CN118101970B - Efficient communication method of monitoring images for ice and snow sports venues based on deep learning - Google Patents

Abstract

The present application relates to the field of video surveillance technology, and specifically to an efficient communication method for monitoring images of ice and snow venues based on deep learning, the method comprising: collecting each video frame in the monitoring video of the ice and snow venue; determining the energy function value of the snow sports track of each sampling point in the video frame based on the intensity value of each sampling point in the video frame under different channels and the position and distance distribution with any optical flow vector; determining the snow information density splitting weight of the vector point based on the energy function value of the snow sports track and the moving position of the centroid of the vector point in the iteration process; optimizing the VQ-LGB compression algorithm to obtain the compressed video frame, and inputting the compressed video frame into the neural network to realize efficient communication of the ice and snow venue. The present application aims to obtain higher quality video compression data of the ice and snow venue and improve communication efficiency.

Description

Efficient communication method of monitoring images for ice and snow sports venues based on deep learning

Technical Field

The present application relates to the field of video surveillance technology, and in particular to an efficient communication method for monitoring images of ice and snow project sites based on deep learning.

Background technique

As a kind of sports, ice and snow events require safety monitoring of participants in daily venues. Since some ice and snow sports are high-speed sports, when people fall in the venue, timely response should be made to prevent the fallen people from being hit and then suffering secondary injuries. However, if the ice and snow venue is too large, long-term pure manual inspection will cause fatigue of the monitoring personnel, resulting in poor supervision.

Therefore, this application uses a deep learning method to collect video frames of ice and snow venues in real time, collects them at the monitoring terminal, identifies the fallen person in the picture and issues a reminder. Because ice and snow venues are usually too large, resulting in too much data collected in real time, the collected video frames need to be compressed and decompressed before and after communication to ensure efficient communication. This application uses the VQ-RGB compression algorithm to further improve the real-time and monitoring quality of ice and snow venue monitoring, and improves the traditional VQ-RGB compression algorithm to improve the communication quality of the monitoring video.

Summary of the invention

In order to solve the above technical problems, the present application provides an efficient communication method for monitoring images of ice and snow project sites based on deep learning to solve the existing problems.

The deep learning-based efficient communication method for monitoring images of ice and snow sports venues in this application adopts the following technical solutions:

An embodiment of the present application provides an efficient communication method for monitoring images of ice and snow sports venues based on deep learning, the method comprising the following steps:

Collect each video frame from the monitoring video of the ice and snow event venue;

Based on the intensity values of each sampling point in the video frame under different channels, the snow color highlight segmentation weight of each sampling point is determined;

The sparse optical flow algorithm is used to obtain the optical flow vectors of adjacent video frames; based on the position distribution of each sampling point in the video frame and any optical flow vector, the degree of coincidence between each sampling point and any optical flow vector is determined;

The optical flow vector non-oscillation degree of any optical flow vector is determined based on the distance distribution of the optical flow vectors in adjacent video frames; the energy function value of the snow sports track at each sampling point of the video frame is determined based on the snow color highlight segmentation weight, optical flow vector non-oscillation degree and direction consistency at the same sampling point of the continuous reference frame;

Vector points in the video frame are randomly generated, and the snow information density splitting weight of the vector point is determined based on the energy function value of the snow sports track and the moving position of the centroid of the vector point in the iteration process during any iteration of the video frame using the LGB algorithm in the VQ-LGB algorithm;

Based on the snow information density split weight optimization VQ-LGB compression algorithm, the compressed video frames are obtained and input into the neural network to achieve efficient communication in the ice and snow sports venues;

Based on the intensity values of each sampling point in the video frame under different channels, the snow color highlight segmentation weight of each sampling point is determined, and the expression is:

In the formula, is the snow color highlight segmentation weight of the mth sampling point, , are the intensity values of the mth sampling point in the dark channel and bright channel video frames, respectively. , , are the intensity values of the mth sampling point in the R, G, and B channels respectively;

Determining the degree of direction fit between each sampling point and any optical flow vector based on the position distribution of each sampling point and any optical flow vector in the video frame includes: connecting each sampling point with the midpoint of any optical flow vector, and taking the angle between the connecting line and any optical flow vector as the degree of direction fit between each sampling point and any optical flow vector;

The step of determining the optical flow vector non-oscillation degree of any optical flow vector based on the distance distribution of the optical flow vectors in adjacent video frames includes:

For any optical flow vector in each video frame, the optical flow vector whose midpoint is closest to the midpoint of any optical flow vector in the video frame of the previous adjacent frame is recorded as the associated optical flow vector of any optical flow vector;

The standard deviation of the lengths of all associated optical flow vectors of any optical flow vector is recorded as the optical flow vector non-oscillation degree of any optical flow vector.

