Dual-layer intelligent discrimination method and system based on ultra-low power aov mode and event triggering

Abstract

The application discloses a kind of based on ultra-low power consumption AOV mode and event trigger double-layer intelligent discrimination method and system, belong to industrial vision intelligent monitoring technical field, including steps, construct by image sensor, millimeter wave radar, multi-in-one environmental sensor and intelligent light supplement unit composition perception layer, configuration image is AOV low frame rate low resolution normal mode, radar and sensor intermittent sampling, low-power operation;Image, radar and environmental data are collected, and are detected by AOV vision and radar, identify effective moving target and generate trigger signal;Response trigger and wake up main control SoC and switch high-definition mode, read illumination, hierarchical control intelligent light supplement;S4 calls NPU to execute double-layer discrimination, AOV fast rough screening, lightweight AI accurate identification;Determine event, generate alarm upload, complete and restore AOV normal;And monitor electric quantity, when low electric quantity, enter extreme value guard, close internal trigger, retain external response.The practicability and reliability of the application are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology for industrial vision products, specifically relating to a dual-layer intelligent discrimination method and system based on ultra-low power AOV mode and event triggering. Background Technology

[0002] In applications such as railway safety monitoring, traditional monitoring equipment typically requires continuous power, resulting in high energy consumption and difficulty in achieving long-term autonomous operation in areas without mains power. While some low-power vision modules exist, most lack intelligent event triggering mechanisms, rendering them ineffective in low-light conditions.

[0003] Current safety monitoring technologies in railway lines and other "no mains power" scenarios face two major challenges: "continuous high energy consumption" and "insufficient intelligent sensing capabilities." On one hand, while high-power continuous monitoring solutions (such as traditional network cameras / IPCs) can achieve 24 / 7 monitoring, their "uninterrupted video stream acquisition and transmission" mode results in power consumption often exceeding 10W. In outdoor environments entirely reliant on solar power, a significant gap exists between limited energy collection capabilities and the high power consumption of the equipment, making it difficult for the system to achieve long-term stable operation during continuous rainy weather and generating massive amounts of invalid data, wasting storage and communication resources. On the other hand, simple low-power triggering solutions (such as using PIR sensors or basic visual motion detection (VMD)) attempt to reduce power consumption through "event triggering," but they have fundamental flaws: PIR (passive infrared) can only sense heat source movement, cannot distinguish between people, vehicles, and animals, is completely ineffective against stationary targets (such as objects left on the tracks), has poor environmental adaptability, and a high false alarm rate. VMD (motion detection) algorithms are simple but easily affected by sudden changes in light, weather, and swaying leaves, resulting in low practicality. Furthermore, VMD cannot function due to poor image quality. Although supplemental lighting measures are in place, the control logic is simple and not linked to intelligent discrimination, often resulting in energy waste.

[0004] Meanwhile, existing solutions suffer from a disconnect between perception and decision-making. Trigger signals require waking up the high-power main system or returning to the cloud for judgment, resulting in high response latency and potential missed detections when the network is poor. The supplementary lighting strategy is based solely on a simple illuminance threshold and is not linked to the vision task, leading to significant energy waste. Furthermore, the lack of a system-level ultra-low-power architecture makes it impossible to balance perception, intelligence, and reliability under extremely low power consumption, which is insufficient to meet the needs of long-term unattended industrial scenarios. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a dual-layer intelligent discrimination method and system based on ultra-low power AOV mode and event triggering. It features low normal power consumption, dual-source triggering to reduce false alarms and missed alarms, local decision-making at the edge, and hierarchical supplementary lighting. It is suitable for long-term stable operation in outdoor scenarios without mains power, and its practicality and reliability are significantly improved.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0007] In a first aspect, the present invention provides a two-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering, comprising the following steps:

[0008] Step S1: Construct a perception layer consisting of an image sensor, millimeter-wave radar, an all-in-one environmental sensor, and an intelligent supplementary lighting unit. Configure the image sensor to operate in AOV low frame rate and low resolution normal mode, while simultaneously enabling the millimeter-wave radar and environmental sensor to operate in an intermittent sampling low-power mode.

