Application of adaptive moving object detection and edge calculation in police physical education training app course

Nowadays, the theoretical knowledge and computer skills of digital signal processing are increasing year by year, and the mode of combining moving target detection industry and edge computing gradually occupies an important position in the research process of processing and pattern recognition in various application fields. For moving target detection and edge algorithm, this paper proposes a technical method based on Gaussian mixture background model and filtering, which combines moving target detection and edge calculation. This method changes the learning rate mechanism of the traditional weight in the background model updating, so that the background model can be continuously effective. Then, the motion information and centroid position information of the detected moving target are taken as the initial input data of the filter, so that the filter can start tracking the target. Finally, by analyzing and understanding the development history of Chinese police sports, analyzing the current situation of police sports teaching and training in Chinese public security colleges, we find the problems existing in police sports teaching and training in public security colleges, and use target detection and edge algorithm. Carry out actual calculation in the police sports teaching app, adjust the curriculum objective system of police sports teaching, update the teaching content, reform the teaching methods, reposition the police sports teaching curriculum, build a high-quality teaching team, fully improve the theoretical system of police sports quality education in public security colleges, and cultivate a large number of compound and applied public security talents with strong comprehensive quality and ability.


Introduction
The precondition of moving target detection is to recognize the moving target in the moving target sequence in the video, so as to obtain the target location and information. The work of moving target detection is to capture and track multiple moving targets in a video in real-time, generate motion tracks, and implement preset actions according to the automatic control theory (Lv et al. 2018). The detection of moving objects as well as various forms of life such as travel and surveillance is being examined (He et al. 2010). The development of moving target detection technology also promotes the development of all walks of life. In the process of analyzing the changes of moving targets, the detection of moving targets is divided into rigid target tracking and non rigid target tracking (Wang et al. 2011). For rigid targets, there are only translational and rotational motions, and non rigid targets are more diverse in shape changes. Moving target tracking is through the matching of information points, such as the matching of target points and inflection points, or direct matching in the form of template motion detection (Wen et al. 2021). Templates are also divided into color templates and texture templates. At present, the most widely used non rigid body tracking algorithm includes active contour model and model matching (Gedik and Alatan 2013).
Police physical education is originally a form of sports training teaching, which also includes law, fighting, and other aspects of knowledge, and is one of the disciplines with high requirements. The development level of police physical education affects the combat effectiveness of the & Lifang Zhen 2008029@muc.edu.cn police force at this stage. As the cradle of professional talents graduated from public security colleges, its vocational skills have a special nature. Police physical education and training in public security colleges have gradually attracted everyone's attention. However, in the past few years, there have been many deficiencies. The ''Arrest Tactics,'' ''police physical fitness,'' ''military sports,'' and ''police tactics'' have replaced the teaching and training forms of police sports. They only pay attention to the police arrest tactics and put the basic level of practical training and the theoretical research of police physical fitness training in the first place. This round not only hinders the exploration of the development of police physical education, but also is very disadvantageous to the exploration of the law of this teaching and training. Therefore, strengthening the construction of police physical education teaching and training discipline is of great importance to the development of police physical education teaching industry in the future.

