WO2022247712A1

WO2022247712A1 - Security warning method and apparatus based on target tracking in microdistrict monitoring video

Info

Publication number: WO2022247712A1
Application number: PCT/CN2022/093648
Authority: WO
Inventors: 秦军瑞; 吴劲; 李启文; 段志奎; 邝伟锋; 许剑锋; 邓锐; 李洋
Original assignee: 广州智慧城市发展研究院
Priority date: 2021-05-24
Filing date: 2022-05-18
Publication date: 2022-12-01
Also published as: CN113365026A; CN113365026B

Abstract

The present application provides a security warning method and apparatus based on target tracking in a microdistrict monitoring video. Video sequence processing is performed on a microdistrict monitoring video, a reading time and a reading position of the video sequence is calculated, probability deviations and route continuity values between a monitoring target and residents in the microdistrict are obtained, a community classification value of the monitoring target belonging to the microdistrict in the video sequence is calculated for comparison with a threshold, and security warning information is sent to a microdistrict manager's mobile device. The present application has the following advantages: (1) information of both reading times and reading positions is added to target video tracking to effectively determine comprehensive time and space information of the video; (2) real-time monitoring is performed for an external person in the microdistrict by means of the probability value of the monitoring target belonging to the microdistrict in the video sequence; (3) by comparison with the probability threshold, security warning notification information is sent to the microdistrict manager's mobile device in real time.

Description

A method and device for community monitoring video target tracking security early warning

technical field

The present disclosure relates to the technical field of target tracking, and in particular to a security warning method and device for target tracking of community monitoring video.

Background technique

The current target tracking technology mainly uses deep learning algorithms to screen and locate monitoring targets only in surveillance videos. Deep learning is a technology widely used in computer vision processing of current videos and images. It decomposes videos and images into image frames. Frame matrix, operate on the frame matrix to obtain the recognition result. However, in the case of using video tracking technology for community security management, simply computing the video is not enough to measure the relevance of the monitoring personnel to the community, and the time and location of the monitoring target are also important technical considerations. Therefore, the current target tracking technology is not enough to calculate the impact of the monitoring target's video, appearance time, and appearance location on the security of the community, and it is difficult to meet the technical requirements of the community security early warning.

Contents of the invention

The present disclosure provides a security warning method and device for community monitoring video target tracking. By processing the video sequence of the community monitoring video and calculating the reading time and reading position of the video sequence, the probability difference between the monitoring target and the residents of the community is obtained. The route coherence value is used to calculate the community classification value of the monitoring target belonging to the community in the video sequence for threshold judgment, and the safety warning information is sent to the mobile device of the community management personnel.

In order to achieve the above object, according to one aspect of the present disclosure, a method and device for security warning of community monitoring video target tracking are provided, the method includes the following steps:

Step 1, add the processing of video reading time and reading position to the monitoring video read in the community, form a data set, and obtain a data set of video sequences;

Step 2, by calculating whether the video sequence containing the monitoring target and the video sequence in the data set can be marked as the same target, the probability bias gap between each video sequence is obtained;

Step 3, by calculating the reading time and reading position of the video sequence of the monitoring target and other video sequences in the data set, the route coherence value between different sequences is obtained;

Step 4, calculate and obtain the community classification value of the monitoring target belonging to the community in the video sequence through the probability deviation gap and the route coherence value;

Step 5: Judging by the community classification value whether the monitoring target in the video sequence is an outsider who does not belong to the community, and if not, send a preset security warning notification message to the administrator's mobile device.

Further, in step 1, the monitoring video read in the community is processed by adding the video reading time and reading position to form a data set and obtain a data set of video sequences. The specific method is: from the monitoring camera or by using Read the surveillance video of the residents in the community in the database that stores the video data captured by the surveillance camera, add the video reading time and reading position to the surveillance video read in the community, and decompose the surveillance video into image frames , and each image frame is marked with its reading time and reading position, thus forming a dataset of video sequences of community residents.

Further, in step 2, by calculating whether the video sequence containing the monitoring target and the video sequence in the data set can be labeled as the same target, the method to obtain the probability deviation gap between each video sequence is: obtained in step 1 In the video sequence data set of the residents of the residential area, the newly read outsiders are used as the monitoring target, and the video sequence of the monitoring target and all the video sequences in the data set are calculated, and the video sequence of the monitoring target is set as the monitoring sequence, Find the probability deviation gap between the monitoring sequence and the sequence in the data set, and calculate whether the monitoring target can be marked as the same target as the target in the data set, so as to measure the probability difference between the target contained in the sequence and the targets in other sequences. deviation.