Preferably, the method of determining the snow sports track energy function value of each sampling point of the video frame based on the snow color highlight segmentation weight, optical flow vector non-oscillation degree and direction consistency at the same sampling point of the continuous reference frame includes:

Recording a preset number of video frames adjacent to any video frame as reference frames of any video frame;

For any optical flow vector in the reference frame, the product of the length of any optical flow vector and the snow color highlight segmentation weight of each sampling point in the reference frame is calculated as the numerator;

Calculate the product of the non-oscillation degree of the optical flow vector of any optical flow vector, the degree of direction consistency between each sampling point in the reference frame and any optical flow vector, and the Euclidean distance between each sampling point in the reference frame and any optical flow vector, and use the sum of the three products and the preset parameter adjustment factor as the denominator;

Calculate the ratio of the numerator to the denominator; and use the sum of the ratios of all optical flow vectors in all reference frames as the energy function value of the snow sports track at each sampling point in any video frame.

Preferably, the method of finding the moving position of the centroid based on the energy function value of the snow sports track and the vector point in the iteration process to determine the snow information density splitting weight of the vector point includes:

For any vector point before any iteration process, a traditional vector point splitting coefficient of any vector point is determined based on any vector point and all its associated sampling points; the associated sampling points are sampling points within the region where any vector point is located;

For each movement of any vector point in the iterative process, the average of the energy function values of the snow sports track of all associated sampling points of any vector point before each movement is calculated, and recorded as the energy value of the vector point before each movement of any vector point; the energy values of the vector points before all the movements are combined into an energy function change vector of any vector point;

The mean value of the first-order derivative vector of the energy function change vector is recorded as the energy change tendency of the snow sports track at any vector point;

Determine the weight of the centroid movement position of each movement of any vector point based on the Euclidean distance between the movement positions of each vector point in the iterative process;

The snow information density splitting weight of any vector point is determined based on the weight of the center of mass movement position, the energy change tendency of the snow sports track and the traditional vector point splitting coefficient.

Preferably, the method of determining a traditional vector point splitting coefficient of any vector point based on any vector point and all associated sampling points thereof includes:

The standard deviation of the grayscale values of all associated sampling points of any vector point is recorded as the distortion of any vector point; the product of the number of associated sampling points of any vector point and the distortion is taken as the traditional vector point splitting coefficient of any vector point.

Preferably, the weight of the centroid movement position of each movement of any vector point is determined based on the Euclidean distance between the movement positions of each vector point in the iterative process, and the expression is:

In the formula, is the weight of the center of mass movement of the hth vector point for the gth movement, is the position of the h-th vector point before the g-th movement, is the position of the hth vector point after G moves, yes and The Euclidean distance between It is the maximum value of the Euclidean distance of all movement times of the h-th vector point.

Preferably, the snow information density splitting weight of any vector point is determined based on the mass center moving position weight, the energy change tendency of the snow sports track and the traditional vector point splitting coefficient, including:

Calculate the product of the vector point energy value before each movement of any vector point and the weight of the center of mass movement position of each movement of any vector point, and calculate the sum of the products under all movement times of any vector point;

Obtaining a maximum value among the energy change tendencies of the snow sports track of all vector points, and calculating a sum of the maximum value and the energy change tendencies of the snow sports track of any vector point;

The product of the two sums and the three traditional vector point splitting coefficients of any vector point is used as the snow information density splitting weight of any vector point.

Preferably, the VQ-LGB compression algorithm is optimized based on the snow information density split weight to obtain compressed video frames, which are input into the neural network to achieve efficient communication of ice and snow project venues, including:

The snow information density splitting weight of any vector point is used to replace the traditional vector point splitting coefficient, and the VQ-LGB compression algorithm is improved to obtain the compressed video frame.

