Intelligent monitoring method for hotel monitoring

Abstract

The invention belongs to the technical field of data analysis and intelligent storage, and provides an intelligent monitoring method for hotel monitoring, which specifically comprises the following steps of: identifying monitoring equipment at each floor in a hotel, acquiring original video data from the monitoring equipment at equal time intervals, constructing a floor passenger flow time sequence through the original video data, and storing the floor passenger flow time sequence into a database; and finally, building a passenger flow mode function in cooperation with the passenger flow time sequence of each floor to obtain a passenger flow mode transition value, and performing intelligent storage control on the monitoring data according to the passenger flow mode transition value. Through real-time quantification of the occurrence degree of the cross-floor passenger flow guiding abnormity in the hotel, the monitoring system can more intelligently judge whether the video data in the corresponding time period has the storage value obviously higher than that in the daily time period, and the high-quality evidence obtaining requirement of a key event is ensured. The huge consumption risk of the storage space and the network bandwidth caused by the fact that high-quality recording is adopted for all time periods only with personnel targets is avoided.

Description

Technical Field

[0001] This invention belongs to the fields of data analysis and intelligent storage technology, and specifically relates to an intelligent monitoring method for hotel surveillance. Background Technology

[0002] In hotel management, surveillance is crucial for operations, cost control, and security strategies. Hotel surveillance cameras often have a large number of cameras and record continuously, resulting in a huge amount of recorded data. Current technologies typically use anomaly detection or human behavior recognition methods to record video segments with storage value at high quality, while the rest are recorded at low quality. This difference in quality is usually reflected in frame rate and resolution.

[0003] However, existing anomaly detection or human behavior recognition methods mostly focus on static image features or behavioral patterns of a single target in a single camera feed. For example, they detect fighting, running, or falling based on a single frame or a local time period, or trigger high-quality recording only when a localized crowd gathers in a certain feed. These methods generally lack modeling and attention to the correlation structure of passenger flow changes over time across multiple floors. They cannot determine which time periods are significant in the overall floor structure based on passenger flow characteristics between floors. This results in a very high threshold for cameras to detect anomalies, prolonged periods of low-quality recording, and insufficient coverage of high-quality recording periods. This fails to fully utilize floor traffic information for more refined screening. Furthermore, the inability to construct cross-floor correlations of passenger flow intensity changes over time makes it impossible to identify abnormal passenger flow behavior. This further increases the risk of missing too much valuable data in hotel security risk assessment and operational anomaly analysis. Therefore, there is an urgent need for an intelligent hotel monitoring method based on multi-floor collaborative monitoring of abnormal passenger flow behavior. Summary of the Invention

[0004] The purpose of this invention is to propose an intelligent monitoring method for hotel surveillance, in order to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create conditions.

[0005] To achieve the above objectives, according to one aspect of the present invention, a smart monitoring method for hotel surveillance is provided, the method comprising the following steps: S100 identifies surveillance equipment on each floor of the hotel. S200 obtains raw video data from the monitoring equipment at equal time intervals; S300 constructs a floor passenger flow time series using raw video data; S400, coordinates the passenger flow time series of each floor to construct the passenger flow pattern function and obtains the passenger flow pattern transition value; S500 intelligently stores and controls monitoring data based on passenger flow pattern transition values; Furthermore, in step S100, the method for identifying monitoring devices on each floor of the hotel is as follows: monitoring devices are deployed on each floor of the hotel, and each monitoring device is associated with corresponding floor identification information to distinguish monitoring devices on different floors. The monitoring equipment is a camera; The preferred monitoring equipment is a fixed-installation network camera, but it can also be an integrated access control machine with video recording function, a building intercom terminal, or other electronic devices with video recording capabilities.

[0006] Each monitoring device is assigned a unique device identifier and its floor information is recorded. This is to enable the raw video data subsequently collected by the monitoring device to be categorized and statistically analyzed according to the floor level. The deployment and identification association methods of the monitoring devices described above are standard practices for those skilled in the art and will not be elaborated upon here.

[0007] Further, in step S200, the method for obtaining raw video data from the monitoring device at equal time intervals is as follows: The preset time interval is 5-15 minutes. Each time interval is a recording point. Each recording point collects monitoring video data once from each monitoring device as raw video data. The raw video data is then stored after being identified by the monitoring device and time information.

[0008] The storage of raw video data here refers to initial storage, not terminal storage. Raw video data is high-quality video data directly acquired, with frame rate, resolution, and other specifications at the highest level available to the detection equipment, or the preset highest specifications. Initial storage refers to the original specifications of any video data, while terminal storage refers to video data after selective data compression. The video data is in the form of a monitoring video stream or video clips. The preset time interval determines the granularity of anomaly detection and intelligent village management in subsequent video data. Too small a granularity can easily distort the passenger flow pattern transition values, while too large a granularity can lead to excessive anomaly coverage and overfitting problems.

[0009] Furthermore, in step S300, the method for constructing the floor passenger flow time series using the original video data is as follows: for any recording point, pedestrian detection is performed on the original video data obtained through the monitoring equipment, the number of passengers in the corresponding field of view is statistically obtained, and the number of passengers corresponding to each monitoring equipment is accumulated by floor to obtain the passenger flow statistics for each floor. A time period is preset as the backtest duration RL, with the value of the backtest duration ranging from 3 to 6 hours. The time period from the current time to its reverse time direction RL is taken as the current backtest segment. For any floor, the passenger flow statistics of each recording point in the current backtest segment are used to form a sequence, which is recorded as the floor passenger flow time series.

