Hot area object aggregation detection method and system based on non-visual acquisition

Abstract

The invention relates to the technical field of object detection, in particular to a hot area object gathering detection method and system based on non-visual collection, and aims at solving the problems that a traditional visual collection detection method is limited in complex illumination, shielding and privacy sensitive areas and complex in data processing. The data preprocessing unit carries out normalization, denoising and correction on data, and the feature extraction module extracts density, distance, movement and other related features. The aggregation judgment module judges the aggregation condition through an MLP model, and the alarm and feedback module gives an alarm to a manager and feeds back detailed data when aggregation exists. The system comprises a non-visual acquisition device layer, a data processing center and a user interface, acquisition devices are reasonably distributed to form a monitoring network, the data processing center completes data processing and analysis, the user interface facilitates parameter setting and information viewing, and the system can be widely applied to object gathering monitoring in public places, large-scale activity sites and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of object detection, and in particular to a method and system for detecting hot zone object aggregation based on non-visual acquisition. Background Art

[0002] In today's society, the detection of gatherings of people or specific objects in a specific area is of great significance and is widely used in many fields such as public place management, large-scale event organization, and security prevention. Traditionally, detection methods based on visual acquisition, such as the use of camera monitoring systems, are more commonly used. However, this vision-based detection method has many difficult-to-overcome problems. First, under complex lighting conditions, such as direct strong light, dim light, or environments with frequent light changes, the image quality obtained by the camera will be significantly reduced, which may lead to inaccurate identification and positioning of objects. For example, at a large outdoor concert at night, the complex light such as the strong light from the stage and the flash in the hands of the audience will seriously interfere with the camera's monitoring of the gathering of the audience.

[0003] Secondly, occlusion also has a great impact on visual acquisition. When objects block each other or there are obstacles blocking them, the camera cannot fully obtain information about all objects, resulting in errors in cluster detection. For example, at a crowded indoor trade fair, people are crowded together, and there are obstacles such as exhibits and display racks. It is difficult for the camera to accurately determine the number and distribution of the obscured part of the crowd.

[0004] Furthermore, as society's awareness of privacy protection continues to increase, privacy issues related to visual acquisition are becoming increasingly prominent. The images collected by the camera contain a large amount of personal privacy information such as appearance and behavior. If the data is not properly managed during storage, transmission and processing, it is easy to cause privacy leakage risks, which limits its application in some places with high privacy requirements, such as hospital wards and school classrooms.

[0005] In addition, detection methods based on visual acquisition often need to process massive amounts of image data, which places high demands on data processing capabilities and storage resources. The entire process from image acquisition, compression, transmission to analysis and processing requires a lot of computing resources and time, increasing the complexity and cost of the system.

[0006] Therefore, according to the above-mentioned related technologies, it is urgent to develop a hot spot object aggregation detection method and system based on non-visual acquisition. Summary of the invention

[0007] In view of this, the purpose of the present invention is to propose a hot spot object cluster detection method and system based on non-visual acquisition, so as to provide a new detection method and system that can effectively overcome the above-mentioned problems, avoid the limitations brought by visual acquisition, and achieve more accurate, efficient and privacy-friendly object cluster detection.

[0008] Based on the above objectives, the present invention provides a method and system for detecting hot spot object aggregation based on non-visual acquisition.

[0009] A hot zone object gathering detection system based on non-visual acquisition, comprising a non-visual acquisition device layer, a data processing center and a user interface;

[0010] The non-visual acquisition device layer includes a data acquisition module;

[0011] The data acquisition module is used to collect data on objects in the hot zone;

[0012] The non-visual acquisition device layer includes an infrared sensor array, a radio frequency identification RFID reader-writer network, and a Bluetooth signal detection device;

[0013] The infrared sensor is used to obtain the temperature information and the general thermal distribution of the object;

[0014] The radio frequency identification RFID reader is used to obtain the identity and location information of the object with the RFID tag;