The compressed video frames are decompressed in the data processing center to obtain decompressed video frames; a label value is manually set for the decompressed video frames, where a label value of 1 represents that a person has fallen, and a label value of 0 represents that no person has fallen;

Input all decompressed video frames and corresponding label values into the convolutional network and output the CNN convolutional network decision function;

The decompressed video frames transmitted in real time are used as the input of the CNN convolutional network decision function, and the label value is output; when the output label value is 1, an alarm is issued; otherwise, no alarm is issued.

This application has at least the following beneficial effects:

The present application calculates the grayscale values of the dark channel, bright channel and three primary color channels of the video frame to obtain the snow color highlight segmentation weight, which is used to accurately distinguish the snow background from the athletes wearing colorful clothes; further, the direction consistency is obtained through the angle formed between the sparse optical flow vector and the sampling point, and the length standard deviation of the optical flow vector in a certain area of the video frame is used to improve the recognition of the optical flow vector caused by the oscillation of background objects and the optical flow vector caused by the athlete's movement, and the snow color highlight segmentation weight is combined to calculate the snow sports track energy function value, and accurately locate the athlete's movement in the video frame; further, according to the snow sports track energy function value and the center of mass movement operation of the vector point in the VQ-LGB algorithm, the vector point energy value and the snow sports track energy change tendency are obtained to indicate whether the vector point is near The athlete positions in the snow monitoring map and whether the past search paths of the vector points are directed towards the athlete positions in the snow monitoring map are used to mine valuable information in historical mobile data; further, the movement of the centroid operation, the energy value of the vector points and the energy change tendency of the snow sports track are combined to calculate the snow information density splitting weight of the vector points, so that when splitting the vector points in the VQ-LGB algorithm, the athlete movement positions in the video frame are tended to be selected, so that when the VQ-LGB algorithm compresses the video frame, more vector points are generated at the athlete positions in the video frame, and fewer vector points are generated at other positions. The compression results with low distortion are obtained faster than traditional algorithms, and higher-quality monitoring compression data of ice and snow venues can be obtained while accelerating the compression speed, improving communication quality and efficiency.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

FIG1 is a flow chart of an efficient communication method for monitoring images of ice and snow sports venues based on deep learning provided by the present application;

FIG2 is a schematic diagram of the degree of consistency of the direction of the optical flow vector;

FIG3 is a flowchart of indicator construction for the video frame compression process.

Detailed ways

In order to further explain the technical means and effects adopted by this application to achieve the predetermined invention purpose, the specific implementation method, structure, features and effects of the efficient communication method of ice and snow project site monitoring images based on deep learning proposed in this application are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

The specific scheme of the efficient communication method for monitoring images of ice and snow project sites based on deep learning provided by this application is described in detail below with reference to the accompanying drawings.

An embodiment of the present application provides an efficient communication method for monitoring images of ice and snow project sites based on deep learning.

Specifically, the following deep learning-based efficient communication method for monitoring images of ice and snow sports venues is provided. Please refer to FIG1. The method includes the following steps:

Step S001, collecting video frames of the monitoring video of the ice and snow sports venue.

Collect videos of ice and snow venues in real time, where the nth frame of a video in a certain monitoring area is recorded as the snow monitoring video frame , the color space of the video frame is RGB color space, and the snow monitoring video frame is further As input, the RGB to grayscale color space conversion algorithm is used, and the output is a video frame The color space conversion algorithm is a well-known technology in the art and will not be described in detail in this embodiment.

At this point, each video frame of the monitoring video of the ice and snow sports venue can be obtained.

Step S002, analyzing the video frame, and optimizing the traditional vector point splitting coefficient of any vector point at each iteration number in the VQ-RGB algorithm.

When the traditional VQ-RGB algorithm performs region segmentation on video frames, the sampling points in the same region are represented by the same vector type to achieve the purpose of video compression. However, in snow monitoring videos, most video frames do not contain snow scenes where people fall down. When the traditional algorithm compresses them, the RGB algorithm consumes a lot of iterative computing resources on the snow during the iterative operation of region segmentation, which reduces the algorithm compression speed and the communication speed of snow monitoring.

Therefore, this embodiment first takes the nth video frame as input, and uses the dark channel algorithm and the bright channel algorithm for calculation, and outputs the dark channel video frame and the bright channel video frame respectively. The dark channel algorithm and the bright channel algorithm are well-known technologies in the art and will not be described in detail.