[0010] The method for detecting pedestrians and counting the number of people in the corresponding field of view from the raw video data obtained through monitoring equipment is as follows: pedestrian detection is implemented using existing target detection algorithms, including pedestrian detection models based on convolutional neural networks, with YOLO algorithm being preferred, Faster R-CNN detection network algorithm, and moving target detection algorithms based on background modeling and foreground segmentation. Those skilled in the art can choose any mature pedestrian detection scheme according to the system performance requirements.

[0011] To avoid double counting of the same person, preferably, the detected person targets are further associated using a multi-target tracking algorithm. This can be done using a conventional tracking framework based on Kalman filtering and Hungarian matching, or using an existing DeepSORT-type multi-target tracking algorithm to associate the trajectories of the same person target appearing across frames within the same time interval.

[0012] At least one virtual counting line or counting area is pre-set in the field of view of the monitoring equipment. When any pedestrian trajectory crosses the virtual counting line or enters the counting area for the first time in the current time interval, the passenger flow count value corresponding to the pedestrian trajectory is incremented by one, and the same pedestrian trajectory is not counted repeatedly in the time interval. Thus, the number of passengers in the field of view of the monitoring equipment in the current time interval is obtained.

[0013] The principle of choosing passenger flow as the input variable for cross-floor passenger flow orientation anomaly monitoring is that the entry and exit behavior of hotel guests leaves observable trajectory segments in the monitoring screens of different floors, including the time to enter the elevator from the lobby, the time to arrive at a certain floor, the behavior of staying in the corridor and entering and leaving the room, etc. Based on a large-scale passenger flow and a long-term reference scale, guests often move between floors through limited vertical transportation channels, and behaviors such as check-in, check-out, and peak elevator use have obvious regularities. Therefore, the intensity changes of passenger flow on each floor in the time dimension and the correlation between floors will show a relatively stable statistical pattern, and this statistical pattern can be used to point to cross-floor passenger flow orientation anomalies.

[0014] Further, in step S400, the method for constructing a passenger flow pattern function and obtaining the passenger flow pattern transition value by coordinating the passenger flow time series of each floor is as follows: For each recording point, construct a passenger flow time series matrix Xt of dimension n×m for each floor, where n represents the total number of floors and m is the sequence length; perform eigenvalue decomposition on the covariance matrix of Xt to obtain the corresponding eigenvalues ​​and eigenvector matrices; extract the eigenvectors corresponding to the first k largest eigenvalues ​​to construct a projection matrix and denote it as the passenger flow projection matrix P. k The global covariant matrix Zt is obtained through linear isomorphic mapping. =Xt·P k Each row of the matrix corresponds to the eigenvector zt of a single floor; the eigenvalue decomposition process of the covariance matrix is ​​obtained in Python by inputting the passenger flow time series matrix into the linalg.svd function of the NumPy library to obtain the corresponding eigenvalues ​​and eigenvector matrix; k is the preset number of eigenvalues ​​to be extracted, and its value range is k∈[1,3]. The design principle of the passenger flow projection matrix and the global covariance matrix lies in reorganizing the scattered observations of the time series corresponding to any time on each floor into a low-dimensional state that reflects the overall passenger flow structure across floors. Performing covariance eigenvalue decomposition on the matrix composed of passenger flow time series from each floor essentially utilizes the law of large numbers and the law of passenger flow conservation. That is, while the behavior of an individual guest going up and down floors is random, under the constraints of limited vertical passageways and check-in / check-out rhythms, multi-floor passenger flow will form a stable, coordinated change pattern in the long term. These normal patterns are concentrated on the first few principal eigenvectors of the covariance matrix. By constructing a passenger flow projection matrix by taking the eigenvectors corresponding to the first few largest eigenvalues ​​and calculating the global covariance matrix, it is equivalent to mapping the original high-dimensional passenger flow data to the most representative cross-floor passenger flow pattern space in history. At this time, each row of the global covariance matrix eigenvector zt represents the coordinates of that floor in the common pattern. This makes the subsequent judgment of anomalies no longer stop at whether the number of people on a certain floor has increased, but instead focus on whether the current cross-floor passenger flow structure deviates from the low-dimensional natural pattern it has formed in history. This provides a computable global state carrier for solving the structural problem of cross-floor passenger flow orientation anomalies.

[0015] Calculate the arithmetic mean of the elements in each column of the global covariance matrix and define it as the global reference vector mt; for each eigenvector zt in the global covariance matrix, calculate its Euclidean norm distance dt=‖zt-mt‖ with the global reference vector mt, and extract the maximum norm offset dmax from it; the maximum norm offset is the maximum Euclidean norm distance; input dmax into the sigmoid function for normalization processing to generate the passenger flow pattern transition value PFMTV=1 / [1+e^(-dmax)].

