Video monitoring system for multi-source data fusion of construction site under internet of things architecture

Abstract

The application relates to the technical field of building monitoring, in particular to a video monitoring system for multi-source data fusion of a building site under an Internet of Things architecture, which solves the problems that, in the process of video monitoring of a building construction site, the construction site environment is complex, the construction scene is various, multi-source data of different building scenes cannot be fused and processed, the building monitoring and control effect is affected, and accurate identification and abnormal processing according to scene data are inconvenient; the video monitoring system comprises a data acquisition module, an acquisition scene identification module, a data fusion module and an application management module; the data acquisition module is used for acquiring real-time scene data of multiple building monitoring scenes from various sensors and video devices; the scene data is composed of video data, environment data and personnel positioning data; the multi-source data of different building scenes are fused and identified, scene abnormalities can be effectively identified, and the monitoring and control effect of the building scene is improved.

Description

Technical Field

[0001] This invention relates to the field of building monitoring technology, specifically a video surveillance system for building sites that integrates multi-source data under an Internet of Things (IoT) architecture. Background Technology

[0002] The construction industry is a vital material production sector and pillar industry of my country's national economy, playing a crucial role in improving living conditions, enhancing infrastructure, and driving economic growth. However, the unique working environment of construction sites, characterized by high personnel mobility and complex large equipment, coupled with a lack of safety awareness among some construction workers, often results in significant safety hazards. Previously, construction sites largely relied on manual scheduling and monitoring to mitigate some of these hazards.

[0003] In the current process of video surveillance of construction sites, the complex environment and diverse construction scenarios make it impossible to fuse multi-source data from different construction scenarios, which affects the effectiveness of building monitoring and control. Furthermore, it is not convenient to make accurate judgments and handle anomalies based on scenario data. Therefore, it does not meet the existing needs. To address this, we propose a video surveillance system for building sites that fuses multi-source data under an Internet of Things (IoT) architecture. Summary of the Invention

[0004] The purpose of this invention is to provide a video monitoring system for building sites that integrates multi-source data under an Internet of Things (IoT) architecture. This system addresses the problems mentioned in the background art, such as the complexity of construction site environments and the diversity of construction scenarios, which prevent the integration of multi-source data from different building scenarios, thus affecting the effectiveness of building monitoring and control, and making it difficult to accurately identify and handle anomalies based on scenario data.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a video surveillance system for multi-source data fusion of building sites under an Internet of Things (IoT) architecture, comprising a data acquisition module, a data acquisition scene recognition module, a data fusion module, and an application management module. The data acquisition module is used to collect real-time scene data of multiple building monitoring scenarios from various sensors and video devices. The scene data consists of video data, environmental data, and personnel positioning data. The scene recognition module is used to receive scene data and sequentially perform scene feature extraction and scene classification recognition. The data fusion module uses a weighted average fusion algorithm and a Kalman filter algorithm to process and analyze scene data from different building monitoring scenarios to obtain a multi-source data fusion set. The data fusion module sets corresponding data thresholds for different scene data according to the type of building monitoring scenario. The data fusion module performs early warning judgment on the multi-source data fusion set based on multiple data thresholds. The application management module is used to issue early warnings for abnormal multi-source data fusion sets through light alarms and sound alarms, and to display them through a visual platform.

[0006] Preferably, the data acquisition module and the acquisition scene recognition module constitute a monitoring terminal mechanism, which is composed of a video data unit, an environmental data unit, a personnel positioning data unit, and a hardware unit.

[0007] Preferably, the hardware unit comprises a protective housing, a support component, a drive component, and a rotating component. The protective housing is fixed within multiple building monitoring scenarios via the support component. The rotating component and the drive component are built into the protective housing, and the rotating component reciprocates within the protective housing via the drive component.

[0008] Preferably, the video data unit includes a camera module, which consists of a rotating base, multiple high-definition cameras, an infrared camera, and multiple fill lights; the environmental data unit consists of a temperature and humidity sensor, a light sensor, and a smoke sensor; and the personnel positioning data unit consists of RFID and Bluetooth beacons.

[0009] Preferably, the video data unit is built into the rotating component, and the environmental data unit and the personnel positioning data unit are fixed to the outer surface of the protective housing. The environmental data unit and the personnel positioning data unit are used to collect environmental data and personnel positioning data.

[0010] Preferably, the scene feature extraction involves extracting color histograms, texture features, and edge information from scene data, and using a convolutional neural network to extract high-level semantic features. The scene classification and recognition utilizes a support vector machine and a deep learning classifier to perform scene recognition and output scene labels and confidence scores.