Further, in step 3, by calculating the reading time and reading position of the video sequence of the monitoring target and other video sequences in the data set, the method of obtaining the route coherence value between different sequences is: calculating the Each reading position forms a positioning route, and the route coherence value between the current positioning in the positioning route and the positioning routes of all sequences in the data set is calculated to measure the degree of fit between the monitoring sequence and the sequences in the data set on the positioning route.

Further, in step 4, the method of calculating and obtaining the community classification value of the monitoring target belonging to the community in the video sequence is as follows: using the probability deviation difference and route coherence value to compare the monitoring sequence in The deviation in the target probability and the degree of fit on the positioning route are calculated to obtain the probability value of the monitoring target belonging to the community in the video sequence as the community classification value, and the community classification value and the preset probability threshold will be conditioned Determine whether the threshold is exceeded.

Further, in step 4, it also includes setting the newly read outsider as the monitoring target x, and its video sequence is S _x , calculating the community classification value of each sequence in x and data values and taking its arithmetic mean As the target x and the community classification value of the dataset.

Further, in step 5, it is judged by the community classification value whether the monitoring target in the video sequence is an outsider who does not belong to the community, and if not, a preset security warning notification is sent to the administrator’s mobile device information, the method is as follows: in step 4, the community classification value and the probability threshold are judged, and if the threshold is exceeded, the server sends safety warning information to the mobile device of the community manager.

A community monitoring video target tracking security warning device includes: a processor, a memory, and a computer program stored in the memory and running on the processor, when the processor executes the computer program, the one In the step in the method of community surveillance video target tracking security early warning, the device for community surveillance video target tracking security early warning runs on a computing device in a desktop computer, a notebook computer, a palmtop computer or a cloud data center.

The beneficial effects of the present disclosure are: the present disclosure provides a community monitoring video target tracking security warning method and device, by processing the video sequence of the community surveillance video, and calculating the reading time and reading position of the video sequence, the monitoring target is obtained The difference between the probability deviation and the route coherence value of the residents in the community, and the community classification value of the monitoring target belonging to the community in the video sequence are obtained for threshold judgment, and the security warning information is sent to the mobile device of the community management personnel, which has the following advantages: (1) The information of reading time and reading position is added to the target video tracking at the same time to effectively measure the comprehensive time and space information of the video; (2) through the probability value of the monitoring target belonging to the community in the video sequence, real-time monitoring is carried out by outsiders in the community; 3) By judging the probability threshold, the security warning notification information is sent to the cell manager's mobile device in real time.

Description of drawings

The above and other features of the present disclosure will be more apparent through a detailed description of the embodiments shown in the drawings. The same reference numerals in the drawings of the present disclosure represent the same or similar elements. Obviously, the appended The drawings are only some embodiments of the present disclosure. For ordinary technical goals in this field, other drawings can also be obtained according to these drawings without creative work. In the drawings:

Fig. 1 shows the flow chart of a kind of community monitoring video target tracking security warning method and device;

Figure 2 shows the calculation flow chart of the probability deviation gap;

Figure 3 shows the calculation flow chart of the community classification value.

Detailed ways

The concept, specific structure and technical effects of the present disclosure will be clearly and completely described below in conjunction with the embodiments and drawings, so as to fully understand the purpose, scheme and effect of the present disclosure. It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

Figure 1 is a flow chart of a community monitoring video target tracking security early warning method and device according to the present disclosure, and a community monitoring video target tracking security early warning method and device according to an embodiment of the present disclosure will be described below in conjunction with Figure 1 device.

The present disclosure proposes a community monitoring video target tracking security warning method and device, which specifically includes the following steps:

Step 5, judging by the community classification value whether the monitoring target in the video sequence is an outsider who does not belong to the community, and if not, send a preset security warning notification message to the administrator's mobile device.