[0016] The global baseline vector mt is composed of the arithmetic mean of each column. The design principle of the global reference vector and maximum norm offset is to construct a reference point for the overall normal state at the current moment within the global state Zt, which has been compressed into the cross-floor passenger flow pattern subspace, and to quantify the severity of the anomaly using the floor that deviates most from this reference point. The arithmetic mean of each row of Zt along the column direction is taken to obtain mt, which reflects the current consensus state of all floors at the current recording point in the historical master pattern coordinate system. Because under natural operating conditions, passenger flow on each floor is constrained by check-in and check-out patterns and the conservation of passenger flow within limited vertical channels, the zt values ​​of multiple floors will fluctuate slightly around this consensus center. The Euclidean norm distance between the feature vector of a single floor and the global reference vector is calculated. This fits the degree to which the floor deviates from the formation in the cross-floor collaborative mode. Taking the maximum value dmax among all floor distances captures the floor that deviates most from the overall structure. For example, situations such as a middle floor suddenly becoming a passageway or multiple guest room floors increasing sharply simultaneously without the support of the lobby will cause the corresponding zt to move away from mt. This directly addresses the technical challenge that the anomaly is not the number of people themselves, but the anomaly in the cross-floor temporal relationship and flow pattern.

[0017] It should be noted that while the passenger flow pattern transition value calculation scheme based on covariance matrix eigenvalue decomposition and maximum norm offset can measure the overall deviation of the current recording point from historical cross-floor passenger flow patterns, the passenger flow time series of all floors are further compressed into a single global offset during projection and normalization. This leads to a certain degree of structural information collapse in the fine-grained transmission relationship of passenger flow between different floors and the abnormal contribution of the dominant floor, affecting the accuracy of targeted storage control and post-event source tracing analysis. To make the identification of abnormal passenger flow more precise, the interpretability of abnormal patterns stronger, and to eliminate the aforementioned structural information collapse and abnormal contribution ambiguity while ensuring that the passenger flow pattern transition value can still be used to drive intelligent storage control, this invention proposes a more interpretable and stable scheme as follows: Preferably, in step S400, the method for constructing a passenger flow pattern function and obtaining the passenger flow pattern transition value by coordinating the passenger flow time series of each floor is as follows: for any floor, the range and standard deviation of its passenger flow time series are recorded as interval drop and interval dispersion, respectively. At the same recording point, the interval drop and interval dispersion are normalized respectively, and the binary pair formed by the interval drop and interval dispersion corresponding to the floor is recorded as the interval phase group; the average value of each interval phase group of the same floor within the backtesting time is calculated to obtain the baseline phase group, and the Euclidean distance between the interval phase group of the current recording point and the baseline phase group is calculated as the baseline deviation of the current recording point. If the baseline deviation of a recording point reaches a maximum value, and the corresponding element of the recording point in the passenger flow time series is greater than the average value of the passenger flow time series, then the recording floor has experienced effective passenger flow enhancement at that recording point; the median value of the passenger flow time series is recorded as the passenger flow level PFL. The preset floor proportion RRFL is used, with a value range of RRFL∈[0.2,0.5]. The floor number of interest is obtained by rounding down the product of the total number of floors and the relevant floor proportion, and is denoted as RFL. Any floor is taken as the parsing layer, and the record points where effective passenger flow enhancement occurs are identified as the mapping start point. The mapping quality of any record point along the time direction from the mapping start point is the number of non-parsing layers where at least one effective passenger flow enhancement occurs between the two record points. The mapping interval is from the record point to its first convergence constraint point along the time direction. The convergence constraint point is the first record point along the time direction from the mapping start point that satisfies the mapping quality being greater than or equal to the number of floors of interest. Each non-parsing layer where effective passenger flow enhancement occurs within the mapping interval is the mapping object. The proportion of the mapping object appearing in all mapping intervals of the parsing layer is the coverage of interest (CAC). The ratio of the number of record points between the record point where the mapping object first experiences effective passenger flow enhancement and the mapping start point to the total number of record points in the mapping interval is the non-mapping coverage (NMC). The design goal of this mapping structure is to explore the structural response to the propagation of sudden abnormal increases in passenger flow across multiple floors. Effective passenger flow increases form a mapping interval that extends forward, triggered by anomalies in the parsing layer. The positioning of non-parsing layers is subject to the natural constraint that a large number of floors must share passenger flow through limited vertical channels. Therefore, it can effectively eliminate local noise in the interval construction. Based on this parsing, cross-floor collaborative events can abstract the collective abnormal passenger flow behavior, such as simultaneous increases in volume across multiple floors or anomalies in the middle floors becoming channels, into a computable time interval, providing a cross-floor temporal skeleton for the subsequent passenger flow pattern function.

[0018] Based on the mapping intervals of each floor, a passenger flow pattern function is constructed to obtain the passenger flow pattern transition value PFMTV: ; Where i1 and i3 are cumulative variables, i2 is the index of the mapping interval, HF is the total number of floors, PFL(i1) represents the passenger flow level of the i1th floor, PFL.max is the maximum value among all passenger flow levels, RZ(i1) is the number of mapping intervals for the i1th floor, min{} is the minimum value function, and CAC i2,i3 and NMC i2,i3 These represent the coverage rate of the 3rd mapped object in the 2nd mapping interval and the coverage rate of the non-mapped object, respectively. When any floor is considered a parse floor, all other floors are considered non-parse floors. This concept is used to divide the mapping intervals for each floor separately; the time direction is the opposite direction of the time direction.

[0019] The design principle of heart coverage and non-mapping coverage is to describe the floors that have long been widely involved in abnormal passenger flow paths within each mapped interval, and to intervene in abnormal floors earlier in time. The heart coverage in the passenger flow pattern function is used to characterize the breadth of its participation in multiple cross-floor coordinated passenger flow events, while the non-mapping coverage is used to characterize the forward position ratio of its intervention time in this interval.