Calculate the snow color highlight segmentation weights for each sampling point in the nth video frame:

In the formula, is the snow color highlight segmentation weight of the mth sampling point, , are the intensity values of the mth sampling point in the dark channel and bright channel video frames, respectively. , , are the intensity values of the mth sampling point in the R, G, and B channels respectively.

In the formula, the background in the snow is mostly the white snow part. The difference between the dark channel intensity value and the bright channel intensity value of the sampling point in this part is small. Therefore, the difference between the dark and bright channels of the m-th sampling point is used as the weight. The larger the difference, the less likely it is the background snow. In the formula, the R, G, and B channels of the m-th sampling point are subtracted and then summed. The larger the value, the greater the difference in the intensity value of the m-th sampling point in the RGB channel, and the brighter the color of the m-th sampling point. At the same time, it is divided by the dark channel intensity value of the m-th sampling point. The smaller the dark channel intensity value, the brighter the color of the m-th sampling point. In the ice and snow venues, in order to ensure safety, athletes often wear bright and bright clothes. Therefore, the brighter the color, the less likely it is the background snow.

Finally, the snow color highlight segmentation weight is obtained. The larger the value, the more the corresponding sampling point is an athlete rather than a background, and the more details of this part should be retained when compressing the video frame.

Furthermore, since the skiing routes of the tracks in the ice and snow venues are usually relatively fixed, the areas where athletes often pass by can be selected according to the changes in the motion states of different areas in the video frames.

First, the nth and n-1th video frames are used as input, and the sparse optical flow algorithm is used to output the sparse optical flow vector set of the two video frames. There are multiple optical flow vectors in the set, and the starting and ending coordinates of the kth optical flow vector are recorded as , , representing the n-1th frame The sampling point of the position moves to The sparse optical flow algorithm is a well-known technology in the art and will not be described in detail.

The angle between the line from the mth sampling point to the midpoint of the kth optical flow vector in the rth video frame and the kth optical flow vector is recorded as the direction fit .

As shown in Figure 2, the arrows in Figure 2 are optical flow vectors, where m represents the mth sampling point in the current frame, k represents the kth optical flow vector pointing to the sampling point m in the previous frame, and Ia represents the angle between the sampling point m and the kth optical flow vector. The smaller Ia is, the closer the mth sampling point is to the motion path of the kth optical flow vector, and when evaluating the motion of the mth sampling point, the greater the influence of the kth optical flow vector on it.

At the same time, for the kth optical flow vector in the rth video frame, get the midpoint position , and at the same time obtain the midpoint position of the optical flow vector in the r-1th frame of the video frame and The optical flow vector with the closest Euclidean distance is used as the associated optical flow vector of the kth optical flow vector in the rth video frame, and so on, the corresponding associated optical flow vector can also be found in the r-2th frame.

Further, N associated optical flow vectors are obtained for the kth optical flow vector in the rth video frame, and the standard deviation of the lengths of these optical flow vectors is calculated and recorded as the optical flow vector non-oscillation degree of the kth optical flow vector in the rth video frame. The smaller the value, the more likely the kth optical flow vector is to be caused by objects that oscillate back and forth in the video frame, such as colorful flags in the wind and trees in the snow. For athletes in ice and snow events, their movements on the track are not swaying back and forth, so their optical flow vector non-oscillation value is higher.

Furthermore, the energy function value of the snow sports track at each sampling point in the video frame is calculated:

In the formula, is the snow sports track energy function value of the mth sampling point of the nth video frame, N is the number of reference frames, and in this embodiment, the value N=50, is the number of optical flow vectors in the r-th reference frame, is the length of the kth optical flow vector of the rth reference frame, is the snow color highlight segmentation weight of the mth sampling point in the rth reference frame, is the degree of consistency between the direction of the mth sampling point and the kth optical flow vector in the rth reference frame, is the Euclidean distance from the midpoint of the kth optical flow vector in the rth reference frame to the mth sampling point, is the non-oscillation degree of the optical flow vector of the kth optical flow vector in the rth reference frame, It is a parameter adjustment factor to prevent the denominator from being 0, and its value is 0.1.

In the formula, the longer the length of the sparse optical flow vector is, the farther the movement of the nearby sampling points is, and the more likely it is that it is a track crossed by athletes of ice and snow events; in the formula, the Euclidean distance and the degree of direction fit are used as weights, and the length of the optical flow vector is used to weight the influence value of the sampling point. The larger the value is, the farther the movement is near the m-th sampling point, and the larger the corresponding energy function value of the snow sports track is; in the formula, the larger the snow color highlight segmentation weight of the m-th sampling point is, the smaller the non-oscillation degree of its optical flow vector is, which means that the m-th sampling point is more likely to be an athlete in motion, and the corresponding energy function value of the snow sports track is larger.