[0015] The Bluetooth signal detection device is used to scan the signal strength and device information of the Bluetooth-enabled object;

[0016] The data processing center includes a data preprocessing unit, a feature extraction module and an aggregation judgment module;

[0017] The data preprocessing unit is used to receive the collected data and preprocess it, normalize the data from different sources, remove noise and outliers, for example, smooth the abnormal temperature fluctuation points in the infrared sensor data, and screen and correct the weak and unstable signals in the RFID or Bluetooth signals;

[0018] The feature extraction module is used to extract features related to object aggregation from the preprocessed data, wherein the features include density features, distance features and movement features of the objects;

[0019] The density feature of the object is obtained by calculating the number of objects detected within a unit area;

[0020] The distance feature of the object is the distance relationship between different objects;

[0021] The movement characteristics of the object are the movement speed and direction of the object;

[0022] For infrared data, the movement and aggregation of objects can be inferred by the changes in the temperature distribution of the hot zone; for RFID and Bluetooth data, the location changes and distribution density of tags or devices can be used for analysis;

[0023] The aggregation judgment module uses the MLP model based on the extracted features to determine whether there is an object aggregation in the hot zone;

[0024] The analysis process of the MLP model for the aggregation judgment module is as follows:

[0025] Step S2031. Input layer setting: The number of input layer nodes is consistent with the number of features extracted from the data. If the density feature, distance feature and movement feature of the object are obtained after feature extraction, the input layer will have 3 nodes for receiving the corresponding feature values ​​as input signals;

[0026] Step S2032. Hidden layer design: The number of hidden layers and the number of neurons in each layer are selected. The number of layers and the number of neurons are determined through experiments and verification. The input layer has n nodes. The first hidden layer is set to have 2n neurons. The neurons receive the output signal from the neurons in the previous layer, multiply it by the corresponding connection weight, and then add a bias term. For the first neuron in the hidden layer, it receives n signals x1, x2, ..., x from the input layer. n , the corresponding connection weight is w 11 , w 12 ,…,w 1n , the bias is b1, and the input sum is For the calculated z1, the ReLU activation function is used to process the output y1=max(0,z1);

[0027] Step S2033. Output layer setting: The output layer has only one node, which is used to output the result of clustering or not. It receives the output signal from the neurons of the last hidden layer. After weighted summation and adding the bias term, the last hidden layer has m neurons, and the output signals are h1, h2, ..., h m , the connection weight is w o1 , w o2 ,…,w om , with a bias of b o , the sum of the inputs to the output layer nodes is The output result is activated by Sigmoid function, and the output

[0028] Step S2034. Training process analysis: Use the labeled data to input the feature vector into the input layer. After calculation and transmission in the hidden layer, a prediction result is finally obtained in the output layer. The loss between the prediction result and the true label is calculated. The true label is y (0 to 1 indicates the absence or presence of aggregation), the prediction result is y0, and the cross entropy loss function L = [ylog(y o )+(1-y)log(1-y o )], update the weights and biases of the model by gradient descent, calculate the gradient of the loss function for each weight and bias, and then update the weights and biases according to a certain learning rate. ij , the update formula is Where α is the learning rate, which is iterated over the entire training dataset, adjusting the weights and biases and calculating the loss between the predicted results and the true labels.

[0029] The user interface includes an alarm and feedback module, which is used to send alarm information to relevant management systems or personnel and feed back detailed aggregated data to users so that corresponding measures can be taken in time;

[0030] The cluster data includes cluster locations and object quantity estimates.

