An embedded monitoring abnormality recognition system based on deep learning

Abstract

This invention belongs to the field of intelligent monitoring technology, specifically a deep learning-based embedded monitoring anomaly recognition system. It includes a device housing and a video acquisition module, a storage module, and a power supply module housed within the housing. The video acquisition module is used to acquire video footage from security scenarios. Through a hardware structure design integrating a local AI inference chip, encrypted storage, adaptive network switching, tiered audible and visual alarms, expansion interfaces, dual power supply, and chip heat dissipation, it achieves complete offline security functions, including all-weather high-precision intelligent recognition under network outages and weak network conditions, tiered on-site early warning, local encrypted evidence storage, flexible multi-device linkage, and continuous monitoring even after power failure. This solves the problems of existing traditional embedded devices lacking dedicated AI inference hardware, unable to implement deep learning models for high-precision real-time analysis, and exhibiting poor system intelligence, integration, and stability, making it difficult to meet the demands of high-precision, low-latency on-site intelligent monitoring.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology, specifically an embedded monitoring anomaly identification system based on deep learning. Background Technology

[0002] Currently, video surveillance is widely used in security scenarios in parks, factories, buildings, and public areas. Existing surveillance systems are mainly divided into two categories: manual monitoring and cloud-based intelligent monitoring. Traditional manual monitoring is characterized by high labor costs, operator fatigue leading to missed detections, and delayed response times, failing to meet the needs of 24 / 7 automated security monitoring.

[0003] Cloud-based intelligent monitoring relies on the network to upload video data to the cloud for analysis. It is highly dependent on bandwidth and cannot work properly in environments with network outages or weak networks. It also suffers from high analysis latency and the inability to provide real-time warnings. At the same time, the remote transmission of large amounts of video data poses risks of privacy leaks and data tampering. Furthermore, cloud deployment and maintenance costs are high and the flexibility for expansion is poor.

[0004] Existing traditional embedded devices lack dedicated AI inference hardware, making it impossible to carry deep learning models to achieve local high-precision real-time analysis. The system has poor intelligence, integration and stability, making it difficult to meet the needs of high-precision, low-latency on-site intelligent monitoring.

[0005] Therefore, the present invention provides an embedded monitoring anomaly identification system based on deep learning. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is as follows: An embedded monitoring anomaly identification system based on deep learning, comprising a device housing and a video acquisition module, a storage module, and a power supply module disposed within the device housing. The video acquisition module is used to acquire on-site video footage of security scenarios. The device housing integrates a dedicated AI inference hardware chip and a local processing main control module. The dedicated AI inference hardware chip, video acquisition module, and storage module are all electrically connected to the local processing main control module. The local processing main control module has a built-in offline deep learning inference model, which can perform local AI analysis and anomaly identification on the on-site video footage without cloud network transmission. The local processing main control module is equipped with a local real-time early warning unit, which, in conjunction with an audible and visual alarm component, can output early warning signals on-site in real time, realizing all-weather local intelligent monitoring in environments with network outages or weak network conditions.

[0008] Preferably, the dedicated AI inference hardware chip is an onboard embedded AI acceleration chip, and the device housing is provided with a heat dissipation structure adapted to the dedicated AI inference hardware chip. The heat dissipation structure includes a heat-conducting copper sheet attached to the chip and a miniature cooling fan arranged opposite the heat-conducting copper sheet.

[0009] Preferably, the device housing integrates a data encryption local storage unit, which is electrically connected to the storage module and the local processing main control module respectively, for local encrypted storage of video data and analysis logs, eliminating the risk of data leakage and tampering caused by remote transmission.

[0010] Preferably, the local processing main control module integrates a network adaptive switching unit, which is connected to the cloud communication module and the local offline processing unit respectively, and can automatically switch between cloud data upload or local full-process analysis, storage and early warning mode according to the network status.

[0011] Preferably, the audible and visual alarm component is embedded in the top of the device housing, including a high-brightness warning light and a high-frequency buzzer, and is linked with a local real-time early warning unit, which can match the corresponding alarm intensity according to the level of abnormal risk.

[0012] Preferably, the side wall of the device housing is provided with a modular expansion interface group, which includes an alarm linkage interface, a peripheral expansion interface and a data docking interface, enabling external security peripherals and linkage with security systems, thereby improving device expandability and reducing cloud operation and maintenance costs.

[0013] Preferably, the power supply module includes an AC power supply unit and a lithium battery energy storage unit. The lithium battery energy storage unit is electrically connected to the local processing main control module and can independently supply power in the event of a power outage, ensuring stable operation of the equipment around the clock.