[0009] Step S2: Real-time acquisition of environmental image data, radar detection data and environmental parameter data through the perception layer; joint detection based on AOV vision algorithm and radar detection results; identification of significant pixel changes or effective moving targets that meet preset conditions; and generation of event trigger signals.

[0010] Step S3: Respond to the event trigger signal, wake up the main control SoC in sleep mode and switch the image sensor to high resolution high definition capture mode, read the ambient illuminance data and dynamically control the start and stop of the intelligent supplement unit and pulse output according to the graded strategy;

[0011] Step S4: The main control SoC calls the NPU to execute a two-layer intelligent discrimination process. The first layer is a coarse screening discrimination based on the AOV low-power fast filtering algorithm, and the second layer is a high-definition image accurate recognition discrimination based on a lightweight AI model.

[0012] Step S5: Determine the event type based on the dual-layer intelligent discrimination result, generate structured alarm data and upload it through the main communication module. After the event is processed, the control system returns to the AOV ultra-low power normal mode.

[0013] Step S6: Monitor the battery level of the energy layer in real time. When the battery level is lower than the preset threshold, enter the extreme guard mode, shut down the internal trigger source and retain only the external trigger response capability.

[0014] Furthermore, in step S1, the AOV ultra-low power constant-view mode has a frame rate set to 1fps±0.2fps, a resolution set to VGA / QCIF level, an image sensor operating voltage reduced to 1.2V–1.5V, and standby power consumption controlled within 10mW; the millimeter-wave radar scanning period is set to 200ms–500ms, and the transmit power is reduced to **-20dBm~-10dBm**; the multi-functional environmental sensor sampling period is 5s–30s, and the single sampling duration does not exceed 100ms.

[0015] Furthermore, in step S2, based on the AOV pixel change threshold algorithm, there are three judgments: inter-frame difference threshold, regional change proportion threshold, and continuous frame change duration threshold. The inter-frame difference threshold is 8–16 gray levels, the regional change proportion threshold is 5%–15%, and the continuous frame change duration threshold is 3 frames or more. Combined with the radar joint judgment of valid moving events logic, the three conditions of distance 5–150m, speed 0.1–30m / s, and echo intensity ≥-60dB are met.

[0016] Furthermore, in step S3, the multi-level illumination data threshold is determined and divided into four levels: high illumination (>3000 Lux), medium illumination (10–3000 Lux), low illumination (0.1–10 Lux), and extremely low illumination (<0.1 Lux); the duration of instantaneous high-brightness supplementary lighting is 10–50 ms, the peak power is 1–3 W; and the frequency of pulsed intermittent supplementary lighting is 10–50 Hz, the duty cycle is 5%–20%; the adaptive dimming supplementary lighting dynamically adjusts the laser power according to the target distance, and sets the power to be higher the farther the target is.

[0017] Furthermore, in step S4, the lightweight AI model adopts a customized YOLOv5s / YOLOv8-tiny model with ≤5M parameters, ≤1TOPS computing power requirement, and ≤50ms inference time; the first-layer fast filtering algorithm adopts a dual mechanism of gray-scale mean difference and edge feature coarse screening, with a filtering accuracy of ≥90% and a single-frame time of ≤5ms; the second-layer accurate recognition supports five types of targets: personnel, vehicles, falling rocks, track foreign objects, and animals, and supports simultaneous detection of multiple targets with an accuracy of ≥95%.

[0018] Furthermore, in step S6, the low battery preset threshold is set to 15%±2%; in the extreme monitoring mode, the system operating current is ≤10mA and the power consumption is ≤50mW; it only responds to three types of external triggers: monitoring center instructions, adjacent device linkage signals, and preset emergency trigger signals. After triggering, it can be temporarily woken up to complete the capture and reporting, and immediately revert to the extreme monitoring mode after completion.

[0019] In a second aspect, the present invention provides a two-layer intelligent discrimination system based on ultra-low power AOV mode and event triggering, which can execute the two-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering as described in any one of the first aspects. The system includes a sensing layer, a processing layer, a communication layer and an energy layer.

[0020] The sensing layer includes a global shutter CMOS image sensor, a millimeter-wave radar module, an all-in-one environmental sensor, and a 940nm VCSEL intelligent fill light unit. The image sensor is configured with AOV low frame rate normal mode and high-definition capture trigger mode. The millimeter-wave radar and environmental sensor perform low-power intermittent sampling, and the intelligent fill light unit outputs through PWM control.