[0020] The more frequently a floor appears within the mapping interval and the earlier it participates in abnormal passenger flow within the interval's time range, the greater its contribution to the transition value. Essentially, this fits the phenomenon of individual floors being repeatedly used as passageways or hidden gathering points, playing a leading or pivotal role in the formation of abnormal passenger flow. This quantifies the degree of deviation of cross-floor passenger flow distribution from the baseline and also identifies the set of floors that play a core role in these deviations. Thus, treating floors as channels or experiencing imbalances in checkout rhythms—collective anomalies—is transformed into a scalar quantification result that can be used for intelligent storage control.

[0021] Beneficial effects: Since the passenger flow pattern transition value is obtained by collaborative modeling based on passenger flow data from each floor, it can effectively quantify the degree of abnormal passenger flow orientation across floors in a multi-floor hotel environment. This provides a traceable mathematical reference variable for the next step of the monitoring system to more intelligently determine the storage value of video data in different time periods, thereby optimizing the storage performance and efficiency of the monitoring system.

[0022] Further, in step S500, the method for intelligent storage control of monitoring data based on passenger flow pattern transition values ​​is as follows: a time period is preset as a fuzzy time window VTT, the value range of the fuzzy time window is VTT∈[0.5,1] hours; a time period is preset as a distribution period DTT, the value range of the distribution period is DTT∈[7,14] natural days; the set of each passenger flow pattern transition value within the current fuzzy time window is taken as the first distribution sequence, and the first distribution sequences within the distribution period are used to form the second distribution sequence; The current fuzzy time window is the time period from the current moment to its reverse time length of VTT. Therefore, the first distribution sequence is the set of passenger flow pattern transition values ​​within the divided VTT time period. The process of constructing the second distribution sequence from the first distribution sequences within the distribution period is based on the same current moment corresponding to the time scale in each natural day. The first distribution sequences corresponding to the VTT time periods within the distribution period are located in the reverse time direction, and their union is used as the second distribution sequence. If the current passenger flow pattern transition value at the current recording point is greater than or equal to the upper quartile of the second distribution sequence and greater than or equal to the average value of the first distribution sequence, then the first passenger flow anomaly condition is met. The average value of the minimum values ​​corresponding to the current passenger flow pattern transition values ​​in the first distribution sequence is used as the fuzzy benchmark. The difference between the current passenger flow pattern transition value and the fuzzy benchmark is the trend pressure amount. The difference between the current record point and the passenger flow pattern transition value of the previous record point is the step pressure amount. The product of the trend pressure amount and the step pressure amount is the transition pressure degree. If the transition pressure degree of a record point is positive and larger than that of the previous record point, then the second passenger flow anomaly condition is met. If the current recording point simultaneously meets the first and second abnormal passenger flow conditions, then it is determined that an abnormal passenger flow has occurred; search for the maximum value of the first transition pressure in the reverse time direction from the current recording point, and store the video data between that recording point and the current recording point in high quality.

[0023] High-quality storage of high-quality video data refers to storing video data with high resolution or high frame rate, or storing raw video data without data compression. Conversely, if high-quality video data storage is not performed, the frame rate or resolution of the raw video data is reduced to reduce the storage pressure on the server or computer.

[0024] If the current recording point does not indicate an abnormal passenger flow, video data compression should not be performed immediately. This is because the current recording point may be re-identified as a recording point indicating an abnormal passenger flow within a fuzzy time window. Therefore, the original video data should only be compressed and stored after a recording point has exceeded the current fuzzy time window.

[0025] The cross-floor passenger flow anomalies referred to in this method do not merely refer to single-point anomalies such as crowd gathering, running, or fighting in a single surveillance video. Rather, they refer to a type of group anomalies where the temporal changes and flow patterns of passenger flow across different floors deviate significantly from historical baselines. These anomalies typically include, but are not limited to, a sudden surge in passenger flow across multiple guest room floors within a short period, without corresponding lobby traffic support; intermediate floors that were previously less frequented becoming statistically abnormal transit or passageways; and a passenger flow diffusion rhythm from lower to higher floors that is significantly inconsistent with normal check-in and check-out times. Hotel events implied by these abnormal passenger flow behaviors include: people not following conventional routes to go up and down floors, floors being used as passageways for visiting other rooms, the emergence of gathering points on floors, abnormal passageways between floors, and check-out rhythms deviating from historical patterns. Because their anomalies are manifested in the cross-floor passenger flow distribution rather than changes in the number of people in a single camera feed, they cannot be directly determined using localized behavior recognition methods based on single camera data.

[0026] Preferably, all undefined variables in this invention, if not explicitly defined, can be manually set thresholds.

[0027] This invention also provides a hotel intelligent monitoring system, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the hotel intelligent monitoring method. The hotel intelligent monitoring system can run on computing devices such as desktop computers, laptops, handheld computers, and cloud data centers. The runnable system may include, but is not limited to, processors, memory, and server clusters. The processor executes the computer program within the following system units: The monitoring and identification unit is used to identify monitoring devices on each floor of the hotel. The raw video data acquisition unit is used to obtain raw video data from the monitoring equipment at equal time intervals. The passenger flow statistics unit is used to construct a floor passenger flow time series from raw video data; The passenger flow pattern function calculation unit is used to coordinate the passenger flow time series of each floor to construct the passenger flow pattern function and obtain the passenger flow pattern transition value.