[0014] The beneficial effects of this invention are as follows:

[0015] The embedded monitoring anomaly identification system based on deep learning described in this invention achieves complete offline security functions through a hardware structure design that integrates a local AI inference chip, encrypted storage, adaptive network switching, tiered audible and visual alarms, expansion interfaces, dual power supply, and chip heat dissipation. This includes all-weather local high-precision intelligent identification in the event of network outages or weak networks, tiered on-site early warning, local encrypted evidence storage, flexible multi-device linkage, and continuous monitoring even after power failure. It effectively solves the problems of existing traditional embedded devices lacking dedicated AI inference hardware, being unable to carry deep learning models to achieve local high-precision real-time analysis, having poor system intelligence, integration, and stability, and being unable to meet the requirements of high-precision, low-latency on-site intelligent monitoring. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Specific implementation examples are given below.

[0020] like Figure 1 As shown in the figure, an embedded monitoring anomaly identification system based on deep learning according to an embodiment of the present invention includes a device housing and a video acquisition module, a storage module and a power supply module disposed within the device housing. The video acquisition module is used to acquire on-site video images of security scenarios. The device housing integrates a dedicated AI inference hardware chip and a local processing main control module. The dedicated AI inference hardware chip, the video acquisition module and the storage module are all electrically connected to the local processing main control module.

[0021] The local processing main control module has a built-in offline deep learning inference model, which can perform local AI analysis and anomaly recognition on on-site video footage without cloud network transmission. The local processing main control module is equipped with a local real-time early warning unit, which is linked with the sound and light alarm components to output early warning signals on-site in real time, realizing all-weather local intelligent monitoring in network outage and weak network environments. When in operation, this system is suitable for security monitoring in all scenarios such as parks, factories, buildings, roads, and squares. It relies on the dedicated AI inference hardware integrated in the shell to perform offline deep learning operations with the local main control, without continuously pushing video streams to the cloud. In the face of scenarios such as unattended nighttime, network failure, and insufficient broadband bandwidth, it can complete the identification of anomalies such as human figures, climbing over, area intrusion, and left-behind objects on-site. After identification, it directly triggers local early warning, solving the shortcomings of traditional manual monitoring that misses detection and lags, cloud monitoring that depends on the network, has poor real-time performance, and is prone to privacy data leakage, realizing 24 / 7 uninterrupted autonomous security monitoring.

[0022] The dedicated AI inference hardware chip is an onboard embedded AI acceleration chip. The device housing has a heat dissipation structure adapted to the dedicated AI inference hardware chip. The heat dissipation structure includes a thermally conductive copper sheet attached to the chip and a miniature cooling fan positioned opposite the thermally conductive copper sheet. During operation, the onboard embedded AI acceleration chip continuously generates a large amount of heat when running deep learning models. In scenarios with poor heat dissipation conditions, such as high-temperature server rooms in factories, outdoor cabinets, and enclosed equipment housings, chip frequency reduction, recognition lag, and device crashes are likely to occur. This heat dissipation structure quickly conducts away the core heat source by tightly attaching the thermally conductive copper sheet to the chip, combined with directional air cooling by the miniature cooling fan, stabilizing the main frequency of the AI ​​inference hardware, ensuring the accuracy and processing speed of local video analysis in high-temperature environments, and extending the long-term outdoor working life of the embedded monitoring equipment.

[0023] The device housing integrates a local encrypted storage unit, which is electrically connected to both the storage module and the local processing control module. This unit is used for locally encrypted storage of video data and analysis logs, eliminating the risk of data leakage and tampering associated with remote transmission. During operation, surveillance videos from parks, residential areas, and public roads involve sensitive privacy data such as facial features and movement trajectories. Traditional cloud-based transmission and storage are prone to theft, tampering, and leakage. This encrypted storage unit encrypts all captured video and AI anomaly recognition logs locally before writing them to the storage module, ensuring the data is never transmitted externally. It is suitable for scenarios with stringent data confidentiality requirements, such as schools, hospitals, and factory areas, mitigating the security risks of video data leakage and malicious tampering at the hardware level.

[0024] The local processing main control module integrates a network adaptive switching unit, which connects to both the cloud communication module and the local offline processing unit. It can automatically switch between cloud data upload and local full-process analysis, storage, and early warning modes based on network conditions. During operation, for areas with unstable network signals, such as remote locations within the factory, underground parking garages, and suburban parks, the network adaptive switching unit automatically detects uplink bandwidth in real time. When the network is normal, it only uploads simplified abnormal alarm images and text to the cloud for aggregation and management, saving bandwidth. In the event of a network outage or weak network, it immediately switches to a pure local offline working mode, fully retaining the full-link functions of video acquisition, AI recognition, encrypted storage, and on-site alarms. This balances centralized management needs with independent offline security capabilities, adapting to complex network security projects.