[0011] Preferably, the data fusion module sequentially obtains the multi-source data fusion set through a weighted average fusion algorithm, a Kalman filter algorithm, and threshold setting and anomaly detection, wherein the weighted average fusion algorithm is FusedDaTa; ; in, .

[0012] Preferably, the Kalman filtering algorithm includes state prediction, error covariance prediction, Kalman gain calculation, and error covariance update; ; in, Let k be the prior state estimate at time k, which is the current state predicted based on the optimal estimate at the previous time. To be modeled as uniform motion, To control the input matrix and represent the effect of external control on the state, To control the input vector; ; in, This represents the covariance and the uncertainty in predicting the state; a larger value indicates a less reliable prediction. This represents the posterior covariance of the previous time step and reflects the reliability of historical estimates. The process noise covariance matrix is ​​used to quantify the inherent uncertainty of the system model; ; in, The Kalman gain dynamically determines the fusion weights of the predicted and observed values. To observe noise; ; in, The updated posterior covariance reflects the remaining uncertainty of the fused state, and I is the identity matrix to ensure that the covariance matrix has consistent dimensions.

[0013] Preferably, the threshold setting and anomaly detection consist of a static threshold, a dynamic threshold, and detection. ; in, The mean represents the central tendency of the dataset. Standard deviation measures the dispersion of data. The multiplier coefficient determines the leniency of the threshold; when k=2, it covers 95% of the data, and when k=3, it covers 99.7% of the data. ; in, This is a time-dependent threshold that adaptively adjusts as real-time data changes. The smoothing coefficient controls the rate of change of the threshold. This data is from the previous time step and is used to dynamically adjust the threshold. ; in, For the merged multi-source data, The current threshold, This is the tolerance factor. As an early warning signal, when , , hour, This triggers an alarm.

[0014] Preferably, the application management module consists of an early warning platform and a visual platform. The visual platform consists of multiple displays, and the early warning platform includes multiple light alarms and multiple sound alarms. The light alarms display different levels of abnormality through color and flashing frequency, and the sound alarms provide alerts by emitting sounds of different frequencies.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention determines the current monitoring area type through a scene recognition module, and then the data acquisition module synchronously acquires multi-source data in that scene, including video streams, environmental sensor data, and personnel positioning information. The acquired raw data is then cleaned, denoised, and formatted to normalize data of different dimensions to a unified numerical range. A weighted average algorithm is used to integrate similar data, and Kalman filtering is used to predict and correct dynamic data, generating a more accurate and reliable fusion dataset, effectively improving monitoring accuracy and system robustness. 2. This invention sets static and dynamic thresholds based on historical data, and identifies abnormal behavior by comparing the fused data with the thresholds. This enables intelligent early warning and judgment of abnormal events such as intrusion, fire, and equipment failure. The application management module visualizes the processing results through a multi-display system, and triggers corresponding light and sound alarm devices according to the level of abnormality, realizing real-time monitoring and emergency response through human-computer interaction. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a flowchart of the IoT building video surveillance and control system of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Please see Figure 1 The present invention provides an embodiment of a video monitoring system for building sites based on the Internet of Things architecture, comprising a data acquisition module, a scene recognition module, a data fusion module, and an application management module. Through the data acquisition module, scene recognition module, data fusion module, and application management module, scene recognition and data acquisition, data preprocessing and standardization, multi-source data fusion processing, anomaly detection and threshold discrimination, and visualization display and alarm response can be achieved.

[0019] The data acquisition module is used to collect real-time scene data of multiple building monitoring scenarios from various sensors and video devices. The scene data consists of video data, environmental data, and personnel positioning data. The scene recognition module is used to receive the scene data and perform scene feature extraction and scene classification recognition in sequence. Scene feature extraction involves extracting color histograms, texture features, and edge information from the scene data, and using convolutional neural networks to extract high-level semantic features. Scene classification recognition uses support vector machines and deep learning classifiers to perform scene recognition and output scene labels and confidence scores. The monitoring terminal mechanism consists of a data acquisition module and a scene recognition module. The monitoring terminal mechanism is composed of a video data unit, an environmental data unit, a personnel positioning data unit, and a hardware unit. The hardware unit consists of a protective shell, a support component, a drive component, and a rotating component. The protective shell is fixed in multiple building monitoring scenes by the support component. The rotating component and the drive component are built into the protective shell. The rotating component swings back and forth inside the protective shell by the drive component. The hardware unit can provide hardware support during the scene data acquisition process.