[0028] The intelligent storage control unit is used to intelligently store and control monitoring data based on passenger flow pattern transition values. The beneficial effects of this invention are as follows: This invention provides an intelligent monitoring method for hotel surveillance. The proposed scheme combines monitoring data to construct a cross-floor correlation function model of the change in passenger flow intensity over time on different floors. This allows for real-time quantification of the degree of abnormal passenger flow guidance within the hotel, making the monitoring system more intelligent in judging whether video data for a given time period has significantly higher storage value than during normal times. Furthermore, it has important warning significance for hotel security risk assessment, early warning identification of emergencies, and diagnosis of operational anomalies. By using the results of abnormal passenger flow identification to differentiate the storage quality of surveillance videos, it can ensure the high-quality evidence collection requirements for critical events while avoiding the huge consumption of storage space and network bandwidth caused by uniformly using high-quality recording for all time periods with only human targets. It also avoids the problem of excessively high thresholds for camera anomaly detection leading to insufficient coverage of high-quality recording time periods. This improves the utilization efficiency of storage resources and the ability to trace back key abnormal events, and significantly reduces the data flow read / write pressure on storage ports and calling ports. Attached Figure Description

[0029] The above and other features of the present invention will become more apparent from the detailed description of the embodiments shown in conjunction with the accompanying drawings. In the accompanying drawings, the same reference numerals denote the same or similar elements. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort. In the drawings: Figure 1 The diagram shows a flowchart of a smart monitoring method for hotel surveillance. Figure 2 The diagram shown is a structural diagram of a hotel monitoring and intelligent surveillance system. Detailed Implementation

[0030] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0031] like Figure 1 The diagram shown is a flowchart of a smart monitoring method for hotels. The following section will combine... Figure 1 This invention describes a smart monitoring method for hotel surveillance according to an embodiment of the present invention, the method comprising the following steps: S100 identifies surveillance equipment on each floor of the hotel. S200 obtains raw video data from the monitoring equipment at equal time intervals; S300 constructs a floor passenger flow time series using raw video data; S400, coordinates the passenger flow time series of each floor to construct the passenger flow pattern function and obtains the passenger flow pattern transition value; S500 intelligently stores and controls monitoring data based on passenger flow pattern transition values; Furthermore, in step S100, the method for identifying monitoring devices on each floor of the hotel is as follows: monitoring devices are deployed on each floor of the hotel, and each monitoring device is associated with corresponding floor identification information to distinguish monitoring devices on different floors. The monitoring equipment is a camera; The preferred monitoring equipment is a fixed-installation network camera, but it can also be an integrated access control machine with video recording function, a building intercom terminal, or other electronic devices with video recording capabilities.

[0032] Each monitoring device is assigned a unique device identifier and its floor information is recorded. This is to enable the raw video data subsequently collected by the monitoring device to be categorized and statistically analyzed according to the floor level. The deployment and identification association methods of the monitoring devices described above are standard practices for those skilled in the art and will not be elaborated upon here.

[0033] Further, in step S200, the method for obtaining raw video data from the monitoring device at equal time intervals is as follows: The preset time interval is 5-15 minutes. Each time interval is a recording point. Each recording point collects monitoring video data once from each monitoring device as raw video data. The raw video data is then stored after being identified by the monitoring device and time information.

[0034] The storage of raw video data here refers to initial storage, not terminal storage. Raw video data is high-quality video data directly acquired, with frame rate, resolution, and other specifications at the highest level available to the detection equipment, or the preset highest specifications. Initial storage refers to the original specifications of any video data, while terminal storage refers to video data after selective data compression. The video data is in the form of a monitoring video stream or video clips. The preset time interval determines the granularity of anomaly detection and intelligent village management in subsequent video data. Too small a granularity can easily distort the passenger flow pattern transition values, while too large a granularity can lead to excessive anomaly coverage and overfitting problems.

[0035] Furthermore, in step S300, the method for constructing the floor passenger flow time series using the original video data is as follows: for any recording point, pedestrian detection is performed on the original video data obtained through the monitoring equipment, the number of passengers in the corresponding field of view is statistically obtained, and the number of passengers corresponding to each monitoring equipment is accumulated by floor to obtain the passenger flow statistics for each floor. A time period is preset as the backtest duration RL, with the value of the backtest duration ranging from 3 to 6 hours. The time period from the current time to its reverse time direction RL is taken as the current backtest segment. For any floor, the passenger flow statistics of each recording point in the current backtest segment are used to form a sequence, which is recorded as the floor passenger flow time series.

[0036] The method for detecting pedestrians and counting the number of people in the corresponding field of view from the raw video data obtained through monitoring equipment is as follows: pedestrian detection is implemented using existing target detection algorithms, including pedestrian detection models based on convolutional neural networks, with YOLO algorithm being preferred, Faster R-CNN detection network algorithm, and moving target detection algorithms based on background modeling and foreground segmentation. Those skilled in the art can choose any mature pedestrian detection scheme according to the system performance requirements.

[0037] To avoid double counting of the same person, preferably, the detected person targets are further associated using a multi-target tracking algorithm. This can be done using a conventional tracking framework based on Kalman filtering and Hungarian matching, or using an existing DeepSORT-type multi-target tracking algorithm to associate the trajectories of the same person target appearing across frames within the same time interval.