Further, in step 1, the monitoring video read in the community is processed by adding the video reading time and reading position to form a data set and obtain a data set of video sequences. The specific method is: from the monitoring camera or by using Read the monitoring video of the residents in the community in the database that stores the video data captured by the monitoring camera, and use the MOT algorithm for the monitoring target video read in the community (references: Ciaparrone G, FL Sánchez, Tabik S, et al .Deep Learning in Video Multi-Object Tracking: A Survey[J].Neurocomputing,2019,381.) or using the SORT algorithm (references: Bewley A, Ge Z, Ott L, et al.Simple Online and Realtime Tracking[ C] 2016 IEEE International Conference on Image Processing (ICIP). IEEE, 2016.) Filter and preprocess the monitoring target video and output the preprocessed frame slice containing the monitoring target, and then add the video reading time and reading position Processing, the reading position is the preset positioning position of the camera or the GPS positioning latitude and longitude position, decomposing the monitoring video into image frames, and marking each image frame with its reading time and reading position, each video The sequence is composed of all decomposed frames and the reading time and reading position corresponding to each frame. Let the video sequence be S, and each frame and its reading time and reading position are recorded as s as a data item. The video sequence S is composed of multiple data items s, and each video sequence S has a target (the target is the identity of the community residents identified in the sequence in the preprocessing), and all the video sequences of the community residents are combined into a video sequence of the community residents The data set T.

Further, in step 2, by calculating whether the video sequence containing the monitoring target and the video sequence in the data set can be marked as the same target, the method to obtain the probability deviation gap between each video sequence is: obtained in step 1 In the video sequence data set of residents in the residential area, calculate the newly read video sequence of outsiders (that is, the video sequence containing the monitoring target) and all the video sequences in the data set, and set the video sequence containing the monitoring target as the monitoring sequence , to find the probability deviation gap between the monitoring sequence and the sequences in the data set, so as to measure the deviation in the probability of the targets contained in the sequence and the targets in other sequences, specifically:

Take the image frame in each data item s in the sequence S, convert the image frame into a 512×512 image frame array f, let f(m,n) be the value of the mth row and nth column of the array f (m , n are all positive integers less than or equal to 512);

Let n be the length of the sequence S, and assume that the value range of the serial number i of the data item s belongs to [1, n], then the image frame matrix of the image frame in the i-th data item s _i in S is f _i , the i-th The reading time corresponding to each data item takes year, month, day, hour, minute, and second as a 6-dimensional array and is represented as d _i , and the corresponding reading position positioning in the ith data item takes longitude o _i and latitude a _i is a 2-dimensional array expressed as l _i =[o _i , a _i ], then the i-th data item _{s i} in S is expressed mathematically as s _i =[f _i , d _i , l _i ];

Let the function K(f,d,l) extract the feature value k _i of the image information, time information and location information in the data item s,

Among them, R(d,l) is the processing process of d _i and l _i , let d(m,n), l(m,n) be the value of the mth row and nth column of the array d _i , l _i , the specific calculation R(d,l)=[exp(d(1,1)*l(1,1))+exp(d(1,1)*l(1,2)]+[exp(d(1, 2)*l(1,1))+exp(d(1,2)*l(1,2)]+[exp(d(1,3)*l(1,1))+exp(d( 1,3)*l(1,2)]+[exp(d(1,4)*l(1,1))+exp(d(1,4)*l(1,2)]+[exp (d(1,5)*l(1,1))+exp(d(1,5)*l(1,2)]+[exp(d(1,6)*l(1,1)) +exp(d(1,6)*l(1,2)], that is, s _i =[f _i , d _i , l _i ] through the function K(s _i )=K(f _i , d _i , l _i ) can get k _i =K(s _i ), then the feature value set of the image information, time information and location information of the video sequence S can be expressed as K(S)={k ₁ ,k ₂ ,...,k _{i- 1} , k _i }, the calculation methods of other eigenvalues are the same;

Assuming different targets a and b in T, process any two videos corresponding to a and b into image frame sequences S _a and S _b marked with reading time and place, and calculate the probability of two different sequences S _a and S _b Deviation gap, to judge the probability deviation D _ab of the target in two videos, the specific calculation is: K(S _a )={k _a,1 ,k _a,2 ,…,k _a,i-1 ,k _{a, i} },K(S _b )＝{k _b,1 ,k _b,2 ,…,k _b,i-1 ,k _b,i },

k _{a, i} and k _{b, i} represent the eigenvalues of data item number i in S _a and S _b respectively.