[0031] The present invention also provides a method for detecting hot zone object aggregation based on non-visual acquisition, comprising the following steps:

[0032] Step S1. Using a non-visual acquisition device layer to collect data;

[0033] Step S2. Using a data processing center to comprehensively process and analyze the collected data for object aggregation detection and alarm functions;

[0034] Step S3. Use a user interface to provide managers with an intuitive operation interface for setting parameters, viewing detection data and alarm information, and feeding back detailed aggregated data to users so that appropriate measures can be taken in a timely manner;

[0035] Preferably, the data collection in step S1 includes the following steps:

[0036] Step S101. Deployment of non-visual acquisition equipment: According to the shape, size and detection requirements of the hot zone, reasonably deploy infrared sensors, RFID readers and writers, and Bluetooth signal detection equipment. For example, install a certain number of infrared sensors around and in the center of a large square, set up RFID readers and writers at entrances and exits and key channels, and evenly distribute Bluetooth signal detection equipment throughout the area;

[0037] Step S102. Data collection and transmission. Each collection device collects data according to the set time interval or trigger condition, and transmits the data to the data processing center through wired or wireless communication. For example, the infrared sensor collects temperature and heat distribution data once a second and transmits it to the data processing center through the WIFI network. The RFID reader transmits the tag information in real time through Ethernet when the tag is detected.

[0038] Preferably, the processing process of the data processing center in step S2 is as follows:

[0039] Step S201. The data preprocessing unit receives the collected data and preprocesses it. It normalizes the data from different sources and removes noise and outliers. For example, the temperature data collected by the infrared sensor is smoothed by a sliding average filter. Assume that the collected temperature sequence is {T1, T2, ..., T n}, set the sliding window size to m, then the temperature value after smoothing For RFID and Bluetooth signal data, the signal strength threshold is used for screening, weak signal data with signal strength lower than the set threshold is removed, and weak and unstable signals are screened and corrected;

[0040] Step S202. Feature extraction module extracts features related to object aggregation from the preprocessed data, including density features, distance features and movement features of objects; the density feature of objects is calculated by calculating the number of objects detected within a unit area; the distance feature of objects is the distance relationship between different objects; the movement feature of objects is the movement speed and movement direction of objects; for infrared data, the movement and aggregation of objects can be inferred by the change in temperature distribution in the hot zone; for RFID and Bluetooth data, the position change and distribution density of tags or devices can be analyzed, and the object density feature is calculated based on the preprocessed data. For objects detected by RFID, the hot zone is divided into several small areas, and the number of tags in each small area is counted as the object density. For the movement feature, the movement speed and movement direction of the device are calculated by the continuously collected Bluetooth device position information. For example, the speed is calculated based on the coordinates (x1, y1) and (x2, y2) of two time points t1 and t2. Direction

[0041] Step S203. The clustering judgment module uses the MLP model, inputs the extracted feature vector into the trained MLP model, and the model outputs the result of clustering or not.

[0042] Preferably, the parameters in step S3 include an aggregation judgment threshold and an alarm mode.

[0043] Preferably, the detection data in step S3 includes cluster locations and object quantity estimates.

[0044] Beneficial effects of the present invention:

[0045] The present invention overcomes the limitation of visual acquisition and uses non-visual acquisition, which is not affected by illumination and partial occlusion. It can effectively realize object gathering detection in various complex environments and avoids the privacy issues that are easily involved in visual acquisition. The present invention uses non-visual acquisition equipment and does not need to obtain visual image information of the object, which is more in line with the requirements of privacy protection and can be widely used in more privacy-sensitive hot spot detection scenarios.

[0046] In the present invention, the data preprocessing unit is used to normalize the data from different sources, remove noise and outliers, improve the data quality, and lay a foundation for more accurate feature extraction and aggregation judgment. Moreover, through the comprehensive analysis of data from a variety of non-visual acquisition devices, the judgment of object aggregation is made more accurate and comprehensive. The model used after training with a large amount of labeled data makes the judgment of object aggregation more accurate, can adapt to the changes in object aggregation patterns in different scenarios, and effectively improves the accuracy of aggregation detection.