[0025] The audible and visual alarm component is embedded in the top of the equipment housing, including a high-brightness warning light and a high-frequency buzzer, and is linked to a local real-time early warning unit. It can match the corresponding alarm intensity according to the level of abnormal risk. When in operation, the audible and visual alarm component is placed on the top of the housing so that gatekeepers and building security personnel can quickly detect it. It distinguishes the multi-level risk alarm logic: low-risk boundary crossing uses low-frequency slow flashing light + low-volume buzzer, while high-risk climbing over or intrusion uses high-frequency strobe light + high-decibel buzzer. It is suitable for scenarios such as warehouses, walls, and parking lots that require on-site visual warning. The graded alarm helps security personnel quickly determine the level of danger, reduce ineffective manual patrols, and improve emergency response efficiency.

[0026] The equipment housing features a modular expansion interface group on its sidewalls. This group includes alarm linkage interfaces, peripheral expansion interfaces, and data interface interfaces, allowing for connection to external security peripherals and linkage with security systems. This enhances the equipment's scalability and reduces cloud-based maintenance costs. During operation, the modular expansion interface group integrates multiple types of universal connectors, adapting to multi-device linkage scenarios in large parks. The alarm linkage interface can connect to turnstiles, barriers, and access control systems; the peripheral expansion interface can connect to external microphones and infrared detectors; and the data interface can connect to a local NVR security host. Security equipment can be expanded without replacing the entire unit. During phased project upgrades or factory expansions, there is no need to repurchase a complete cloud monitoring system, significantly reducing project deployment and subsequent maintenance costs and offering greater flexibility for expansion.

[0027] The power supply module includes a mains power supply unit and a lithium battery energy storage unit. The lithium battery energy storage unit is electrically connected to the local processing main control module and can independently supply power in the event of a power outage, ensuring stable operation of the equipment around the clock. During operation, in scenarios such as power outages in the factory area, maintenance of municipal power lines, or failures in the power supply lines for outdoor monitoring, conventional monitoring equipment will shut down directly, creating a security monitoring vacuum. The power supply module features a dual-power supply design. When the mains power is normal, it provides the main power supply and supplements the built-in lithium battery. In the event of a power outage, it seamlessly switches to independent power supply from the lithium battery, ensuring continuous operation of AI recognition, video recording and storage, and audible and visual alarm functions, meeting the uninterrupted security monitoring needs of areas with unstable power supply such as factories, industrial parks, and old residential areas.

[0028] Working principle: This system is applicable to security monitoring in all scenarios such as parks, factories, buildings, roads, and squares. It relies on dedicated AI inference hardware integrated in the shell to achieve offline deep learning operations with local main control, without the need to continuously push video streams to the cloud. In the face of scenarios such as unattended nighttime monitoring, network failure, and insufficient broadband bandwidth, it can complete the identification of anomalies such as human figures, climbing over walls, area intrusion, and left-behind objects on-site. After identification, it directly triggers local warnings, solving the shortcomings of traditional manual monitoring that misses detections and lags, cloud monitoring that depends on the network, has poor real-time performance, and is prone to privacy data leakage, thus achieving 24 / 7 uninterrupted autonomous security monitoring.

[0029] When the onboard embedded AI acceleration chip runs deep learning models, it continuously generates a lot of heat. In scenarios with poor heat dissipation conditions, such as high-temperature computer rooms in factories, outdoor cabinets, and enclosed equipment housings, chip frequency reduction, recognition lag, and equipment crashes are likely to occur. This heat dissipation structure uses thermally conductive copper sheets to closely adhere to the chip and quickly conduct away the core heat source. Combined with a miniature cooling fan for directional air cooling, it stabilizes the main frequency of the AI ​​inference hardware, ensures the accuracy and processing speed of local video analysis in high-temperature environments, and extends the long-term outdoor working life of the embedded monitoring equipment.

[0030] Surveillance videos from parks, residential areas, and public roads contain sensitive and private data such as people's faces and movement trajectories. Traditional cloud transmission and storage are prone to theft, tampering, and leakage. This encrypted storage unit encrypts all collected videos and AI anomaly recognition logs locally and writes them to the storage module. The data is never transmitted outside the storage module. It is suitable for scenarios such as schools, hospitals, and factory areas where data confidentiality compliance is required. It avoids the security risks of video data leakage and malicious tampering of recordings from the hardware level.

[0031] For areas with unstable network signals, such as remote locations in factories, underground parking garages, and suburban industrial parks, the network adaptive switching unit automatically detects uplink bandwidth in real time. When the network is normal, it only uploads simplified abnormal alarm images and texts to the cloud for summary management, saving bandwidth. In the event of a network outage or weak network, it immediately switches to a pure local offline working mode, fully retaining the full-link functions of video acquisition, AI recognition, encrypted storage, and on-site alarms, taking into account both centralized management needs and offline independent security capabilities, and adapting to complex network security projects.