[0021] The processing layer includes a main control SoC with integrated NPU, an MCU power management unit, and a memory and storage unit. The MCU power management unit is responsible for system timing control and power consumption scheduling. The main control SoC responds to the trigger signal to wake up and execute dual-layer intelligent discrimination. The NPU runs a lightweight AI model to achieve accurate target recognition.

[0022] The communication layer includes an SFP optical module, a gigabit Ethernet interface, and a wireless backup module. The SFP optical module is the primary high-speed data transmission channel, and the backup module is adapted for network anomaly scenarios.

[0023] The energy layer includes solar panels, lithium iron phosphate battery packs, and an MPPT controller. The MPPT controller manages charging and power supply and provides tiered power supply support for the system.

[0024] Furthermore, the global shutter CMOS image sensor supports dynamic switching between AOV low-power mode and high-definition capture mode. In AOV mode, it outputs images with a resolution of 640×480 and below, and the operating current is ≤8mA; in high-definition mode, it outputs images with a resolution of 1920×1080 and above, and the operating current is ≤120mA. The sensor is connected to the main control SoC through MIPI-CSI / Ethernet dual interfaces, and supports frame synchronization triggering and mode switching hardware control.

[0025] The millimeter-wave radar module adopts the FMCW frequency-modulated continuous wave system, supports three-dimensional detection of distance, speed, and angle, with a detection range of 5–150m, an angle coverage of ±60°, and a refresh rate of 10–20Hz. It is connected to the MCU power management unit via UART / SPI, supports parameter configuration, data transmission, and sleep / wake-up hardware control, with sleep power consumption ≤1mW and operating power consumption ≤80mW.

[0026] Furthermore, the dual-layer intelligent discrimination is integrated into the NPU inside the main control SoC, including an AOV fast filtering submodule, an NPU accurate recognition submodule, a confidence verification submodule, and a timing consistency verification submodule. The AOV fast filtering submodule includes solidified inter-frame difference, edge detection, and region proportion triple filtering logic. The NPU accurate recognition submodule deploys a lightweight YOLO model and supports multi-objective parallel inference. The verification submodule performs confidence threshold filtering and continuous frame result comparison on the recognition results.

[0027] Furthermore, the MCU power management unit integrates a system-level power management module, a low-battery monitoring module, an extreme value protection control module, and a multi-level power supply control module. The system-level power management module is responsible for module sleep / wake-up timing, power on / off control, and dynamic voltage regulation. The low-battery monitoring module collects lithium battery SOC, voltage, and current in real time with an accuracy of ≤1%. When triggered, the extreme value protection control module cuts off power to the image sensor, radar, NPU, and main communication module, retaining only the MCU and basic communication monitoring circuitry. The multi-level power supply control module supports multiple voltage outputs of 1.2V / 1.8V / 3.3V / 5V.

[0028] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: The dual-layer intelligent discrimination method and system based on ultra-low power AOV mode and event triggering provided by this invention, by constructing a system-level ultra-low power architecture, reduces the normal power consumption to the level of 2.5W, which is significantly lower than the traditional 10W and above solutions, and can be adapted to long-term stable operation in solar-powered scenarios without mains power; by adopting a dual-source triggering mechanism of AOV and millimeter-wave radar, combined with triple algorithm filtering and three-dimensional radar joint judgment, the false alarm and missed alarm rates are effectively reduced, and reliable perception is achieved under all-weather conditions; the dual-layer intelligent discrimination is pushed down to the device end, supporting local autonomous decision-making and low-latency response, without relying on the cloud, and can still provide stable alarms in scenarios with poor network conditions; intelligent supplementary lighting is achieved through graded illumination matching, and supplementary lighting output is precisely controlled as needed, taking into account both imaging quality and low power consumption requirements; at the same time, an extreme duty mode is set to ensure the normal survival of the device under low power conditions and retain external response capabilities, effectively extending the battery life cycle and comprehensively improving the practicality and reliability of unattended outdoor scenarios. Attached Figure Description

[0029] Figure 1 The flowchart shows the two-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering provided by the present invention.

