Safety monitoring method and system for hotel

Abstract

The invention provides a safety monitoring method and system for a hotel. The method comprises the following steps: acquiring a hotel image; constructing a security monitoring model; a safety monitoring model is used for recognizing the crowd gathering condition and the body movement; when the safety monitoring model identifies an abnormal condition, information transmission is carried out on the abnormal condition to corresponding equipment, so that measures can be taken in time, and the safety of guests is ensured. According to the safety monitoring method and system for the hotel provided by the invention, whether crowds are gathered or not can be quickly identified, when the crowd density is greater than a preset value, the crowds are considered to be gathered, the information needs to be pre-warned and reported, and at the moment, the gathered crowds are subjected to body movement identification, so that the crowd density is improved. When the preset dangerous limb movement is identified, the information that security personnel need to intervene when the security risk exists is sent to the related personnel, and the video within a period of time is sent to the related personnel, so that the related personnel can intervene and stop in time.

Description

Technical Field

[0001] The invention belongs to the technical field of hotel security, and more specifically, relates to a security monitoring method and system for a hotel. Background Art

[0002] With the rapid development of the hotel industry, there are more and more hotels and travelers, and safety management is becoming more and more important. As a densely populated area, in some specific environments, it is easy for guests to have quarrels with each other or between guests and hotel staff, and even escalate to physical conflicts. Hotel safety management is the most important part of hotel operations.

[0003] At present, most hotel security management systems adopt a monitoring room-centered model, which is not conducive to dynamic management. Human monitoring cannot provide timely warnings. In addition, the means of data processing are relatively backward, the data processing efficiency is low, and the hotel security management efficiency is low. Summary of the invention

[0004] The purpose of the present invention is to provide a security monitoring method and system for a hotel, aiming to solve the technical problems of the monitoring room-centered model in the prior art, in which manual monitoring cannot provide timely warnings and the hotel security management efficiency is low.

[0005] To achieve the above object, the technical solution adopted by the present invention is: to provide a security monitoring method for a hotel, the following steps:

[0006] Collect hotel images;

[0007] Build a security monitoring model; use the security monitoring model to identify crowd gatherings and body movements;

[0008] When the security monitoring model identifies an abnormal situation, it transmits information about the abnormal situation to the corresponding equipment so that timely measures can be taken to ensure the safety of the guests.

[0009] Preferably, the construction of the security monitoring model comprises the following steps:

[0010] Preprocess the image;

[0011] Identify the gathering of guests;

[0012] When the crowd gathering meets the preset requirements, human body movements are identified to determine whether there are safety issues.

[0013] Preferably, the identifying of the gathering of guests comprises the following steps:

[0014] Identify the number of guests in the image;

[0015] According to the actual area of ​​the scene in the picture and the recognition result of the number of guests, the gathering of guests is recognized and calculated;

[0016] In this step, the identification calculation formula for the aggregation situation is:

[0017] Where M is the density of guests, in units of per square meter, j is the correction coefficient; S1 is the area of ​​a single guest object in the image; S2 is the actual area of ​​the guest; S3 is the total area in the image; and N is the number of guests in the image.

[0018] Preferably, when the crowd gathering situation meets the preset requirements, identifying human body movements includes the following steps:

[0019] Determine the input image;

[0020] The image is input into the aggregation recognition module for recognition; the aggregation recognition module includes: after the first convolution structure processing, the formula is:

[0021] X r =conv 3D (X);

[0022] Among them, X r Represents the features after reducing the number of channels; the dimensional features are [N,C / r,T,W,H]; r is the ratio of reducing the number of channels; conv 3D Indicates that a 1×1×1 convolution kernel is used to reduce the dimension of the feature channel; X represents the input feature, and the feature dimension is [N, C, T, W, H].

[0023] Preferably, the aggregation identification module comprises the following steps:

[0024] For the reduced eigenvector X r First, the channel is changed, and then the motion information is represented by calculating the time difference between the features after channel transformation and the adjacent features;

[0025] M(t)=conv(X r (t+1))-X r (t);

[0026] Where M(t) represents the motion feature vector at time t; the dimensional feature is [N, C / r, W, H]; the value range of t is (1, T-1); conv represents a 3D convolution layer with a 3×3×3 convolution kernel; X r (t+1)) transform each channel of the feature;

[0027] Through the global average pooling layer, the channel-level weight relationship is obtained by using the full connection operation and activation function, and finally the weight and original feature information are fused by element multiplication method;

[0028] Q=Res(Sigmoid(QL(P(QL(Res(PC(M)))))));

[0029] Among them, Q represents the action correlation weight of each channel in the feature vector; the dimension feature is [N, C, T, 1, 1]; PC represents the global average spatial pooling of the feature vector; Res represents the matrix transformation of the feature vector; QL represents the fully connected layer; P represents the ReLU activation function;

[0030] By using residual connection, motion information is added to the original feature information;

[0031]

[0032] Among them, X m represents the output of the motion perception module; Represents matrix multiplication.