[0047] The present invention has a rapid alarm mechanism, which can give timely and effective alarms and provide detailed feedback information, which helps managers to deal with problems that may be caused by object aggregation more efficiently. In addition, the user interface provides managers with an intuitive operation interface, which can easily set parameters, view monitoring data and alarm information. This greatly reduces the threshold for using the system, enables managers to easily operate and manage the entire detection system, and improves the practicality and ease of use of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 It is a module diagram of a hot zone object gathering detection system based on non-visual acquisition in the present invention;

[0050] Figure 2 The present invention is a flowchart of a method for detecting hot spot object aggregation based on non-visual acquisition. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0052] like Figure 1 As shown, a hot zone object gathering detection system based on non-visual acquisition includes a non-visual acquisition device layer, a data processing center and a user interface;

[0053] The non-visual acquisition device layer includes a data acquisition module, which is used to collect data on objects in the hot zone;

[0054] The non-visual acquisition device layer includes infrared sensor arrays, radio frequency identification RFID reader-writer networks, and Bluetooth signal detection devices. These devices can obtain relevant characteristic information of objects. For example, infrared sensors can obtain the temperature information and approximate thermal distribution of objects, RFID readers can obtain the identity and location information of objects with RFID tags, and Bluetooth signal detection devices can scan the signal strength and device information of objects with Bluetooth enabled.

[0055] The data processing center includes a data preprocessing unit, a feature extraction module and an aggregation judgment module;

[0056] The data preprocessing unit is used to receive the collected data and preprocess it, normalize the data from different sources, remove noise and outliers, for example, smooth the abnormal temperature fluctuation points in the infrared sensor data, and screen and correct the weak and unstable signals in the RFID or Bluetooth signals;

[0057] The feature extraction module is used to extract features related to object aggregation from the preprocessed data. These features include density features, distance features, and movement features of the objects.

[0058] The density feature of the object is obtained by calculating the number of objects detected per unit area;

[0059] The distance feature of the object is the distance relationship between different objects;

[0060] The movement characteristics of the object are the movement speed and direction of the object;

[0061] For infrared data, the movement and aggregation of objects can be inferred by the changes in the temperature distribution of the hot zone; for RFID and Bluetooth data, the location changes and distribution density of tags or devices can be used for analysis;

[0062] The clustering judgment module uses the MLP model based on the extracted features to determine whether there is a cluster of objects in the hot zone;

[0063] The analysis process of the MLP model for the aggregation judgment module is as follows:

[0064] Step S2031. Input layer setting: The number of input layer nodes is consistent with the number of features extracted from the data. For example, if after feature extraction, the density feature, distance feature, movement feature, etc. of the object are obtained, if there are n such features, then the input layer will have n nodes, and these nodes will receive the corresponding feature values ​​as input signals. Taking hot spot object detection as an example, assuming that the density feature is the number of objects per square meter (a value), the distance feature is the average distance between objects (a value), and the movement feature includes the average movement speed and movement direction (two values), then there are a total of 4 features, and the input layer has 4 nodes;

[0065] Step S2032. Hidden layer design: The number of hidden layers and the number of neurons in each layer are selected. Generally speaking, the number of layers and the number of neurons need to be determined through experiments and verification. You can start with a simple structure, such as setting 1-3 hidden layers. The number of neurons in each layer can be determined based on the number of nodes in the input layer and the expected complexity of the output layer. For example, if the input layer has n nodes, the first hidden layer can be set to have 2n neurons. The weights between neurons are randomly initialized. These weights determine the intensity of transmission and processing of input signals between neurons. Each neuron receives the output signal from the neurons in the previous layer, multiplies it by the corresponding connection weight, and then adds a bias term. For example, for the first neuron in the hidden layer, assume that it receives n signals x1, x2, ..., x from the input layer. n , the corresponding connection weight is w 11 , w 12 ,…,w 1n , the bias is b1, then its input sum For the calculated z1, the ReLU activation function is used to process it to get the output y1=max(0,z1). The role of the activation function is to introduce nonlinear factors so that the multilayer perceptron can handle complex nonlinear relationships. Without the activation function, the multilayer perceptron is equivalent to a linear model, and its expression ability will be greatly limited.