[0020] The video data unit includes a camera module, which consists of a rotating base, multiple high-definition cameras, an infrared camera, and multiple supplementary lights. The environmental data unit consists of a temperature and humidity sensor, a light sensor, and a smoke sensor. The personnel positioning data unit consists of RFID and Bluetooth beacons. The video data unit is built into the rotating assembly. The environmental data unit and the personnel positioning data unit are fixed to the outer surface of the protective housing. The environmental data unit and the personnel positioning data unit are used to collect environmental data and personnel positioning data.

[0021] The data fusion module uses a weighted average fusion algorithm and a Kalman filter algorithm to process and analyze scene data from different building monitoring scenarios to obtain a multi-source data fusion set. The data fusion module sets corresponding data thresholds for different scene data according to the type of building monitoring scenario. The data fusion module performs early warning judgment on the multi-source data fusion set based on multiple data thresholds. The data fusion module obtains the multi-source data fusion set by sequentially using a weighted average fusion algorithm, a Kalman filter algorithm, and threshold setting and anomaly detection. The weighted average fusion algorithm is FusedDaTa. ; in, .

[0022] The Kalman filter algorithm includes state prediction, error covariance prediction, Kalman gain calculation, and error covariance update. ; in, Let k be the prior state estimate at time k, which is the current state predicted based on the optimal estimate at the previous time. To be modeled as uniform motion, To control the input matrix and represent the effect of external control on the state, To control the input vector; ; in, This represents the covariance and the uncertainty in predicting the state; a larger value indicates a less reliable prediction. This represents the posterior covariance of the previous time step and reflects the reliability of historical estimates. The process noise covariance matrix is ​​used to quantify the inherent uncertainty of the system model; ; in, The Kalman gain dynamically determines the fusion weights of the predicted and observed values. To observe noise; ; in, The updated posterior covariance reflects the remaining uncertainty of the fused state, and I is the identity matrix to ensure that the covariance matrix has consistent dimensions.

[0023] Threshold setting and anomaly detection consist of static thresholds, dynamic thresholds, and detection methods. ; in, The mean represents the central tendency of the dataset. Standard deviation measures the dispersion of data. The multiplier coefficient determines the leniency of the threshold; when k=2, it covers 95% of the data, and when k=3, it covers 99.7% of the data. ; in, This is a time-dependent threshold that adaptively adjusts as real-time data changes. The smoothing coefficient controls the rate of change of the threshold. This data is from the previous time step and is used to dynamically adjust the threshold. ; in, For the merged multi-source data, The current threshold, This is the tolerance factor. As an early warning signal, when , , hour, This triggers an alarm.

[0024] The application management module is used to issue early warnings for abnormal multi-source data fusion sets through light alarms and sound alarms, and to display them through a visual platform. The application management module consists of an early warning platform and a visual platform. The visual platform consists of multiple displays, and the early warning platform includes multiple light alarms and multiple sound alarms. The light alarms display different levels of abnormality through color and flashing frequency, and the sound alarms provide alarm prompts by emitting sounds of different frequencies.

[0025] Please see Figure 2 In summary, the current monitoring area type is first determined by the scene recognition module, and then the data acquisition module synchronously acquires multi-source data in this scene, including video streams, environmental sensor data, and personnel positioning information. The acquired raw data is then cleaned, denoised, and formatted to normalize data of different dimensions to a unified numerical range, preparing for subsequent fusion processing. A weighted average algorithm is used to integrate similar data, and Kalman filtering is used to predict and correct dynamic data, generating a more accurate and reliable fusion dataset, effectively improving monitoring accuracy. Based on historical data, static and dynamic thresholds are set. By comparing the fused data with the thresholds, abnormal behavior is identified, enabling intelligent early warning and judgment of abnormal events such as intrusion, fire, and equipment failure. The application management module visualizes the processing results through a multi-monitor system. At the same time, it triggers corresponding light and sound alarm devices according to the level of abnormality, realizing real-time monitoring and emergency response through human-computer interaction.

[0026] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A video surveillance system for building sites based on IoT architecture, incorporating multi-source data fusion, including a data acquisition module, a scene recognition module, a data fusion module, and an application management module: The data acquisition module is used to collect real-time scene data of multiple building monitoring scenarios from various sensors and video devices. The scene data consists of video data, environmental data, and personnel positioning data. The scene recognition module is used to receive scene data and sequentially perform scene feature extraction and scene classification recognition. The data fusion module uses a weighted average fusion algorithm and a Kalman filter algorithm to process and analyze scene data from different building monitoring scenarios to obtain a multi-source data fusion set. The data fusion module sets corresponding data thresholds for different scene data according to the type of building monitoring scenario. The data fusion module performs early warning judgment on the multi-source data fusion set based on multiple data thresholds. The application management module is used to issue early warnings for abnormal multi-source data fusion sets through light alarms and sound alarms, and to display them through a visual platform.