[0033] Preferably, the aggregation identification module comprises the following steps:

[0034] Perform a global average pooling layer on the input image;

[0035] X q = PC(X); PC represents global average spatial pooling of feature vectors; X represents input feature vectors; feature dimensions are [N, C, T, W, H]; X q Represents the feature vector after spatial pooling; the feature dimension is [N,C,T,1,1];

[0036] V=Sigmoid(conv1(P(conv1(X,K,J),1,C)));

[0037] Where V represents the importance map of learning; conv1 represents temporal convolution; the parameters are the input tensor, kernel size, and number of output channels; P represents the activation function; K represents the size of the convolution kernel; C represents the number of channels of the input feature; J represents the number of channels after the first conv1; J <C;

[0038]

[0039] Z represents the enhanced feature map; Represents matrix multiplication.

[0040] Preferably, the aggregation identification module comprises the following steps:

[0041] Connect the YD module and SJ module in parallel:

[0042] Among them, X represents the input of the residual block, with the dimension [N, C, T, W, H]; YD represents the motion perception module; SJ represents the temporal attention module; Indicates element addition; X L Represents the output of the first residual convolutional neural module; dimension is [N,C,T,W,H].

[0043] Preferably, the aggregation identification module comprises the following steps:

[0044] The input is downsampled;

[0045] Get output, output output X s for:

[0046] X s =X3+X s .

[0047] The present invention also provides a security monitoring system for a hotel, which is used to execute a security monitoring method for a hotel as described in any one of the above items.

[0048] Preferably, it includes:

[0049] A collection module, used to collect hotel images;

[0050] Model module, used to build a security monitoring model; use the security monitoring model to identify crowd gathering and body movements;

[0051] The identification module is used to transmit information about abnormal situations to corresponding devices when the security monitoring model identifies abnormal situations, so that timely measures can be taken to ensure the safety of guests.

[0052] The beneficial effect of a security monitoring method and system for hotels provided by the present invention is that: compared with the prior art, the security monitoring method and system for hotels provided by the present invention can quickly identify whether a crowd is gathered. When the crowd density is greater than a preset value, it can be considered that the crowd is gathered and is prone to security risks, and it is necessary to report the information with an early warning, and at this time, the gathered crowd will also be identified for physical movements. When a preset dangerous physical movement is identified, a message "there is a security risk and security personnel need to intervene" will be sent to relevant personnel, and a video within a period of time will be sent to relevant personnel so that the relevant personnel can intervene in time to stop it. In hotel activities, there are many situations where people need to gather together, so the security monitoring method and system for hotels can effectively improve the recognition accuracy of "security risk events (such as group fights, fighting incidents, etc.)" by first identifying the gathering of people and then identifying physical movements in two stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, 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 invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 A structural block diagram of a security monitoring method for a hotel provided by an embodiment of the present invention;

[0055] Figure 2 The present invention provides a structural block diagram of a security monitoring system for a hotel according to an embodiment of the present invention.

[0056] In the figure: 1. Acquisition module; 2. Model module; 3. Recognition module. DETAILED DESCRIPTION

[0057] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0058] Please also read Figure 1 and Figure 2 Now, a security monitoring method for a hotel provided by the present invention is described. The security monitoring method for a hotel comprises the following steps:

[0059] S1, collect hotel images;

[0060] In this step, cameras are installed in the hotel's corridors, corners and other areas to achieve all-round monitoring of the hotel.

[0061] S2. Build a security monitoring model; use the security monitoring model to identify crowd gathering and body movements.

[0062] The security monitoring model includes a crowd gathering recognition module and a human motion recognition module;

[0063] The implementation process of step S2 can be carried out through the following steps:

[0064] S2.1. Preprocess the collected image; specifically, perform data enhancement processing on the collected image. Use image segmentation algorithm to process the data and remove the background.

[0065] In this step, an image of any frame of the surveillance video of the interval acquisition camera is used as the acquired image.

[0066] S2.2. Identify the gathering of guests; specifically, use a gathering identification module to identify the gathering of guests.