[0066] Step S2033. Output layer setting: The output layer has only one node, which is used to output the result of clustering or not. This node receives the output signal from the neurons in the last hidden layer, and also undergoes weighted summation and bias term addition operations. Assuming that the last hidden layer has m neurons, the output signals are h1, h2, ..., h m , the connection weight is w o1 , w o2 ,…,w om , with a bias of b o , then the sum of the inputs of the output layer nodes The output result uses the Sigmoid activation function. For example, the output result is a probability value, such as 0-1, which represents the probability of aggregation. That is, the Sigmoid activation function is used, that is, the output In this way, o The closer the value is to 1, the higher the probability of aggregation exists, and the closer it is to 0, the lower the probability of aggregation exists.

[0067] Step S2034. Training process analysis: When training the MLP model, use the prepared labeled data, input the feature vector into the input layer, and after calculation and transmission in the hidden layer, finally get a prediction result in the output layer. Calculate the loss between the prediction result and the true label. The true label is y (0 to 1 means no or no aggregation), the prediction result is y0, and the cross entropy loss function L = [ylog(y o )+(1-y)log(1-y o )], update the weights and biases of the model by gradient descent, calculate the gradient of the loss function for each weight and bias, and then update the weights and biases according to a certain learning rate. For example, for weight w ij , the update formula is Where α is the learning rate, and this process will be iterated over the entire training dataset, continuously adjusting the weights and biases until the loss function converges to a smaller value.

[0068] The user interface includes an alarm and feedback module. When it is determined that objects are gathering, an alarm message is sent to the relevant management system or personnel, and detailed gathering data can be fed back to the user so that appropriate measures can be taken in time.

[0069] Detailed aggregation data includes aggregation locations and object count estimates.

[0070] In order to better understand the above embodiments, Figure 2 As shown, the present invention also provides a flowchart of a method for detecting hot zone object aggregation based on non-visual acquisition, comprising the following steps:

[0071] Step S1. Use non-visual acquisition device layer to collect data:

[0072] The data collection in step S1 includes the following steps:

[0073] Step S101. Deployment of non-visual acquisition equipment: According to the shape, size and detection requirements of the hot zone, reasonably deploy infrared sensors, RFID readers and writers, and Bluetooth signal detection equipment. For example, install a certain number of infrared sensors around and in the center of a large square, set up RFID readers and writers at entrances and exits and key channels, and evenly distribute Bluetooth signal detection equipment throughout the area;

[0074] Step S102. Data collection and transmission. Each collection device collects data according to the set time interval or trigger condition, and transmits the data to the data processing center through wired or wireless communication. For example, the infrared sensor collects temperature and heat distribution data once a second and transmits it to the data processing center through the WIFI network. The RFID reader transmits the tag information in real time through Ethernet when the tag is detected.

[0075] Step S2. Using the data processing center to comprehensively process and analyze the collected data to achieve the detection and alarm functions of object aggregation;

[0076] The processing process of the data processing center in step S2 is as follows:

[0077] Step S201. The data preprocessing unit receives the collected data and preprocesses it. It normalizes the data from different sources and removes noise and outliers. For example, the temperature data collected by the infrared sensor is smoothed by a sliding average filter. Assume that the collected temperature sequence is {T1, T2, ..., T n}, set the sliding window size to m, then the temperature value after smoothing For RFID and Bluetooth signal data, the signal strength threshold is used for screening, weak signal data with signal strength lower than the set threshold is removed, and weak and unstable signals are screened and corrected;