Finally, the energy function value of the snow sports track is obtained. The larger the value, the more the m-th sampling point is on the track of the ice and snow venue. Correspondingly, more vector points should be generated for the sampling points of this part in the VQ-LGB algorithm to retain the detailed information of this part, while fewer vector points are generated for other parts, which helps to improve the communication quality and communication rate of the monitoring video.

Furthermore, the LGB algorithm in the VQ-LGB algorithm is improved to calculate the snow information density splitting weights of the vector points respectively, as follows:

The LGB algorithm is an optimized iterative algorithm in the VQ-LGB compression algorithm. At the beginning of the iteration, multiple vector points are randomly generated in the video frame. According to the Euclidean distance from the sampling point to the vector point, the vector point with the closest distance is used as the belonging vector point of the sampling point. Each vector point has multiple associated sampling points.

When the traditional algorithm is iterating, it will calculate the centroid of all the associated sampling points of the h-th vector point to move the h-th vector point until all the vector points are at the centroid of their associated sampling points. At this time, the standard deviation of the grayscale values of all the associated sampling points of the h-th vector point is calculated as the distortion of the h-th vector point. If the distortion is too large, it is considered that the vector point does not segment the video frame in detail. At this time, the vector point with a large distortion or a large number of associated sampling points is selected for splitting. During a certain iteration, the product of the distortion of the h-th vector point and the number of its associated sampling points is recorded as the traditional vector point splitting coefficient of the h-th vector point. .

Furthermore, from the last split vector point to the current split vector point, the hth vector point has multiple movements to find the center of mass. Assuming that from the last split vector point to the current split vector point, the hth vector point has undergone G movements to find the center of mass, then before the gth movement, the mean value of the energy function of the snow sports track of all associated sampling points is calculated, which is recorded as the vector point energy value , and then form the energy function change vector of the hth vector point in the movement process of this iteration , calculate the first-order derivative vector of the vector, and average the first-order derivative vector as the energy change tendency of the snow sports track , the positive or negative value represents whether the hth vector point is approaching or moving away from the high information area of the video frame during the search, and its absolute value represents the intensity of approaching or moving away. At the same time, the position before the gth movement is recorded as .

Calculate the snow information density splitting weight of the vector point:

In the formula, is the snow information density splitting weight of the h-th vector point, The traditional vector point splitting coefficient of the h-th vector point, is the energy change tendency of the snow sports track at the hth vector point, is the maximum value of the energy change tendency of the snow sports track among H vector points, is the total number of times the h-th vector point moves in this iteration to find the center of mass, is the energy value of the h-th vector point before the g-th movement, is the weight of the center of mass movement of the hth vector point for the gth movement;

is the position of the h-th vector point before the g-th movement, is the position of the hth vector point after G moves, yes and The Euclidean distance between It is the maximum value of the Euclidean distance of all movement times of the h-th vector point.

In the formula, the larger the traditional vector point splitting coefficient is, the greater the distortion of the h-th vector point is and the more associated sampling points are, the greater the probability of splitting a new vector point is; in the formula, the energy change tendency of the snow sports track is first non-negatively operated. After non-negative operation, the larger the value is, the closer the h-th vector point is to the athlete's position in the video frame when searching for the center of mass translation operation, and the more information needs to be retained in this scene. The corresponding h-th vector point has a greater probability of splitting a new vector point.

In the formula, the Euclidean distance from the current vector point position to the energy value of each vector point is reversely normalized as the weight of the center of mass movement position. The closer the historical center of mass movement search position is to the current vector point, the greater the weight. Then the vector point energy values are weighted and added. The larger the value, the closer the current position of the h-th vector point is to the player position in the video frame, and the greater the probability of the corresponding h-th vector point splitting into a new vector point.

Further, the video frame As input, the VQ-LGB compression algorithm is used for calculation. During the calculation process, the traditional vector point splitting coefficient of the hth vector point is Replaced with the snow information density splitting weight of the h-th vector point, and output as a compressed video frame Data. Among them, the indicator construction flow chart of the video frame compression process is shown in Figure 3.