[0067] The implementation of this step includes the following processes:

[0068] S2.2.1. Identify the number of guests in the image;

[0069] The operation process of the cluster recognition module is as follows: image input - feature extraction of the image - use of the perceptual structure to obtain the global feature vector and the local feature map - connect the global feature vector and the local feature map to obtain the perceptual feature vector - output the prediction result based on the perceptual feature vector.

[0070] Specifically, a feature extractor of the VGG16 network structure is used to extract features from the input. The feature extractor can effectively extract low-level and high-level features of the image. The perception structure includes a global context unit and a local context unit. The global context unit passes through a global pooling layer to obtain a global feature vector. The global feature vector includes the size and shape of the image. The local context unit passes through a convolution layer and a maximum pooling layer to obtain a local feature map. The local feature map includes the distribution and density of the crowd. The global feature vector and the local feature map are respectively reduced in dimension by two fully connected layers to obtain two low-dimensional feature vectors, and the two low-dimensional feature vectors are connected to obtain a context-aware feature vector. The context-aware feature vector is mapped to the prediction result of the number of guests through a fully connected layer.

[0071] The preprocessed images are input into the cluster recognition module for recognition to obtain the predicted result of the number of guests.

[0072] S2.2.2. According to the actual area of ​​the scene in the picture and the recognition result of the number of guests, the gathering situation of the guests is identified and calculated;

[0073] In this step, the identification calculation formula for the aggregation situation is:

[0074] Where M is the density of guests, in units of per square meter, j is the correction coefficient; S1 is the area of ​​a single guest object in the image; S2 is the actual area of ​​the guest; S3 is the total area in the image; and N is the number of guests in the image.

[0075] In this step, when the value of M is greater than the preset value, the system enters the warning state and sends the image to the receiving device of the relevant personnel. At the same time, the video of a certain time range containing the image is intercepted and sent to the receiving device of the relevant personnel. At the same time, the video is transferred to the next step for action recognition.

[0076] In this step, when the value of M is greater than the preset value, the captured image is sent to the receiving device of the relevant personnel. At the same time, a video of 5-10 seconds after the captured image is captured and sent to the receiving device of the relevant personnel, and the video is transferred to the next step for action recognition.

[0077] S2.3. When the crowd gathering meets the preset requirements, human body movements are identified to determine whether there are safety issues.

[0078] The implementation of this step includes the following processes:

[0079] S2.3.1. Determine an input image; specifically, the input image is: when the value of M is greater than a preset value, a video of 5-10 seconds after the captured image.

[0080] S2.3.2, input the image into the cluster recognition module for recognition;

[0081] In this step, the operation process of the aggregation recognition module is: processed by the first convolution structure - processed by the maximum pooling layer - processed by the first residual convolution neural module - processed by the second residual convolution neural module - processed by the third residual convolution neural module - enter the average pooling layer processing - enter the fully connected layer structure processing, and identify the work.

[0082] In this step, after the first convolution structure processing, the first convolution structure is a 3D convolution layer structure. The processing of the first convolution structure can improve the efficiency of the entire operation process. The formula is, X r =conv 3D (X);

[0083] Among them, X r Represents the features after reducing the number of channels; the dimensional features are [N,C / r,T,W,H]; r is the ratio of reducing the number of channels; conv 3D Indicates that a 1×1×1 convolution kernel is used to reduce the dimension of the feature channel; X represents the input feature, and the feature dimension is [N, C, T, W, H].

[0084] In this step, the output feature vector of feature vector X after the maximum pooling layer is X M .

[0085] In this step, the structures of the first residual convolution neural module, the second residual convolution neural module and the third residual convolution neural module are the same. The specific process of the first residual convolution neural module processing includes YD module processing, and the YD module processing includes the following steps:

[0086] For the reduced eigenvector X r First, the channel is changed, and then the motion information is represented by calculating the time difference between the features after channel transformation and the adjacent features;

[0087] M(t)=conv(X r (t+1))-X r (t);

[0088] Where M(t) represents the motion feature vector at time t; the dimensional feature is [N, C / r, W, H]; the value range of t is (1, T-1); conv represents a 3D convolution layer with a 3×3×3 convolution kernel; X r (t+1)) transforms each channel of the feature.

[0089] Through the global average pooling layer, the channel-level weight relationship is obtained using the fully connected operation and activation function, and finally the weight and original feature information are fused using the element-wise multiplication method.

[0090] Q=Res(Sigmoid(QL(P(QL(Res(PC(M)))))));

[0091] Among them, Q represents the action correlation weight of each channel in the feature vector; the dimension feature is [N,C,T,1,1]; PC represents the global average spatial pooling of the feature vector; Res represents the matrix transformation of the feature vector; QL represents the fully connected layer; and P represents the ReLU activation function.