[0030] Figure 2 The diagram shows the system of a two-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering provided by the present invention. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0032] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0033] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0034] like Figure 1 As shown, this embodiment of the invention provides a dual-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering. In scenarios with no mains power and low power constraints, it is based on the low power consumption of AOV constant-viewing, achieves event triggering through multi-source heterogeneous sensing fusion, and then completes accurate identification through a dual-layer intelligent discrimination mechanism. Finally, it achieves controllable energy consumption throughout the entire lifecycle using a graded power consumption strategy. The method includes the following steps:

[0035] Step S1: Construct a perception layer consisting of an image sensor, millimeter-wave radar, an all-in-one environmental sensor, and an intelligent supplementary lighting unit. Configure the image sensor to operate in AOV low frame rate and low resolution normal mode, while simultaneously enabling the millimeter-wave radar and environmental sensor to operate in an intermittent sampling low-power mode.

[0036] Step S2: Real-time acquisition of environmental image data, radar detection data and environmental parameter data through the perception layer; joint detection based on AOV vision algorithm and radar detection results; identification of significant pixel changes or effective moving targets that meet preset conditions; and generation of event trigger signals.

[0037] Step S3: Respond to the event trigger signal, wake up the main control SoC in sleep mode and switch the image sensor to high resolution high definition capture mode, read the ambient illuminance data and dynamically control the start and stop of the intelligent supplement unit and pulse output according to the graded strategy;

[0038] Step S4: The main control SoC calls the NPU to execute a two-layer intelligent discrimination process. The first layer is a coarse screening discrimination based on the AOV low-power fast filtering algorithm, and the second layer is a high-definition image accurate recognition discrimination based on a lightweight AI model.

[0039] Step S5: Determine the event type based on the dual-layer intelligent discrimination result, generate structured alarm data and upload it through the main communication module. After the event is processed, the control system returns to the AOV ultra-low power normal mode.

[0040] Step S6: Monitor the battery level of the energy layer in real time. When the battery level is lower than the preset threshold, enter the extreme guard mode, shut down the internal trigger source and retain only the external trigger response capability.

[0041] In this embodiment, in step S1, a perception layer is constructed consisting of a global shutter CMOS image sensor, a millimeter-wave radar module, an all-in-one environmental sensor array, and a 940nm VCSEL intelligent fill light unit. The image sensor is configured to enter the AOV ultra-low power constant viewing mode, with a frame rate of 1fps±0.2fps, a resolution of VGA / QCIF level, an operating voltage of 1.2V–1.5V, and a standby power consumption of no more than 10mW.

[0042] Simultaneously, a millimeter-wave radar is configured with a scanning period of 200ms–500ms and a transmission power of -20dBm to -10dBm in a low-power continuous scanning state; an all-in-one environmental sensor is configured with a sampling period of 5s–30s and a single sampling duration of no more than 100ms in an intermittent sampling state, so that the overall perception layer can maintain basic perception capabilities with the lowest energy consumption.

[0043] In step S2, during the AOV normal operation phase, low-resolution image frames, millimeter-wave radar point cloud data, temperature and humidity, rainfall, illuminance, gas concentration and other multi-dimensional environmental parameters are continuously collected.

[0044] Based on the AOV pixel change threshold algorithm, anomaly detection is performed using three conditions: an inter-frame difference threshold of 8–16 gray levels, a regional change percentage threshold of 5%–15%, and a continuous frame change duration threshold of 3 frames or more. At the same time, it is combined with a three-dimensional joint judgment logic based on millimeter-wave radar distance of 5–150m, speed of 0.1–30m / s, and echo intensity ≥-60dB. Only when image anomaly and radar target simultaneously meet the preset conditions is it judged as a valid movement event and an event trigger signal is generated, thereby achieving a highly reliable event triggering mechanism with low false alarms.

[0045] In step S3, the event trigger signal triggers the MCU power management unit to output a wake-up command, switching the hibernation-state main control SoC to the working state.