[0078] Step S202. Feature extraction module extracts features related to object aggregation from the preprocessed data, including density features, distance features and movement features of objects; the density feature of objects is calculated by calculating the number of objects detected within a unit area; the distance feature of objects is the distance relationship between different objects; the movement feature of objects is the movement speed and movement direction of objects; for infrared data, the movement and aggregation of objects can be inferred by the change in temperature distribution in the hot zone; for RFID and Bluetooth data, the position change and distribution density of tags or devices can be analyzed, and the object density feature is calculated based on the preprocessed data. For objects detected by RFID, the hot zone is divided into several small areas, and the number of tags in each small area is counted as the object density. For the movement feature, the movement speed and movement direction of the device are calculated by the continuously collected Bluetooth device position information. For example, the speed is calculated based on the coordinates (x1, y1) and (x2, y2) of two time points t1 and t2. Direction

[0079] Step S203. The clustering judgment module uses the MLP model to input the extracted feature vector into the trained MLP model, and the model outputs the result of clustering or not;

[0080] Step S3. The user interface provides an intuitive operation interface for managers to set parameters (such as aggregation judgment thresholds and alarm methods), view detection data and alarm information, and can feed back detailed aggregation data to users so that corresponding measures can be taken in a timely manner. The detection data includes the aggregation location and the estimated number of objects. When it is determined that aggregation has occurred, the data processing center sends alarm information to the managers through text messages, emails or system messages, and displays detailed information of the aggregation area on the user interface, such as marking the aggregation location on the hot zone map, displaying the estimated number of aggregated objects, etc., so that managers can take corresponding measures such as evacuation and guidance.

[0081] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention is limited to these examples. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0082] The present invention is intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A hot spot object gathering detection system based on non-visual acquisition, characterized in that: Includes non-visual acquisition device layer, data processing center and user interface; The non-visual acquisition device layer includes a data acquisition module; The data acquisition module is used to collect data on objects in the hot zone; The non-visual acquisition device layer includes an infrared sensor array, a radio frequency identification RFID reader-writer network, and a Bluetooth signal detection device; The data processing center includes a data preprocessing unit, a feature extraction module and an aggregation judgment module; The data preprocessing unit is used to receive the collected data, preprocess it, normalize the data from different sources, and remove noise and outliers; The feature extraction module is used to extract features related to object aggregation from the preprocessed data; The clustering judgment module uses the MLP model based on the extracted features to determine whether there is a cluster of objects in the hot zone.

2. The hot zone object gathering detection system based on non-visual acquisition according to claim 1 is characterized in that: The infrared sensor is used to obtain the temperature information and approximate thermal distribution of the object; the radio frequency identification RFID reader is used to obtain the identity and location information of the object with the RFID tag; the Bluetooth signal detection device is used to scan the signal strength and device information of the object with the Bluetooth function turned on.

3. The hot zone object gathering detection system based on non-visual acquisition according to claim 1 is characterized in that: The features related to object aggregation extracted by the feature extraction module from the preprocessed data include density features, distance features and movement features of the objects. The density features of the objects are obtained by calculating the number of objects detected per unit area, the distance features of the objects are the distance relationship between different objects, and the movement features of the objects are the movement speed and direction of the objects.

4. The hot zone object gathering detection system based on non-visual acquisition according to claim 1 is characterized in that: The aggregation judgment module uses the MLP model based on the extracted features to determine whether there is an object aggregation in the hot zone; The analysis process of the MLP model for the aggregation judgment module is as follows: Step S2031. Input layer setting: The number of input layer nodes is consistent with the number of features extracted from the data. After feature extraction, the density feature, distance feature and movement feature of the object are obtained. The input layer has 3 nodes for receiving the corresponding feature values ​​as input signals; Step S2032. Hidden layer design: Determine the number of hidden layers and the number of neurons in each layer. The input layer has n nodes, and the first hidden layer has 2n neurons. The neurons receive the output signal from the neurons in the previous layer, multiply it by the corresponding connection weight, and then add the bias term. For the first neuron in the hidden layer, it receives n signals x1, x2, ..., x from the input layer. n , the corresponding connection weight is w 11 , w 12 ,…,w 1n , the bias is b1, and the input sum is For the calculated z1, the ReLU activation function is used to process the output y1=max(0,z1); Step S2033. Output layer setting: The output layer has only one node, which is used to output the result of clustering or not. It receives the output signal from the neurons of the last hidden layer. After weighted summation and adding the bias term, the last hidden layer has m neurons, and the output signals are h1, h2, ..., h m , the connection weight is w o1 , w o2 ,…,w om , with a bias of b o , the sum of the inputs to the output layer nodes is The output result is activated by Sigmoid function, and the output Step S2034. Training process analysis: Use the labeled data to input the feature vector into the input layer. After calculation and transmission in the hidden layer, a prediction result is finally obtained in the output layer. The loss between the prediction result and the true label is calculated. The true label is y, and the prediction result is y0. The cross entropy loss function L = [ylog(y o )+(1-y)log(1-y o )], update the weights and biases of the model by gradient descent, calculate the gradient of the loss function for each weight and bias, and then update the weights and biases. For weight w ij , the update formula is Where α is the learning rate, which is iterated over the entire training dataset, adjusting the weights and biases and calculating the loss between the predicted results and the true labels.