Further, in step 3, by calculating the reading time and reading position of the video sequence of the monitoring target and other video sequences in the data set, the route coherence value between different sequences is obtained. The method is: calculate the Each reading position forms a positioning route, and calculates the route coherence value of the current positioning of the monitoring sequence in the positioning route and the positioning routes of all sequences in the data set to measure the degree of fit between the monitoring sequence and the sequences in the data set on the positioning route, specifically The method is as follows: According to the characteristic value of the detection target from the current reading position l _i to the next reading position positioning l _i+1 , to obtain its route coherence value, use the ith data item in the sequence S described in step 2 The longitude and dimension of the corresponding reading position in , is a 2-dimensional array l _i , which is connected according to the target’s moving trajectory {l ₁ ,l ₂ ,…,l _i-1 ,l _i } to form a route L _i , and the target is taken The next reading position of locating l _i+1 , calculate the route coherence value of both reading positions of l _i and l _i+1

To measure the matching degree of the connection between the two locations on the route.

Further, in step 4, the method of calculating and obtaining the community classification value of the monitoring target belonging to the community in the video sequence is as follows: using the probability deviation difference and route coherence value to compare the monitoring sequence in The deviation in the target probability and the degree of fit on the positioning route are calculated to obtain the probability value of the monitoring target belonging to the community in the video sequence as the community classification value, and the community classification value and the preset probability threshold will be conditioned Determine whether the threshold is exceeded, specifically:

Step 4.1, take the different sequences S _a and S _{b of any two targets a and b} in the set T and pass the probability deviation D _ab of the two sequences, and position l _i and position l _i+ of each reading position in each route L _i The route coherence value C ^i,i+1 of ₁ is calculated to get {C ^1,2 ,C ^2,3 ,…,C ^i-1,i ,C ^i,i+1 };

Step 4.2, set the community classification value between target a and b as β, then

Indicates that when targets a and b have displacement actions from location l _i to location l _i+1 in route L _i , the community classification value between the two targets is calculated as

Step 4.3: Loop through the set T to extract two sequences, repeat steps 4.1 to 4.2 to obtain and record the classification values of each community, until the complete traversal of the set T, take the arithmetic mean of the community classification values of all records The value is μ, which is used as the probability threshold of the community classification value;

Step 4.4, set the newly read outsider as the monitoring target x, and its video sequence as S _x , calculate the community classification values of x and each sequence in the set T, and take the arithmetic mean as the community of the target x and the set T The classification value λ.

Further, in step 5, it is judged by the community classification value whether the monitoring target in the video sequence is an outsider who does not belong to the community, and if not, a preset security warning notification is sent to the administrator’s mobile device information, the specific method is: judge the community classification value λ and the probability threshold μ in step 4, if it is lower than the threshold value, it means that the community classification value of the monitoring target x and the community is lower than normal, and it is an abnormal situation. The server sends safety warning information to the mobile devices of the community managers.

Said a kind of community monitoring video target tracking security early warning device comprises: processor, memory and the computer program that is stored in said memory and run on said processor, when said processor executes said computer program, realizes said A step in a community monitoring video target tracking security early warning method, wherein the community monitoring video target tracking security early warning device runs on a computing device in a desktop computer, a notebook computer, a palmtop computer or a cloud data center.

The present disclosure provides a security warning method and device for community monitoring video target tracking. By processing the video sequence of the community monitoring video and calculating the reading time and reading position of the video sequence, the probability difference between the monitoring target and the residents of the community is obtained. Based on the coherence value with the route, the community classification value of the monitoring target belonging to the community in the video sequence is obtained for threshold judgment, and the safety warning information is sent to the mobile device of the community management personnel. In the target video tracking, the information of reading time and reading position is added at the same time to effectively measure the comprehensive time and space information of the video; through the probability value of the monitored target belonging to the community in the video sequence, real-time monitoring is carried out by outsiders in the community; by judging the probability Threshold, real-time security warning notification information is sent to the mobile devices of the community managers.