[0092] By using residual connection, motion information is added to the original feature information;

[0093]

[0094] Among them, X m represents the output of the motion perception module; Represents matrix multiplication.

[0095] The specific process of the first residual convolution neural module processing also includes SJ module processing, and the SJ module processing includes the following steps:

[0096] Perform a global average pooling layer on the input image;

[0097] X q = PC(X); PC represents global average spatial pooling of feature vectors; X represents input feature vectors; feature dimensions are [N, C, T, W, H]; X q Represents the feature vector after spatial pooling; the feature dimension is [N,C,T,1,1].

[0098] V=Sigmoid(conv1(P(conv1(X,K,J),1,C)));

[0099] Where V represents the learned importance map; conv1 represents temporal convolution. The parameters are the input tensor, kernel size, and number of output channels. P represents the activation function. K represents the size of the convolution kernel. C represents the number of channels of the input feature. J represents the number of channels after the first conv1. <C。

[0100]

[0101] Z represents the enhanced feature map. Represents matrix multiplication.

[0102] The specific process of the first residual convolution neural module processing also includes connecting the YD module and the SJ module in parallel:

[0103]

[0104] Among them, X represents the input of the residual block, with the dimension [N, C, T, W, H]; YD represents the motion perception module; SJ represents the temporal attention module; Indicates element addition; X L Represents the output of the first residual convolutional neural module. The dimension is [N,C,T,W,H].

[0105] In this step, in the second residual convolution neural module, the input is X L The steps performed are the same as those of the first residual convolution neural module. The output of the second residual convolution neural module is X L2 In the third residual convolution neural module, the input is X L2The steps performed are the same as those of the first residual convolution neural module. The output of the third residual convolution neural module is X L3 .

[0106] The specific process of the first residual convolution neural module also includes converting X L3 The input is convolved in the residual block, and the output is X3. The residual block includes 3 convolutional layers.

[0107] The specific process of the first residual convolution neural module processing also includes downsampling the input to obtain X s ;

[0108] The specific process of the first residual convolution neural module processing also includes obtaining the output, output X s for:

[0109] X s =X3+X s .

[0110] S3. When the security monitoring model identifies an abnormal situation, it transmits information about the abnormal situation to the corresponding equipment so that timely measures can be taken to ensure the safety of the guests.

[0111] In this step, abnormal actions are identified.

[0112] When the security monitoring model detects body movements in the presence of an abnormal crowd, it will send warning information to the receiving terminal so that relevant personnel can take appropriate measures in a timely manner.

[0113] The present invention also provides a security monitoring system for hotels, please refer to Figure 1 and Figure 2 , used to execute a security monitoring method for a hotel as described in any one of the above. Specifically, it includes: an acquisition module 1, a model module 2 and a recognition module 3, wherein the acquisition module 1 is used to acquire hotel images; the model module 2 is used to build a security monitoring model; the security monitoring model is used to identify crowd gathering and body movements; the recognition module 3 is used to transmit information about the abnormal situation to the corresponding device when the security monitoring model identifies an abnormal situation, so as to take timely measures to ensure the safety of the guests.

[0114] The present invention provides a security monitoring method and system for hotels. Compared with the prior art, the present invention can quickly identify whether a crowd is gathered. When the crowd density is greater than a preset value, it can be considered that the crowd is gathered, and the information needs to be reported for early warning. At this time, the gathered crowd will also be identified for physical movements. When a preset dangerous physical movement is identified, the information "there is a security risk and security personnel need to intervene" will be sent to relevant personnel, and the video within a period of time will be sent to the relevant personnel so that the relevant personnel can intervene in time to stop it. In hotel activities, there are many situations where people need to gather together. Therefore, the security monitoring method and system for hotels can identify "security risk events (such as group fights, fighting incidents, etc.)" by first identifying the gathering of people and then identifying physical movements in two stages. It can effectively improve the recognition accuracy of "security risk events (such as group fights, fighting incidents, etc.)", so that relevant personnel can quickly take stop measures and effectively improve the efficiency of hotel safety management.

[0115] The above description is only a preferred embodiment of the present invention and does not limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A security monitoring method for a hotel, characterized in that: The following steps are involved: Collect hotel images; Build a security monitoring model; use the security monitoring model to identify crowd gathering and body movements in the hotel; When the security monitoring model identifies an abnormal situation, it transmits information about the abnormal situation to the corresponding equipment so that timely measures can be taken to ensure the safety of the guests.