2. The video surveillance system for building sites based on the Internet of Things architecture according to claim 1, characterized in that: The data acquisition module and the scene recognition module together form the monitoring terminal mechanism, which consists of a video data unit, an environmental data unit, a personnel positioning data unit, and a hardware unit.

3. The video surveillance system for building sites based on the Internet of Things architecture according to claim 2, characterized in that: The hardware unit comprises a protective housing, a support component, a drive component, and a rotating component. The protective housing is fixed within multiple building monitoring scenarios via the support component. The rotating component and the drive component are built into the protective housing, and the rotating component reciprocates within the protective housing via the drive component.

4. The video surveillance system for building site multi-source data fusion under the Internet of Things architecture according to claim 3, characterized in that: The video data unit includes a camera module, which consists of a rotating base, multiple high-definition cameras, an infrared camera, and multiple fill lights. The environmental data unit consists of a temperature and humidity sensor, a light sensor, and a smoke sensor. The personnel positioning data unit consists of RFID and Bluetooth beacons.

5. The video surveillance system for building site multi-source data fusion under the Internet of Things architecture according to claim 4, characterized in that: The video data unit is built into the rotating component, and the environmental data unit and the personnel positioning data unit are fixed on the outer surface of the protective shell. The environmental data unit and the personnel positioning data unit are used to collect environmental data and personnel positioning data.

6. The video surveillance system for building site multi-source data fusion under the Internet of Things architecture according to claim 5, characterized in that: The scene feature extraction involves extracting color histograms, texture features, and edge information from scene data, and using convolutional neural networks to extract high-level semantic features. The scene classification and recognition utilizes support vector machines and deep learning classifiers to perform scene recognition and output scene labels and confidence scores.

7. The video surveillance system for building site multi-source data fusion under the Internet of Things architecture according to claim 1, characterized in that: The data fusion module sequentially obtains the multi-source data fusion set through a weighted average fusion algorithm, a Kalman filter algorithm, and threshold setting and anomaly detection. The weighted average fusion algorithm is FusedDaTa. ; in, .

8. The video surveillance system for building site multi-source data fusion under the Internet of Things architecture according to claim 7, characterized in that: The Kalman filtering algorithm includes state prediction, error covariance prediction, Kalman gain calculation, and error covariance update. ； in, Let k be the prior state estimate at time k, which is the current state predicted based on the optimal estimate at the previous time. To be modeled as uniform motion, To control the input matrix and represent the effect of external control on the state, To control the input vector; ； in, This represents the covariance and the uncertainty in predicting the state; a larger value indicates a less reliable prediction. This represents the posterior covariance of the previous time step and reflects the reliability of historical estimates. The process noise covariance matrix is ​​used to quantify the inherent uncertainty of the system model; ； in, The Kalman gain dynamically determines the fusion weights of the predicted and observed values. To observe noise; ； in, The updated posterior covariance reflects the remaining uncertainty of the fused state, and I is the identity matrix to ensure that the covariance matrix has consistent dimensions.

9. The video surveillance system for building site multi-source data fusion under the Internet of Things architecture according to claim 8, characterized in that: The threshold setting and anomaly detection consist of static thresholds, dynamic thresholds, and detection methods. ; in, The mean represents the central tendency of the dataset. Standard deviation measures the dispersion of data. The multiplier coefficient determines the leniency of the threshold; when k=2, it covers 95% of the data, and when k=3, it covers 99.7% of the data. ； in, This is a time-dependent threshold that adaptively adjusts as real-time data changes. The smoothing coefficient controls the rate of change of the threshold. This data is from the previous time step and is used to dynamically adjust the threshold. ； in, For the merged multi-source data, The current threshold, This is the tolerance factor. As an early warning signal, when , , hour, This triggers an alarm.

10. The video surveillance system for building site multi-source data fusion under the Internet of Things architecture according to claim 9, characterized in that: The application management module consists of an early warning platform and a visual platform. The visual platform consists of multiple displays, and the early warning platform includes multiple light alarms and multiple sound alarms. The light alarms display different levels of abnormality through color and flashing frequency, and the sound alarms provide alerts by emitting sounds of different frequencies.

Images