the any semantic region according to the optical flow of the any semantic region and the gray level difference between the any semantic region and the reference region includes: Acquiring the optical flow amplitude of each pixel point in any semantic region by using Farneback optical flow method, and accumulating all the optical flow amplitudes to obtain a motion intensity characteristic value of any semantic region; According to the position of each pixel poin