Related work
The literature skillfully combines HS and LK algorithms, uses the improved filtering method and the advantages of the two algorithms to deal with the interference of other objects in the scene, and achieves good detection results (Zhang et al. 2020). According to the Horn and Schunck optical flow algorithm of gradient, an improved optical flow method of Gaussian pyramid is proposed in the literature. This method combines the moving target segmentation with the maximum interclass variance and mathematical morphology operation and carries out HS calculation, which effectively reduces the number of iterations of optical flow calculation, improves the detection efficiency of the algorithm, and basically meets the requirements of real-time (Peng et al. 2019;Lou et al. 2019). Aiming at the problem of slow convergence of Gaussian mixture model, an improved algorithm is proposed to improve the convergence rate without affecting the stability of the model. This method is realized by replacing the global static holding factor with an adaptive learning rate, which is calculated by Gaussian mixture. This method can achieve rapid convergence, but the effect of moving target detection needs to be improved (Yu and Sapiro 2011). The algorithm can adaptively select the number of Gaussian models, which reduces the calculation time and improves the segmentation accuracy. However, the artificial introduction of negative prior coefficients will lead to the unstable weight update of Gaussian mixture, which will affect the accuracy of moving target detection (Yan and Tang 2019). Aiming at the problem that the classical Gaussian mixture algorithm is too sensitive to non-stationary scenes, an optimization method of Gaussian mixture model is proposed in the literature. It optimizes the matching Gaussian mode selection, model updating, and background display of the detection algorithm flow (Zobay 2014). In the process of detection, the mode weight and the matching degree of the mode itself are comprehensively considered, and the matching Gaussian mode can overcome the background disturbance and reduce the false detection rate (Kang and Doh 2020). However, the detection results are still disturbed by shadows. Documents suggest that on the eve of the founding of new China, the military Commission and the Central Military Commission established public security departments (Zhang et al. 2017). In the program of joint study, it is stipulated that our country should establish unified forces, and our troops are the firm strength of the people (Gill 2020). The mission of the army is to protect our country, and to safeguard the revolutionary construction and all legitimate rights and interests of the Chinese people.'' According to the literature, when the first national public security conference was held, Zhou Enlai pointed out that ''the safety of the country is half that of the public security department, and you bear half the responsibility for the safety of the country (Heinecken 2014). The army is prepared but not used, and you need it every day.'' This fully shows that the party and the government have certain differences in the nature and tasks of the people's public security and the people's army. At that time, the people's police force of new China had just been established (Kai and Xiaojun 2014). In addition to the land reform, bandit suppression, and other work tasks, most of the people's police force was deployed by the cadres of the people's Liberation Army. Therefore, the education and training of public security personnel by the people's public security organs during this period was basically carried out in accordance with the contents and requirements of the military sports training of the people's Liberation Army (Grossheim 2018).
3 Design of adaptive moving target detection algorithm

Adaptive moving target detection
Adaptive moving target detection has good computing effect, fast speed, simple content, and easier to understand. Even so, in the in-depth exploration of the algorithm, adaptive moving target detection still has the following two defects: Adaptive moving target detection is a preset target in the detection process, but in the actual process, it is impossible to keep the size completely unchanged, which leads to the error of target information and the failure of target detection; adaptive moving target detection cannot improve the occlusion of the target in the tracking process, which will also lead to the failure of target detection.

Adaptive moving target detection scale
The traditional adaptive moving target detection algorithm is a fixed form. When the size changes, the target area cannot match the size completely, so it is not conducive to positioning. In order to achieve the goal of stable target detection, the background information of the scale module is added, and the error of the background information is reduced by detecting the change of the scale of the target detection information.
3.1.1.1 Rough estimation of scale Under the continuous adjustment of moving target detection, the scaled moving target and the initial target size are sorted and scaled; the hog feature is extracted and operated on the moving target, and then, the hog feature is detected by correlation filtering, from which the maximum value of detection is deleted, and the maximum size obtained is the optimal size. The parameter value of the set initial size is R 0 , the maximized parameter value is R H , and the minimized parameter value is R L . Under the filtering method, the size of the adjusted size is set to L 1 9 W 1 .
3.1.1.2 Accurate estimation of scale Rough scale can keep the speed of the algorithm maximized and has little impact on the scale in the short-term effect, but the perception of scale changes is weak in the long-term, so it affects the final quality of effect collection, resulting in the failure of target tracking. Rough estimation has a great impact on the subsequent detection of target size. In the long run, the calculation will lead to the failure of target detection. Therefore, using accurate estimation can maximize the balance between accuracy and speed, so as to meet the effect of target detection. Finally, unify the sizes of all moving targets, extract the hog features, and use the scale module to screen the maximum value. The best size that affects the maximum value is: In formula (1), q 0 is the maximum value filtered after scale filtering calculation. Using the known data, the target template and coefficient are as follows: where g is the learning rate. As shown in Fig. 1, using a combination of rough estimation and accurate estimation, first mark the approximate target size with rough estimation to obtain the initial size L 1 9 W 1 and then, use the edge calculation method to get the accurate estimated size, which cannot only grasp the change of size, but also have a great impact on the tracking effect.