[0038] At least one virtual counting line or counting area is pre-set in the field of view of the monitoring equipment. When any pedestrian trajectory crosses the virtual counting line or enters the counting area for the first time in the current time interval, the passenger flow count value corresponding to the pedestrian trajectory is incremented by one, and the same pedestrian trajectory is not counted repeatedly in the time interval. Thus, the number of passengers in the field of view of the monitoring equipment in the current time interval is obtained.

[0039] Further, in step S400, the method for constructing a passenger flow pattern function and obtaining the passenger flow pattern transition value by coordinating the passenger flow time series of each floor is as follows: For each recording point, construct a passenger flow time series matrix Xt of dimension n×m for each floor, where n represents the total number of floors and m is the sequence length; perform eigenvalue decomposition on the covariance matrix of Xt to obtain the corresponding eigenvalues ​​and eigenvector matrices; extract the eigenvectors corresponding to the first k largest eigenvalues ​​to construct a projection matrix and denote it as the passenger flow projection matrix P. k The global covariant matrix Zt is obtained through linear isomorphic mapping. =Xt·P k Each row of the matrix corresponds to the eigenvector zt of a single floor; the eigenvalue decomposition process of the covariance matrix is ​​obtained in Python by inputting the passenger flow time series matrix into the linalg.svd function of the NumPy library to obtain the corresponding eigenvalues ​​and eigenvector matrix; k is the preset number of eigenvalues ​​to be extracted, and its value range is k∈[1,3]. Calculate the arithmetic mean of the elements in each column of the global covariance matrix and define it as the global reference vector mt; for each eigenvector zt in the global covariance matrix, calculate its Euclidean norm distance dt=‖zt-mt‖ with the global reference vector mt, and extract the maximum norm offset dmax from it; the maximum norm offset is the maximum Euclidean norm distance; input dmax into the sigmoid function for normalization processing to generate the passenger flow pattern transition value PFMTV=1 / [1+e^(-dmax)].

[0040] The global baseline vector mt is composed of the arithmetic mean of each column. Further, in step S400, the method for constructing a passenger flow pattern function and obtaining the passenger flow pattern transition value by coordinating the passenger flow time series of each floor is as follows: for any floor, the range and standard deviation of its passenger flow time series are recorded as interval drop and interval dispersion, respectively. At the same recording point, the interval drop and interval dispersion are normalized respectively, and the binary pair formed by the interval drop and interval dispersion corresponding to the floor is recorded as the interval phase group; the average value of each interval phase group of the same floor within the backtesting time is calculated to obtain the baseline phase group, and the Euclidean distance between the interval phase group of the current recording point and the baseline phase group is calculated as the baseline deviation of the current recording point. If the baseline deviation of a recording point has a maximum value, and the corresponding element of the recording point in the passenger flow time series is greater than the average value of the passenger flow time series, then the recording floor has experienced effective passenger flow enhancement at that recording point; the median value of the passenger flow time series is recorded as the passenger flow level PFL. The preset floor proportion RRFL is used, with a value range of RRFL∈[0.2,0.5]. The floor number of interest is obtained by rounding down the product of the total number of floors and the relevant floor proportion, and is denoted as RFL. Any floor is taken as the parsing layer, and the record points where effective passenger flow enhancement occurs are identified as the mapping start point. The mapping quality of any record point along the time direction from the mapping start point is the number of non-parsing layers where at least one effective passenger flow enhancement occurs between the two record points. The mapping interval is from the record point to its first convergence constraint point along the time direction. The convergence constraint point is the first record point along the time direction from the mapping start point that satisfies the mapping quality being greater than or equal to the number of floors of interest. Each non-parsing layer where effective passenger flow enhancement occurs within the mapping interval is the mapping object. The proportion of the mapping object appearing in all mapping intervals of the parsing layer is the coverage of interest (CAC). The ratio of the number of record points between the record point where the mapping object first experiences effective passenger flow enhancement and the mapping start point to the total number of record points in the mapping interval is the non-mapping coverage (NMC). Based on the mapping intervals of each floor, a passenger flow pattern function is constructed to obtain the passenger flow pattern transition value PFMTV: ; Where i1 and i3 are cumulative variables, i2 is the index of the mapping interval, HF is the total number of floors, PFL(i1) represents the passenger flow level of the i1th floor, PFL.max is the maximum value among all passenger flow levels, RZ(i1) is the number of mapping intervals for the i1th floor, min{} is the minimum value function, and CAC i2,i3 and NMC i2,i3 These represent the coverage rate of the 3rd mapped object in the 2nd mapping interval and the coverage rate of the non-mapped object, respectively. When any floor is considered a parse floor, all other floors are considered non-parse floors. This concept is used to divide the mapping intervals for each floor separately; the time direction is the opposite direction of the time direction.