[0032] The sound and light alarm components are positioned on the top of the housing for easy detection by factory gatekeepers and building security personnel. They differentiate between multiple risk alarm logics: low-risk boundary crossings use low-frequency slow flashing lights and low-volume buzzers, while high-risk climbing over or intrusions use high-frequency strobe lights and high-decibel buzzers. They are suitable for scenarios requiring on-site visual warnings, such as warehouses, walls, and parking lots. The graded alarms help security personnel quickly determine the level of danger, reduce ineffective manual patrols, and improve emergency response efficiency.

[0033] The modular expansion interface group integrates multiple types of universal docking terminals, adapting to multi-device linkage scenarios in large parks. The alarm linkage interface can be connected to turnstiles, barriers, and access control systems, while the peripheral expansion interface can connect to external microphones and infrared detectors. The data docking interface can be connected to a local NVR security host. Security equipment can be expanded without replacing the entire unit. When projects are upgraded in phases or factories are expanded, there is no need to repurchase a complete set of cloud monitoring equipment, which greatly reduces project deployment and subsequent operation and maintenance costs and provides greater flexibility for expansion.

[0034] In scenarios such as power outages in factory areas, municipal line maintenance, and power supply line failures for outdoor monitoring, conventional monitoring equipment shuts down directly, creating a security monitoring vacuum. The power supply module features a dual-power supply design, providing primary power and supplementing the built-in lithium battery when the mains power is normal. In the event of a power outage, it seamlessly switches to independent power supply from the lithium battery, ensuring continuous operation of AI recognition, video recording and storage, and audible and visual alarm functions. This meets the uninterrupted security monitoring needs of areas with unstable power supply, such as factories, industrial parks, and old residential areas.

[0035] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A deep learning-based embedded monitoring anomaly identification system, comprising a device housing and a video acquisition module, a storage module, and a power supply module disposed within the device housing, wherein the video acquisition module is used to acquire video footage from a security scene, characterized in that: The device housing integrates a dedicated AI inference hardware chip and a local processing main control module. The dedicated AI inference hardware chip, video acquisition module, and storage module are all electrically connected to the local processing main control module. The local processing main control module has a built-in offline deep learning inference model, which can perform local AI analysis and anomaly recognition on the on-site video footage without cloud network transmission. The local processing main control module is equipped with a local real-time early warning unit, which is linked with the sound and light alarm component to output early warning signals on-site in real time, realizing all-weather local intelligent monitoring in network outage and weak network environments.

2. The embedded monitoring anomaly identification system based on deep learning according to claim 1, characterized in that: The dedicated AI inference hardware chip is an onboard embedded AI acceleration chip. The device housing is equipped with a heat dissipation structure adapted to the dedicated AI inference hardware chip. The heat dissipation structure includes a heat-conducting copper sheet attached to the chip and a miniature cooling fan arranged opposite the heat-conducting copper sheet.

3. The embedded monitoring anomaly identification system based on deep learning according to claim 2, characterized in that: The device housing integrates a local encrypted data storage unit, which is electrically connected to the storage module and the local processing main control module. It is used for local encrypted storage of video data and analysis logs, eliminating the risk of data leakage and tampering caused by remote transmission.

4. The embedded monitoring anomaly identification system based on deep learning according to claim 3, characterized in that: The local processing main control module integrates a network adaptive switching unit, which is connected to the cloud communication module and the local offline processing unit respectively. It can automatically switch between cloud data upload or local full-process analysis, storage, and early warning modes according to the network status.

5. The embedded monitoring anomaly identification system based on deep learning according to claim 1, characterized in that: The audible and visual alarm component is embedded in the top of the device housing and includes a high-brightness warning light and a high-frequency buzzer. It is linked with the local real-time early warning unit and can match the corresponding alarm intensity according to the level of abnormal risk.

6. The embedded monitoring anomaly identification system based on deep learning according to claim 5, characterized in that: The device housing has a modular expansion interface group on its side wall. The modular expansion interface group includes an alarm linkage interface, a peripheral expansion interface and a data docking interface, which can connect to external security peripherals and link with security systems to improve the device's expandability and reduce cloud operation and maintenance costs.

7. The embedded monitoring anomaly identification system based on deep learning according to claim 1, characterized in that: The power supply module includes an AC power supply unit and a lithium battery energy storage unit. The lithium battery energy storage unit is electrically connected to the local processing main control module and can independently supply power in the event of a power outage, ensuring stable operation of the equipment around the clock.

Images