Compared with the traditional VQ-LGB compression algorithm, the improved method described in this embodiment generates more vector points at the athlete positions in the video frame and fewer vector points at other positions, and can obtain low-distortion compression results faster than traditional algorithms. It can obtain higher-quality monitoring video compression data of ice and snow venues while accelerating the compression speed and improving communication quality and efficiency.

Further compress the video frame The data is transmitted to the data processing center of the ice and snow project.

Step S003, based on the compressed video frames, use a deep learning algorithm to process and achieve efficient communication at the ice and snow venues.

The data processing center of the ice and snow project uses compressed video frames The data is input, decompressed using the decompression algorithm of the VQ-LGB compression algorithm, and the output is the decompressed video frame .

Further, obtain the data set, which is multiple decompressed video frames , where the label value of each frame is set to 1 or 0, indicating whether there is a person falling in the decompressed video frame. The label value of 1 represents that there is a person falling, and 0 represents that there is no person falling.

The data set is used as the input of the CNN convolutional network, and the output is the CNN convolutional network decision function, wherein the specific construction method of the CNN convolutional network model is a well-known technology and will not be repeated here.

Decompressed video frames for real-time transmission As the input of the CNN convolutional network decision function, the output is a value of 1 or 0. When the output value is 1, it means that someone has fallen in the monitored ice and snow venue and an alarm is issued.

At this point, efficient communication of video frames at ice and snow sports venues is achieved.

It should be noted that the above sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Modifications to the technical solutions recorded in the aforementioned embodiments, or equivalent replacement of some of the technical features therein, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims (4)

1. The high-efficiency communication method for the ice and snow project site monitoring image based on deep learning is characterized by comprising the following steps of:

Collecting each video frame in the monitoring video of the ice and snow project site;

Determining the snow color highlighting and dividing weight of each sampling point based on the intensity values of each sampling point of the video frame under different channels;

acquiring optical flow vectors of adjacent video frames by adopting a sparse optical flow algorithm; determining the directional fitness of each sampling point and any optical flow vector based on the position distribution of each sampling point and any optical flow vector in the video frame;

Determining an optical flow vector non-concussion degree of any optical flow vector based on a distance distribution of the optical flow vectors in adjacent video frames; determining a snowfield motion track energy function value of each sampling point of a video frame based on the snowfield color salient segmentation weight, the optical flow vector non-concussion degree and the direction fitness of the continuous reference frame at the same sampling point;

Randomly generating vector points in a video frame, adopting an LGB algorithm in a VQ-LGB algorithm to find the moving position of the centroid in the iteration process of the vector points based on the snowfield motion race track energy function value in any iteration process of the video frame, and determining the snowfield information density splitting weight of the vector points;

Optimizing a VQ-LGB compression algorithm based on the snow information density splitting weight to obtain a compressed video frame, and inputting the compressed video frame into a neural network to realize efficient communication of ice and snow project sites;

The determining the snowfield motion race track energy function value of each sampling point of the video frame based on the snowfield color salient segmentation weight, the optical flow vector non-concussion degree and the direction fitness degree of the continuous reference frame at the same sampling point comprises the following steps:

Recording the preset number of video frames adjacent to any video frame as reference frames of any video frame;

For any optical flow vector in the reference frame, calculating the product of the length of any optical flow vector and the snow color salient division weight of each sampling point in the reference frame as a molecule;

Calculating three products of the non-concussion degree of the optical flow vector of any optical flow vector, the directional fitness of each sampling point in the reference frame and any optical flow vector, and the Euclidean distance between each sampling point in the reference frame and any optical flow vector, and taking the sum of the three products and a preset parameter adjusting factor as a denominator;

Calculating the ratio of the numerator to the denominator; taking the sum of the ratio values of all the optical flow vectors in all the reference frames as a snow movement track energy function value of each sampling point of any video frame;

the method for determining the snow information density splitting weight of the vector point based on the snow motion track energy function value and the moving position of the centroid of the vector point in the iterative process comprises the following steps:

for any vector point before any iteration process, determining a traditional vector point splitting coefficient of any vector point based on any vector point and all relevant sampling points thereof; the associated sampling points are sampling points in the area where any vector point is located;