Principle of target occlusion judgment
As shown in Fig. 2, four stages of occlusion process under the same image are listed: According to the characteristics of relevant edge calculation results, when the target is occluded, the maximum influence value is calculated by using the edge. When the occluded area is small, there may be some special cases. According to various situations, the accuracy can be evaluated according to the matched target under the edge calculation, so as to judge whether the target is occluded. The calculation formula is shown in (3): In this paper, the occlusion of adaptive moving target detection algorithm is judged by edge calculation. Suppose a is the edge calculation process, PX is the current edge calculation value, M is the average value between two adjacent values under the edge calculation, t is a certain threshold, and the value of T is subject to the actual situation. When the target occlusion is large, increase the threshold T; When the target occlusion is small, reduce the threshold t. According to formula (4), the edge calculation process can be obtained.

Design of moving target denoising algorithm
With the continuous progress of modern science, sports goals are closely related to human daily life, involving more and more industries and a wide range of applications. Moving target information transmission is one of the effective ways, but in the process, due to the noise impact of its own and other external environment, the effective target information is greatly reduced. Therefore, how to improve the accuracy of moving targets is one of the obstacles of target processing technology. The two denoising methods are spatial filtering and frequency domain filtering.

Spatial filtering
Spatial domain refers to the horizontal plane on the moving target, and the processing of spatial domain is to directly operate the elements on the moving target. Spatial filtering is used to filter moving targets in the spatial domain, which is also one of the methods used in moving target processing.
Smooth linear spatial filter The smooth linear space filter replaces the value of the element points on the moving target with the value corresponding to the filter template. This filter is called ''mean filter''. Mean filter is widely used in reducing the signal noise of moving targets. Different mean filters have different principles and mechanisms to deal with mean, which can be divided into four modes: Statistical sorting filter Statistical sorting filter is a kind of ''nonlinear spatial filter''. Its basic principle is to sort pixels in the moving target area around the filter and replace the statistical sorting value obtained. The core idea of this method is to set the result value of sorting as the response value of filter. Therefore, the sorting results and values are different. Statistical sorting filters can be divided into three forms: median filter, minimum filter, and maximum filter, among which median filter is the most widely used. It uses the median value of the pixel field to replace the value of the pixel, which is expressed as follows:

Frequency domain filtering
Low-frequency filtering in Butterworth form is Butterworth filtering in n stages at point D0 in Butterworth form, and its functional formula is as follows: Gauss low-pass filter: The functional expression of Gaussian low-pass filter is: 3.3 Target detection algorithm design

Algorithm structure
The target detection algorithm designs a checkbox generation network, which generates huge candidate areas, including background information and quality problems. The boxes obtained are not all used as subsequent network input, because they are numerous and most of them do not contain targets. Therefore, the target detection algorithm is to compare the intersection and union ratio (IOU) between the real box and the candidate box of the target and then, compare the IOU value of each box with the set threshold to screen the desired samples. S 1 and S 2 represent the target frame area and candidate frame area, then: In Eq. (12), N cls represents the number of candidate frames calculated by classification regression after equalization; N reg indicates the number of anchor frames; i represents the anchor box index. If the anchor box is a positive sample, the anchor box corresponds to the foreground background classification result of the prediction box; Because there is a large gap between N cls and N reg in value, it is set as the balance weight value; t i represents the position vector of the prediction frame; Each component is calculated by the following formula: The classification loss function L cls is calculated as follows: The location regression loss function L reg is calculated as follows: 4 Edge computing deployment