[0046] The system synchronously controls the image sensor to switch from AOV low-resolution mode to full-resolution high-definition capture mode. It reads real-time illumination data and performs multi-level illumination threshold determination, classifying it into four levels: high illumination (>3000 Lux), medium illumination (10–3000 Lux), low illumination (0.1–10 Lux), and extremely low illumination (<0.1 Lux). Based on different illumination levels, it dynamically matches various illumination strategies, including no supplementary lighting, instantaneous high-brightness supplementary lighting (lasting 10–50ms, peak power 1–3W), pulsed intermittent supplementary lighting (frequency 10–50Hz, duty cycle 5%–20%), and adaptive dimming supplementary lighting, ensuring precise matching of supplementary lighting output with scene illumination and target distance.

[0047] In step S4, after the main control SoC is woken up, it drives the built-in NPU to load the YOLOv5s or YOLOv8-tiny lightweight AI model. The model has ≤5M parameters, ≤1TOPS computing power requirement, and ≤50ms inference time; and executes a two-layer intelligent discrimination process.

[0048] The first layer is an AOV low-computing-power fast filtering algorithm, which uses a dual mechanism of grayscale mean difference and edge feature coarse screening to filter out invalid anomalies such as sudden changes in light, rain and snow interference, and vegetation shaking. The filtering accuracy is ≥90% and the single frame time is ≤5ms.

[0049] The second layer is a high-definition image precision recognition algorithm that performs feature extraction and target classification on captured images. It can identify five types of targets: people, vehicles, falling rocks, foreign objects on the track, and animals, with an accuracy rate of ≥95%. It supports parallel reasoning for multiple targets simultaneously.

[0050] In step S5, after the dual-layer discrimination results are verified by confidence and timing consistency, the true event type is determined. Structured alarm data containing timestamps, location information, target attributes, and high-definition captured images is generated and encrypted and uploaded to the remote monitoring center via an SFP optical module. After event processing is completed, high-power modules are shut down step-by-step, restoring the system to the AOV ultra-low power normal operation mode and maintaining low-power cyclic operation.

[0051] In step S6, the SOC value of the lithium battery in the energy layer is monitored in real time. When the SOC value is lower than 15%±2%, the extreme value guard mode is entered.

[0052] In extreme duty mode, all internal triggering sources such as AOV sensing, radar scanning, and active detection are turned off, and the system operating current is ≤10mA and power consumption is ≤50mW. Only the communication link monitoring and external trigger response capabilities are retained. It can respond to the instructions of the monitoring center, the linkage signals of adjacent devices, and the preset emergency trigger signals. After being triggered, it is temporarily woken up to complete the capture and reporting. After processing, it immediately falls back to extreme duty mode to ensure the system survival and critical response capabilities under low power conditions.

[0053] refer to Figure 2 As shown, this embodiment of the invention provides a dual-layer intelligent discrimination system based on ultra-low power AOV mode and event triggering. It is an integrated low-power intelligent monitoring device with a four-layer architecture. The system consists of a perception layer, a processing layer, a communication layer, and an energy layer. Data interaction and control linkage between each layer are achieved through standard interfaces. The specific structure and functions are as follows:

[0054] The sensing layer consists of a global shutter CMOS image, a 77GHz or 24GHz FMCW millimeter-wave radar, an all-in-one environmental sensor array, and a 940nm red-exposure-free VCSEL intelligent fill light unit.

[0055] Specifically, the global shutter CMOS image sensor supports dual-mode switching between AOV low-power mode (resolution ≤640×480, operating current ≤8mA) and high-definition capture mode (resolution ≥1920×1080, operating current ≤120mA), and connects to the main control SoC via MIPI-CSI or Ethernet port.

[0056] The millimeter-wave radar supports three-dimensional detection at distances of 5–150m and angles of ±60°, with a sleep power consumption of ≤1mW and an operating power consumption of ≤80mW.

[0057] The all-in-one environmental sensor integrates temperature, humidity, rainfall, illuminance, and gas concentration detection functions; the intelligent supplementary lighting unit supports three supplementary lighting modes: instantaneous high brightness, pulse intermittent, and adaptive dimming, with precise output control by the MCU.

[0058] The processing layer consists of the RV1126B main control SoC with integrated 3TOPS level NPU, the MCU power management unit with integrated MPPT and multiple DC-DC converters, LPDDR4 memory, and eMMC storage unit.