While the description of the present disclosure has been presented with considerable detail and in particular has described a few described embodiments, it is not intended to be limited to any such details or embodiments or to any particular embodiment, effectively encompassing the intended scope of the present disclosure. Furthermore, the disclosure has been described above in terms of embodiments foreseeable by the inventors for the purpose of providing a useful description, and insubstantial modifications of the disclosure which are not presently foreseeable may still represent equivalent modifications of the disclosure.

Claims

A community surveillance video target tracking security warning method, characterized in that the method comprises the following steps:

Step 1, add the processing of video reading time and reading position to the monitoring video read in the community, form a data set, and obtain a data set of video sequences;

Step 2, by calculating whether the video sequence containing the monitoring target and the video sequence in the data set can be marked as the same target, the probability bias gap between each video sequence is obtained;

Step 3, by calculating the reading time and reading position of the video sequence of the monitoring target and other video sequences in the data set, the route coherence value between different sequences is obtained;

Step 4, calculate and obtain the community classification value of the monitoring target belonging to the community in the video sequence through the probability deviation gap and the route coherence value;

Step 5, judging by the community classification value whether the monitoring target in the video sequence is an outsider who does not belong to the community, and if not, send a preset security warning notification message to the administrator's mobile device.
A kind of community monitoring video target tracking security warning method according to claim 1, it is characterized in that, in step 1, to the monitoring video read in the community, carry out the processing of adding video reading time and reading position, form Data set, the method of obtaining the data set of the video sequence is: read the monitoring video of the residents in the community from the monitoring camera or the database used to store the video data taken by the monitoring camera, decompose the monitoring video into image frames, and divide each Each image frame is marked with its reading time and reading position, thus forming a data set of video sequences of residents in the community.
According to claim 1, a community monitoring video target tracking security early warning method is characterized in that, in step 2, by performing calculations on whether the video sequence containing the monitoring target and the video sequence in the data set can be marked as the same target , the method to obtain the probability deviation gap between each video sequence is: in the video sequence data set obtained in step 1, take the newly read outsider as the monitoring target, and combine the monitoring target video sequence with all the video sequences in the data set Calculate the sequence, set the video sequence of the monitoring target as the monitoring sequence, find the probability deviation gap between the monitoring sequence and the sequence in the data set, calculate whether the monitoring target can be marked as the same target with the target in the data set, to measure the deviation in probability of targets contained in this sequence from targets in other sequences.
A kind of community monitoring video target tracking security warning method according to claim 1, it is characterized in that, in step 3, by the video sequence of monitoring target and the reading time of other video sequences in the data set, the reading position is calculated , the method to obtain the route coherence value between different sequences is: calculate each reading position in the monitoring sequence to form a positioning route, calculate the route coherence value between the current position in the positioning route and the positioning routes of all sequences in the data set, to measure the The degree of fit between the monitoring sequence and the sequence in the data set on the positioning route.
According to claim 1, a kind of community monitoring video target tracking security warning method, it is characterized in that, in step 4, through the probability deviation difference and route coherence value, calculate and obtain the community belonging to the monitoring target in the video sequence. The method of the class value is: compare the deviation of the monitoring sequence on the target probability and the degree of fit on the positioning route through the probability deviation gap and the route coherence value, and calculate the probability value of the monitoring target belonging to the community in the video sequence as the community Classification value, the community classification value and the preset probability threshold will be used to judge whether the threshold is exceeded.
A kind of community monitoring video target tracking security warning method according to claim 1, it is characterized in that, in step 4, also comprise, the outsider that reads newly is set as monitoring target x, and its video sequence is S x , Calculate the community classification value of each sequence in x and data values and take the arithmetic mean as the community classification value of the target x and data set.
A kind of community monitoring video target tracking security warning method according to claim 1, it is characterized in that, in step 5, judge whether the monitoring target in this video sequence is the outsider that does not belong to this community by community classification value, If it does not belong to, the preset security warning notification information is sent to the administrator's mobile device. The method is: in step 4, the community classification value and the probability threshold are judged. The mobile device sends a safety warning message.
A community monitoring video target tracking security early warning device is characterized in that the community monitoring video target tracking security early warning device includes: a processor, a memory, and a computer stored in the memory and running on the processor program, when the processor executes the computer program, it realizes the steps in a method for tracking security warnings of community surveillance video targets in claim 1, and the device for tracking security security warnings of community surveillance video targets runs on a desktop computer , laptop, PDA, or computing device in a cloud data center.