Overall framework design
The edge computing architecture in video detection is shown in Fig. 3. It can be seen from Fig. 3 that the computing card is connected with the camera and can only receive and process single line data, which is to reduce consumption and ensure the timeliness of video transmission. If each video card is connected to multiple cameras, transmission data will be greatly reduced, affecting its time-lapse data.
First of all, in terms of time-lapse data, the signal process of video legend includes encoding and caching. If the transmission calculation is carried out over a long distance, a high delay will be formed when the video arrives at the central processing server, which will have a great impact on the timeliness and accuracy of all aspects. After adding edge computing, the video processing server moves to the edge node, which can greatly speed up the propagation of data and reduce the delay effect. Secondly, in terms of server bearing data, if a single server is used for processing, the bearing capacity is huge, so the server computing capacity is required to be very high. However, after the introduction of edge computing, the server is only responsible for receiving the transmission data, which reduces the additional computing, thus reducing the computing burden. Finally, in terms of system performance, because the computing power of the central server has been fixed, there is no other expansion ability in computing more processing system data, and it cannot adapt to the changes of the new environment. After the introduction of edge computing, the server does not need to calculate redundant tasks, so it does not need to expand, and its performance has been greatly improved in all aspects.

Time consumption performance analysis
First of all, in the server-based centralized processing scheme, the time required from video data acquisition to early warning response is shown in equation, which requires more processing processes and time consumption.
where T0 represents the total time consumption of video data transmission and processing, t server camera represents the time required for video data collected by the camera to be transmitted to the server, t decoding represents the time required for decoding after the server receives video data, and t detect_ 0 refers to the total time required for the server to detect foreign objects in multi-channel video data. After edge calculation is introduced, the time required from video data acquisition to early warning response is shown in Eq. 18. The processing tasks on the server are greatly reduced, and the time delay is small.
Compare the time consumption t0 and T1 of video detection under the server-based centralized processing scheme and the edge computing deployment scheme, the difference between the two is shown in Eq. (19).

Power performance analysis
The power consumption required to complete a detection in the server-based centralized processing scheme is as follows (20): Among them, W 0 represents video data acquisition, and w camera represents the power consumption of video data acquisition, because the server carries too many decoding tasks. The processing capacity is huge. After introducing edge computing, the power consumption required to complete a detection in video detection is as follows (21): Similarly, by comparing the power consumption W0 and W1 of video detection before and after the introduction of edge computing, it can be found that the part of w camera contains both, and the difference between the two is shown in the following formula: Therefore, merge the two moving targets in the same space and then, remove the seams to merge the two moving targets into one.

Detailed description of moving target stitching and fusion algorithm
For moving target stitching, the most basic stitching method is selected. The algorithm steps are as follows: (a) According to the confidence and error rate of the data, the number of samples to be selected is obtained. (b) Randomly select four pairs of matching feature points.
(c) Select two pairs of matching points and use the following formula to calculate the characteristic value t of the matching point pair: Where L X is the gradient in the X direction, L y is the gradient in the y direction, and L xy is the product of the gradient in the xy direction. When merging three or more continuous moving targets, the formula is as follows: For moving target fusion, weighted average fusion algorithm is used to eliminate the stitching trace. Taking two moving targets as an example, if the two moving targets are represented by I1 (i, j) and I2 (I, j), respectively, and the fused moving target is represented by I (i,j), then: Iði; jÞ ¼ aI 1 ði; jÞ þ ð1 À aÞI 2 ði; jÞ ð 25Þ

Curriculum design demand analysis
At present, the teaching and training of Police Physical Education in public security colleges basically follows the teaching and training mode of ordinary physical education in ordinary colleges. This teaching mode of weekly classes for the purpose of improving students' physique and health level obviously cannot meet the needs of professional skill training of police physical education. The purpose of training is to continuously improve students' movement foundation and realize the improvement of skill level by continuously strengthening training. As a teaching module, police physical education is a compulsory course in public security colleges and a subject that every student must complete during school. However, with the continuous progress and development of society, a single discipline cannot meet the daily needs. The elective contents involved in the police physical education curriculum were collected through questionnaires, interviews, and visits. The survey is shown in Table 1.