[0041] Further, in step S500, the method for intelligent storage control of monitoring data based on passenger flow pattern transition values ​​is as follows: a time period is preset as a fuzzy time window VTT, the value range of the fuzzy time window is VTT∈[0.5,1] hours; a time period is preset as a distribution period DTT, the value range of the distribution period is DTT∈[7,14] natural days; the set of each passenger flow pattern transition value within the current fuzzy time window is taken as the first distribution sequence, and the first distribution sequences within the distribution period are used to form the second distribution sequence; The current fuzzy time window is the time period from the current moment to its reverse time length of VTT. Therefore, the first distribution sequence is the set of passenger flow pattern transition values ​​within the divided VTT time period. The process of constructing the second distribution sequence from the first distribution sequences within the distribution period is based on the same current moment corresponding to the time scale in each natural day. The first distribution sequences corresponding to the VTT time periods within the distribution period are located in the reverse time direction, and their union is used as the second distribution sequence. If the current passenger flow pattern transition value at the current recording point is greater than or equal to the upper quartile of the second distribution sequence and greater than or equal to the average value of the first distribution sequence, then the first passenger flow anomaly condition is met. The average value of the minimum values ​​corresponding to the current passenger flow pattern transition values ​​in the first distribution sequence is used as the fuzzy benchmark. The difference between the current passenger flow pattern transition value and the fuzzy benchmark is the trend pressure amount. The difference between the current record point and the passenger flow pattern transition value of the previous record point is the step pressure amount. The product of the trend pressure amount and the step pressure amount is the transition pressure degree. If the transition pressure degree of a record point is positive and larger than that of the previous record point, then the second passenger flow anomaly condition is met. If the current recording point simultaneously meets the first and second abnormal passenger flow conditions, then it is determined that an abnormal passenger flow has occurred; search for the maximum value of the first transition pressure in the reverse time direction from the current recording point, and store the video data between that recording point and the current recording point in high quality.

[0042] High-quality storage of high-quality video data refers to storing video data with high resolution or high frame rate, or storing raw video data without data compression. Conversely, if high-quality video data storage is not performed, the frame rate or resolution of the raw video data is reduced to reduce the storage pressure on the server or computer.

[0043] If the current recording point does not indicate an abnormal passenger flow, video data compression should not be performed immediately. This is because the current recording point may be re-identified as a recording point indicating an abnormal passenger flow within a fuzzy time window. Therefore, the original video data should only be compressed and stored after a recording point has exceeded the current fuzzy time window.

[0044] An embodiment of the present invention provides a hotel monitoring intelligent monitoring system, such as... Figure 2 The diagram shown is a structural diagram of a hotel monitoring intelligent monitoring system according to the present invention. This embodiment of the hotel monitoring intelligent monitoring system includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the above-described embodiment of the hotel monitoring intelligent monitoring method.

[0045] The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program in a unit of the following system: The monitoring and identification unit is used to identify monitoring devices on each floor of the hotel. The raw video data acquisition unit is used to obtain raw video data from the monitoring equipment at equal time intervals. The passenger flow statistics unit is used to construct a floor passenger flow time series from raw video data; The passenger flow pattern function calculation unit is used to coordinate the passenger flow time series of each floor to construct the passenger flow pattern function and obtain the passenger flow pattern transition value.

[0046] The intelligent storage control unit is used to intelligently store and control monitoring data based on passenger flow pattern transition values. The aforementioned intelligent hotel monitoring system can run on computing devices such as desktop computers, laptops, handheld computers, and cloud servers. The system that can run on this intelligent hotel monitoring system may include, but is not limited to, processors and memory. Those skilled in the art will understand that the examples described are merely illustrations of an intelligent hotel monitoring system and do not constitute a limitation on such a system. It may include more or fewer components, combinations of certain components, or different components. For example, the intelligent hotel monitoring system may also include input / output devices, network access devices, buses, etc.

[0047] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the hotel monitoring and intelligent surveillance system, connecting various parts of the system via various interfaces and lines.

[0048] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the hotel monitoring intelligent monitoring system by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0049] Although the invention has been described in considerable detail and particularly with regard to several of the described embodiments, it is not intended to limit itself to any of these details or embodiments or any particular embodiment, thereby effectively covering the intended scope of the invention. Furthermore, the invention has been described above with respect to embodiments foreseeable by the inventors in order to provide a useful description, and non-substantial modifications to the invention that have not yet been foreseen may still represent equivalent modifications.

Claims

1. A hotel monitoring intelligent monitoring method, characterized in that, The method comprises the following steps: S100, identifying monitoring devices on each floor in the hotel respectively; S200, obtaining original video data from the monitoring devices at equal time intervals; S300, constructing a floor passenger flow time sequence through the original video data; S400, constructing a passenger flow mode function to obtain a passenger flow mode transition value in cooperation with the passenger flow time sequences of each floor; S500, intelligently storing and controlling the monitoring data according to the passenger flow mode transition value.

2. The intelligent monitoring method for hotel monitoring according to claim 1, wherein, In step S100, the method of identifying monitoring devices on each floor in the hotel is as follows: monitoring devices are arranged on each floor of the hotel, and corresponding floor identification information is associated with each monitoring device to distinguish monitoring devices on different floors. The monitoring device is a camera.

3. The intelligent monitoring method for hotel monitoring as claimed in claim 1, wherein, In step S200, the method of obtaining original video data from the monitoring devices at equal time intervals is as follows: The preset time interval is 5-15 minutes, each time interval is taken as a recording point, and monitoring video data is collected from each monitoring device once as original video data. After the original video data is identified by the monitoring device and time information, it is stored.

4. The intelligent monitoring method for hotel monitoring as claimed in claim 1, wherein, In step S300, the method of constructing a floor passenger flow time sequence through the original video data is as follows: for any recording point, the original video data obtained through the monitoring device is subjected to pedestrian detection, the number of passengers in the corresponding field of view is counted, and the number of passengers corresponding to each monitoring device is accumulated in units of floors to obtain passenger flow statistics of each floor. A preset time period is taken as a backtesting length RL, and the value range of the backtesting length is 3-6 hours. A current backtesting segment is taken from the current time to the RL period in the reverse time direction. For any floor, the passenger flow statistics of each recording point in the current backtesting segment form a sequence, denoted as a floor passenger flow time sequence.