2. A security monitoring method for a hotel as claimed in claim 1, characterized in that: The construction of the security monitoring model includes the following steps: Preprocess the image; Identify the gathering of guests; When the crowd gathering meets the preset requirements, human body movements are identified to determine whether there are safety issues.

3. A security monitoring method for a hotel as claimed in claim 2, characterized in that: The identification of the gathering of guests includes the following steps: Identify the number of guests in the image; According to the actual area of ​​the scene in the picture and the recognition result of the number of guests, the gathering of guests is recognized and calculated; In this step, the identification calculation formula for the aggregation situation is: Where M is the density of guests, in units of per square meter, j is the correction coefficient; S1 is the area of ​​a single guest object in the image; S2 is the actual area of ​​the guest; S3 is the total area in the image; and N is the number of guests in the image.

4. A security monitoring method for a hotel as claimed in claim 3, characterized in that: When the crowd gathering situation meets the preset requirements, the human body motion recognition includes the following steps: Determine the input image; The image is input into the aggregation recognition module for recognition; the aggregation recognition module includes: after the first convolution structure processing, the formula is: X r =conv 3D (X); Among them, X r Represents the features after reducing the number of channels; the dimensional features are [N,C / r,T,W,H]; r is the ratio of reducing the number of channels; conv 3D Indicates that a 1×1×1 convolution kernel is used to reduce the dimension of the feature channel; X represents the input feature, and the feature dimension is [N, C, T, W, H].

5. A security monitoring method for a hotel as claimed in claim 4, characterized in that: The aggregation identification module includes the following steps: For the reduced eigenvector X r First, the channel is changed, and then the motion information is represented by calculating the time difference between the features after channel transformation and the adjacent features; M(t)=conv(X r (t+1))-X r (t); Where M(t) represents the motion feature vector at time t; the dimensional feature is [N, C / r, W, H]; the value range of t is (1, T-1); conv represents a 3D convolution layer with a 3×3×3 convolution kernel; X r (t+1)) transform each channel of the feature; Through the global average pooling layer, the channel-level weight relationship is obtained by using the full connection operation and activation function, and finally the weight and original feature information are fused by element multiplication method; Q=Res(Sigmoid(QL(P(QL(Res(PC(M))))))); Among them, Q represents the action correlation weight of each channel in the feature vector; the dimension feature is [N, C, T, 1, 1]; PC represents the global average spatial pooling of the feature vector; Res represents the matrix transformation of the feature vector; QL represents the fully connected layer; P represents the ReLU activation function; By using residual connection, motion information is added to the original feature information; Among them, X m represents the output of the motion perception module; Represents matrix multiplication.

6. A security monitoring method for a hotel as claimed in claim 5, characterized in that: The aggregation identification module includes the following steps: Perform a global average pooling layer on the input image; X q = PC(X); PC represents global average spatial pooling of feature vectors; X represents input feature vectors; feature dimensions are [N, C, T, W, H]; X q Represents the feature vector after spatial pooling; the feature dimension is N,C,T,1,1]; V=Sigmoid(conv1(P(conv1(X,K,J),1,C))); Where V represents the importance map of learning; conv1 represents temporal convolution; the parameters are the input tensor, kernel size, and number of output channels; P represents the activation function; K represents the size of the convolution kernel; C represents the number of channels of the input feature; J represents the number of channels after the first conv1; J <C; Z represents the enhanced feature map; Represents matrix multiplication.

7. A security monitoring method for a hotel as claimed in claim 6, characterized in that: The aggregation identification module includes the following steps: Connect the YD module and SJ module in parallel: Among them, X represents the input of the residual block, with the dimension [N, C, T, W, H]; YD represents the motion perception module; SJ represents the temporal attention module; Indicates element addition; X L Represents the output of the first residual convolutional neural module; the dimension is [N, C, T, W, H].

8. A security monitoring method for a hotel as claimed in claim 7, characterized in that: The aggregation identification module includes the following steps: The input is downsampled; Get output, output output X a for: X s =X3+X s 。 9. A security monitoring system for a hotel, characterized in that: Used to execute a security monitoring method for a hotel as described in any one of claims 1 to 8.

10. A security monitoring system for a hotel as claimed in claim 9, characterized in that: include: A collection module, used for collecting hotel images; Model module, used to build security monitoring models; Use security monitoring models to identify crowd gatherings and body movements; The identification module is used to transmit information about abnormal situations to corresponding devices when the security monitoring model identifies abnormal situations, so that timely measures can be taken to ensure the safety of guests.

Images