System course module design
The teaching of police physical education is the core content of physical education, which is essential in the process of cultivating students' all-round development. The teaching management under the current network app technology mode can be improved to improve the teaching quality. The management of this teaching module is divided into two parts, as shown in Fig. 4. Through the establishment of these two parts, it can provide good help for the teaching resources of the Police Physical Education Institute. The teaching management module is a management mode that introduces the teaching management system and teaching information into the curriculum. In this mode, the students' grades at each stage are analyzed through the results of physical education stage tests and examinations. The information interaction module allows students to evaluate and supervise the police physical education curriculum. Under the situation of this module, the management of the teaching department can be supervised. Through the acquisition of police physical education teaching courses, students can realize the autonomous learning mode without regional and time constraints. At the same time, they can also cultivate knowledge related to theory and physical and mental health and make up for the vacancy of theoretical knowledge.
5 Application effect analysis of police physical education teaching and training app course

Research object
The subjects of the experiment were 20 male police members aged 24-28 in a municipal police team, and 10 people in each group were divided into experimental group and control group. Conduct a questionnaire survey and discuss to determine that the technical and tactical contents should include five items: defense control, obstacle climbing, police weapons, water safety, and police tactics. See Table 2 for the comparison results of the two groups and Table 3 for the comparison results of technical and tactical results. After statistical analysis, p [ 0.05 shows that there is no significant difference between the experimental group and   the control group before the experiment, which meets the experimental research conditions. The specific test contents and scoring standards are as follows.

Experimental design
The experimental group used the police physical education teaching and training app designed by the combination of target detection and edge calculation for training, and the control group used traditional training methods for training. During the eight-week period from April 2022 to June 2022, the experimental group and the control group were trained. The routine training contents of the two groups in the morning and evening were consistent. During the daytime training, the experimental group was trained with the training contents formulated by app, and the control group was trained with traditional exercises. After the eight-week training, the two groups were tested.

Index analysis
À Speed quality It can be seen from Table 4 that there is a significant gap in the test data of police physical education reflected by the two groups of data, in which the experimental group is higher than the control group. According to the data in Table 5, the results of the experimental group are higher than those of the control group in the comparison of p-values of defense control technology and climbing obstacle climbing technology, and the results of the two groups are more obvious than those before defense control technology and climbing obstacle climbing technology.

Evaluation of police physical education teaching and training app curriculum design
At present, the positioning of physical education teaching in the police physical education institute is very unclear, and it is confused with the subject teaching in ordinary schools, which not only limits the development of the Police Physical Education Institute, but also limits the future development history of the Physical Education Institute. Due to the particularity of police college education, usually, the police sports management department itself does not understand the internal rules and regulations of the school, so there are many performance deviations in police sports teaching and training. The investigation results of the significance of police physical education teaching and curriculum arrangement show that the current significance of police physical education college teaching is very vague, the curriculum arrangement is very unreasonable, and there are many problems to be understood (Table 6).

Conclusion
Moving target detection is a kind of computer tracking target detection. In real life, due to many objective factors, it brings many problems to the subsequent movement detection form. the movement states in the image are to distinguish the moving and static background in the background state, eliminate the irrelevant moving target references, and carry out automatic machine processing and screening. It is not only a compound the computer workload, which increases the accuracy and efficiency. At this stage, the teaching mode used in the teaching and training of police physical education is the teaching mode on the physical education curriculum of ordinary schools, which is mainly based on improving students' physical quality, and cannot adapt to the daily teaching of professional police physical education. Through the popularization and research of the app of the Police Physical Education Institute, we can fundamentally solve the problem, increase the difficulty of training in the police physical education teaching, and pass on skills and knowledge to students through video analysis and standardized action essentials, so as to continuously improve the training level. Therefore, in the police physical education teaching, we should continue to increase the training intensity and difficulty, not only based on the usual basic training, but also implement practical training, such as increasing elective courses, which plays a role in promoting the quality improvement of students in the police physical education institute from multiple perspectives.
Funding The authors have not disclosed any funding.
Data availability Data will be made available on request.

Declarations
Conflict of interest The authors declare that they have no conflict of interests.
Ethical approval This article does not contain any studies with human participants performed by any of the authors.