[0059] Specifically, the MCU power management unit integrates system-level power management, low battery monitoring, extreme value control, and multi-level power supply control modules, which can realize module sleep wake-up, power on / off, dynamic voltage regulation, SOC monitoring, and graded power supply output.

[0060] The main control SoC integrates a dual-layer intelligent discrimination function, including an AOV fast filtering submodule, an NPU accurate identification submodule, a confidence verification submodule, and a timing consistency verification submodule, to complete event filtering, target identification, and result verification.

[0061] Memory is used for temporary data caching, and storage units are used to store model, configuration, and historical data.

[0062] The communication layer consists of an SFP optical module, a gigabit Ethernet port, a 4G / 5G wireless module, and an RS485 spare interface.

[0063] Specifically, the SFP optical module serves as the primary high-speed encrypted transmission channel, responsible for uploading alarm data, images, and device status; Gigabit Ethernet, wireless modules, and RS485 serve as backup communication links, adapting to network anomalies or remote communication interruptions to ensure data transmission redundancy.

[0064] The energy layer consists of solar panels, lithium iron phosphate battery packs, and MPPT charging management controllers.

[0065] Specifically, solar panels collect solar energy, which is then used by the MPPT controller to achieve maximum power point tracking and charge the lithium batteries. The lithium iron phosphate battery pack serves as an energy storage unit, providing tiered power supply for the entire system, including low-power supply under normal conditions and ultra-low-power supply under extreme conditions, ensuring long-term stable operation of the system in environments without mains power.

[0066] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A two-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering, characterized in that, Includes the following steps: Step S1: Construct a perception layer consisting of an image sensor, millimeter-wave radar, an all-in-one environmental sensor, and an intelligent supplementary lighting unit. Configure the image sensor to operate in AOV low frame rate and low resolution normal mode, while simultaneously enabling the millimeter-wave radar and environmental sensor to operate in an intermittent sampling low-power mode. Step S2: Real-time acquisition of environmental image data, radar detection data and environmental parameter data through the perception layer; joint detection based on AOV vision algorithm and radar detection results; identification of significant pixel changes or effective moving targets that meet preset conditions; and generation of event trigger signals. Step S3: Respond to the event trigger signal, wake up the main control SoC in sleep mode and switch the image sensor to high resolution high definition capture mode, read the ambient illuminance data and dynamically control the start and stop of the intelligent supplement unit and pulse output according to the graded strategy; Step S4: The main control SoC calls the NPU to execute a two-layer intelligent discrimination process. The first layer is a coarse screening discrimination based on the AOV low-power fast filtering algorithm, and the second layer is a high-definition image accurate recognition discrimination based on a lightweight AI model. Step S5: Determine the event type based on the dual-layer intelligent discrimination result, generate structured alarm data and upload it through the main communication module. After the event is processed, the control system returns to the AOV ultra-low power normal mode. Step S6: Monitor the battery level of the energy layer in real time. When the battery level is lower than the preset threshold, enter the extreme guard mode, shut down the internal trigger source and retain only the external trigger response capability.

2. The dual-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering as described in claim 1, characterized in that, In step S1, the AOV ultra-low power constant-view mode is set with a frame rate of 1fps±0.2fps, a resolution of VGA / QCIF, an image sensor operating voltage of 1.2V–1.5V, and standby power consumption of less than 10mW; the millimeter-wave radar scanning period is set to 200ms–500ms, and the transmit power is reduced to -20dBm to -10dBm; the multi-functional environmental sensor sampling period is 5s–30s, and the single sampling duration does not exceed 100ms.

3. The dual-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering as described in claim 1, characterized in that, In step S2, based on the AOV pixel change threshold algorithm, there are three judgments: inter-frame difference threshold, regional change proportion threshold, and continuous frame change duration threshold. The inter-frame difference threshold is 8-16 gray levels, the regional change proportion threshold is 5%-15%, and the continuous frame change duration threshold is 3 frames or more. Combined with the radar joint judgment of valid moving events logic, the three conditions of distance 5-150m, speed 0.1-30m / s, and echo intensity ≥-60dB are met.