5. The hot zone object gathering detection system based on non-visual acquisition according to claim 1 is characterized in that: The user interface includes an alarm and feedback module, which is used to send alarm information to relevant management systems or personnel and feed back detailed aggregated data to users so that corresponding measures can be taken in time; The cluster data includes cluster locations and object quantity estimates.

6. A method for detecting hot spot object aggregation based on non-visual acquisition, characterized in that: The following steps are involved: Step S1. Using a non-visual acquisition device layer to collect data; Step S2. Using a data processing center to comprehensively process and analyze the collected data for object aggregation detection and alarm functions; Step S3. A user interface is used to provide an intuitive operation interface for setting parameters, viewing detection data and alarm information, and feeding back detailed aggregated data to the user so that corresponding measures can be taken in a timely manner.

7. The method for detecting hot spot object aggregation based on non-visual acquisition according to claim 6, characterized in that: The data collection in step S1 includes the following steps: Step S101. Deployment of non-visual acquisition equipment: deploy infrared sensors, RFID readers and writers, and Bluetooth signal detection equipment according to the shape, size, and detection requirements of the hot zone; Step S102: Data collection and transmission. Each collection device collects data according to a set time interval or trigger condition, and transmits the data to a data processing center via wired or wireless communication.

8. The method for detecting hot spot object aggregation based on non-visual acquisition according to claim 6, characterized in that: The processing process of the data processing center in step S2 is as follows: Step S201. The data preprocessing unit receives the collected data and preprocesses it. For the temperature data collected by the infrared sensor, the sliding average filter method is used for smoothing. For the RFID and Bluetooth signal data, the signal strength threshold is used for screening, and the weak signal data and unstable signals are screened and corrected. Step S202. Feature extraction module, extracts the density feature, distance feature and movement feature of the object from the preprocessed data; for infrared data, infers the movement and aggregation of the object through the change of temperature distribution in the hot zone; for objects detected by RFID, the hot zone is divided into several small areas, and the number of tags in each small area is counted as the object density; for movement features, the moving speed and direction of the device are calculated through the continuously collected Bluetooth device location information, and the speed is calculated according to the coordinates (x1, y1) and (x2, y2) at two time points t1 and t2. Direction Step S203. The clustering judgment module uses the MLP model, inputs the extracted feature vector into the trained MLP model, and the model outputs the result of clustering or not.

9. The method for detecting hot spot object aggregation based on non-visual acquisition according to claim 8, characterized in that: The smoothing process in step S201 is as follows: The temperature sequence collected by the infrared sensor is {T1, T2, ..., T n }, set the sliding window size to m, then the temperature value after smoothing 10. The method for detecting hot spot object aggregation based on non-visual acquisition according to claim 6, characterized in that: The parameters in step S3 include an aggregation judgment threshold and an alarm mode, and the detection data in step S3 include an aggregation position and an estimated number of objects.

Images