5. The intelligent monitoring method for hotel monitoring as claimed in claim 1, wherein, In step S400, the method of constructing the passenger flow mode function and obtaining the passenger flow mode transition value in cooperation with the passenger flow time sequence of each floor is: constructing a passenger flow time sequence matrix X t with a dimension of n x m for each recording point of each floor, where n represents the total number of floors, and m is the sequence length; performing eigenvalue decomposition on the covariance matrix of X t to obtain the corresponding eigenvalues and eigenvector matrix; intercepting the eigenvectors corresponding to the first k largest eigenvalues to construct a projection matrix and recording it as a passenger flow projection matrix P k , obtaining a global covariance matrix Z t = X t · P k through linear isomorphism mapping, where each row of the matrix corresponds to the eigenvector z t of a single floor; The arithmetic mean of each column element of the global covariance matrix is calculated, which is defined as a global reference vector mt. For each feature vector zt in the global covariance matrix, the Euclidean norm distance dt=‖zt-mt‖ of the global reference vector mt is calculated, and the maximum norm offset dmax is extracted from it. The maximum norm offset is the maximum Euclidean norm distance. dmax is input into a sigmoid function for normalization processing to generate a passenger flow mode transition value PFMTV=1 / [1+e^(-dmax)].

6. The intelligent monitoring method for hotel monitoring as claimed in claim 1, wherein, In step S400, the method for constructing the passenger flow mode function to obtain the passenger flow mode transition state value in cooperation with the passenger flow time series of each floor is: for any floor, the range difference and the standard deviation of the passenger flow time series are recorded as the interval drop and the interval dispersion respectively, under the same recording point, the interval drop and the interval dispersion are normalized respectively, and the binary group composed of the interval drop and the interval dispersion of the floor is recorded as the interval phase state group; the average value of each interval phase state group of the same floor in the back test duration is calculated to obtain the baseline phase state group, the Euclidean distance between the interval phase state group of the current recording point and the baseline phase state group is calculated as the baseline deviation of the current recording point, if the baseline deviation of a recording point appears a maximum value, and the corresponding element of the recording point in the passenger flow time series is greater than the average value of the passenger flow time series, then the effective passenger flow enhancement of the recording point of the floor occurs; the median value of the passenger flow time series is recorded as the passenger flow level PFL; The preset concerned floor proportion RRFL has a value range of RRFL∈[0.2, 0.5]; the product of the total number of floors and the relevant floor proportion is rounded down to obtain the concerned floor quantity, which is recorded as RFL; Any floor is taken as an analysis layer, the recording point where the effective passenger flow enhancement occurs is taken as a mapping starting point, the mapping quality of any recording point along the time direction of the mapping starting point is the number of non-analysis layers where at least one effective passenger flow enhancement occurs between the two recording points, the mapping interval is from the recording point to the first convergent constraint point along the time direction of the mapping starting point, and the convergent constraint point is the first recording point along the time direction of the mapping starting point which satisfies that the mapping quality is greater than or equal to the concerned floor quantity; the non-analysis layers where the effective passenger flow enhancement occurs in the mapping interval are mapping objects, the proportion of the mapping objects appearing in all mapping intervals of the analysis layer is the concerned coverage rate CAC, and the ratio of the number of recording points between the recording point where the effective passenger flow enhancement of the mapping object first occurs in the mapping interval and the mapping starting point to the total number of recording points in the mapping interval is the non-mapping coverage rate NMC; the passenger flow mode function is constructed according to the mapping interval division of each floor to obtain the passenger flow mode transition state value.

7. The intelligent monitoring method for hotel monitoring as claimed in claim 1, wherein, In step S500, the method for intelligently storing and controlling the monitoring data according to the passenger flow mode transition state value is: a time period is preset as a fuzzy time window; a time period is preset as a distribution period; a set composed of the passenger flow mode transition state values in the current fuzzy time window is taken as a first distribution sequence, and a second distribution sequence is formed by the first distribution sequences in the distribution period; if the passenger flow mode transition state value of the current recording point is greater than or equal to the upper quartile of the second distribution sequence and greater than or equal to the average value of the first distribution sequence, the first passenger flow abnormal condition is met; The average value of the minimum value corresponding to the passenger flow mode transition state value of the current first distribution sequence is taken as a fuzzy reference, the difference between the current passenger flow mode transition state value and the fuzzy reference is a trend pressure, the difference between the passenger flow mode transition state values of the current recording point and the previous recording point is a step pressure, and the product of the trend pressure and the step pressure is a transition pressure; if the transition pressure of a recording point is positive and greater than that of the previous recording point, the second passenger flow abnormal condition is met; If the current record point satisfies the first passenger flow abnormal condition and the second passenger flow abnormal condition simultaneously, it is judged that the passenger flow abnormality occurs; the first maximum value of the transfer state pressure in the time reverse direction is searched from the current record point, and the video data between the record point and the current record point is stored in high quality.

8. A hotel monitoring intelligent monitoring system, characterized in that, The hotel monitoring intelligent monitoring system comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps in the hotel monitoring intelligent monitoring method according to any one of claims 1-7, and the hotel monitoring intelligent monitoring system is used in a desktop computer, a notebook computer, a palm computer, and a computing device of a cloud data center.

Images