For each movement of any vector point in the iterative process, calculating the average value of the snow motion track energy function values of all the associated sampling points of any vector point before each movement, and recording the average value as the vector point energy value before each movement of any vector point; forming energy function change vectors of any vector point by all the vector point energy values before movement;

The mean value of the first-order reciprocal vector of the energy function change vector is recorded as the energy change trend of the snow sports track at any vector point;

Determining centroid movement position weights of each movement of any vector point based on Euclidean distances between each movement position of any vector point in the iterative process;

determining the snow information density splitting weight of any vector point based on the centroid moving position weight, the snow movement track energy variation tendency and the traditional vector point splitting coefficient;

the determining the traditional vector point splitting coefficient of any vector point based on any vector point and all the associated sampling points comprises the following steps:

the standard deviation of gray values of all relevant sampling points of any vector point is recorded as the distortion degree of any vector point; taking the product of the number of the associated sampling points of any vector point and the distortion degree as a traditional vector point splitting coefficient of any vector point;

The centroid movement position weight of each movement of any vector point is determined based on the Euclidean distance between each movement position of any vector point in the iterative process, and the expression is as follows:

In the method, in the process of the invention, Is the centroid movement position weight of the g-th movement of the h-th vector point,Is the position before the g-th movement of the h-th vector point,Is the position of the h vector point after the G-th shift,Is thatAnd (3) withThe euclidean distance between the two,Is the maximum value of the Euclidean distance of all the moving times of the h vector point;

The determining the snow information density splitting weight of any vector point based on the centroid moving position weight, the snow movement track energy change trend and the traditional vector point splitting coefficient comprises the following steps:

calculating the product of the vector point energy value before each movement of any vector point and the centroid movement position weight of each movement of any vector point, and calculating the sum of the products under all movement times of any vector point;

Obtaining the maximum value in the snow movement track energy change trend of all vector points, and calculating the sum value of the maximum value and the snow movement track energy change trend of any vector point;

Taking the product of the two sum values and the traditional vector point splitting coefficient of any vector point as the snow information density splitting weight of any vector point;

The method for optimizing the VQ-LGB compression algorithm based on the snow information density splitting weight to obtain a compressed video frame, inputting the compressed video frame into a neural network to realize the efficient communication of the ice and snow project site comprises the following steps:

Replacing the traditional vector point splitting coefficient with the snowfield information density splitting weight of any vector point, and improving the VQ-LGB compression algorithm to obtain a compressed video frame;

Decompressing the compressed video frame in a data processing center to obtain a decompressed video frame; manually setting a label value for the decompressed video frame, wherein the label value is 1, which indicates that a person falls down, and 0 indicates that no person falls down;

inputting all decompressed video frames and corresponding tag values into a convolutional network, and outputting a CNN convolutional network decision function;

Taking the decompressed video frames transmitted in real time as the input of a CNN convolutional network decision function, and outputting a tag value; when the output tag value is 1, an alarm is sent out; otherwise, no alarm is issued.

2. The method for efficiently communicating the monitoring image of the ice and snow project field based on the deep learning according to claim 1, wherein the determining the snow color highlighting and dividing weight of each sampling point based on the intensity values of each sampling point of the video frame under different channels is expressed as follows:

In the method, in the process of the invention, The snow color highlighting segmentation weight for the mth sample point,、The m-th sample point is the intensity value in the dark channel and bright channel video frames,、、The intensity values of the mth sample point in the R, G, B channels are respectively.

3. The method for efficient communication of ice and snow project site monitoring images based on deep learning as claimed in claim 1, wherein determining the directional fitness of each sampling point and any optical flow vector based on the position distribution of each sampling point and any optical flow vector in the video frame comprises: connecting each sampling point with the midpoint of any optical flow vector, and taking the included angle between the connecting line and any optical flow vector as the direction coincidence degree of each sampling point and any optical flow vector.

4. The method for efficient communication of ice and snow project site monitor images based on deep learning as set forth in claim 1, wherein determining the optical flow vector non-concussion degree of any one of the optical flow vectors based on the distance distribution of the optical flow vectors in the adjacent video frames comprises:

for any optical flow vector in each video frame, marking the optical flow vector of the midpoint closest to the midpoint of any optical flow vector in the video frame of the adjacent previous frame as the associated optical flow vector of any optical flow vector;

the standard deviation of the lengths of all associated optical flow vectors of any optical flow vector is taken as the optical flow vector non-concussion of any optical flow vector.