4. The dual-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering as described in claim 1, characterized in that, In step S3, the multi-level illumination data threshold is determined and divided into four levels: high illumination (>3000Lux), medium illumination (10–3000Lux), low illumination (0.1–10Lux), and extremely low illumination (<0.1Lux); the duration of instantaneous high-brightness supplementary lighting is 10–50ms, the peak power is 1–3W; and the frequency of pulsed intermittent supplementary lighting is 10–50Hz, with a duty cycle of 5%–20%; the adaptive dimming supplementary lighting dynamically adjusts the laser power according to the target distance, and sets the power to be higher the farther the target is.

5. The dual-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering as described in claim 1, characterized in that, In step S4, the lightweight AI model adopts a customized YOLOv5s / YOLOv8-tiny model with ≤5M parameters, ≤1TOPS computing power requirement, and ≤50ms inference time. The first-layer fast filtering algorithm uses a dual mechanism of gray-scale mean difference and edge feature coarse screening, with a filtering accuracy of ≥90% and a single-frame time of ≤5ms. The second-layer accurate recognition supports five types of targets: personnel, vehicles, falling rocks, track debris, and animals, and supports simultaneous detection of multiple targets with an accuracy of ≥95%.

6. The dual-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering as described in claim 1, characterized in that, In step S6, the low battery preset threshold is set to 15% ± 2%; In the extreme monitoring mode, the system operating current is ≤10mA and the power consumption is ≤50mW; it only responds to three types of external triggers: monitoring center instructions, adjacent device linkage signals, and preset emergency trigger signals. After being triggered, it can be temporarily woken up to complete the capture and reporting, and immediately revert to the extreme monitoring mode.

7. A two-layer intelligent discrimination system based on ultra-low power AOV mode and event triggering, characterized in that, The system can execute the dual-layer intelligent discrimination method based on ultra-low power AOV mode and event triggering as described in any one of claims 1 to 6, wherein the system includes a sensing layer, a processing layer, a communication layer and an energy layer; The sensing layer includes a global shutter CMOS image sensor, a millimeter-wave radar module, an all-in-one environmental sensor, and a 940nm VCSEL intelligent fill light unit. The image sensor is configured with AOV low frame rate normal mode and high-definition capture trigger mode. The millimeter-wave radar and environmental sensor perform low-power intermittent sampling, and the intelligent fill light unit outputs through PWM control. The processing layer includes a main control SoC with integrated NPU, an MCU power management unit, and a memory and storage unit. The MCU power management unit is responsible for system timing control and power consumption scheduling. The main control SoC responds to the trigger signal to wake up and execute dual-layer intelligent discrimination. The NPU runs a lightweight AI model to achieve accurate target recognition. The communication layer includes an SFP optical module, a gigabit Ethernet interface, and a wireless backup module. The SFP optical module is the primary high-speed data transmission channel, and the backup module is adapted for network anomaly scenarios. The energy layer includes solar panels, lithium iron phosphate battery packs, and an MPPT controller. The MPPT controller manages charging and power supply and provides tiered power supply support for the system.

8. The dual-layer intelligent discrimination system based on ultra-low power AOV mode and event triggering as described in claim 7, characterized in that, The global shutter CMOS image sensor supports dynamic switching between AOV low-power mode and high-definition capture mode. In AOV mode, it outputs images with a resolution of 640×480 and below, and the operating current is ≤8mA; in high-definition mode, it outputs images with a resolution of 1920×1080 and above, and the operating current is ≤120mA. The sensor is connected to the main control SoC through MIPI-CSI / Ethernet dual interfaces, and supports frame synchronization triggering and mode switching hardware control. The millimeter-wave radar module adopts the FMCW frequency-modulated continuous wave system, supports three-dimensional detection of distance, speed, and angle, with a detection range of 5–150m, an angle coverage of ±60°, and a refresh rate of 10–20Hz. It is connected to the MCU power management unit via UART / SPI, supports parameter configuration, data transmission, and sleep / wake-up hardware control, with sleep power consumption ≤1mW and operating power consumption ≤80mW.

9. The dual-layer intelligent discrimination system based on ultra-low power AOV mode and event triggering as described in claim 8, characterized in that, The dual-layer intelligent discrimination is integrated into the NPU inside the main control SoC, including an AOV fast filtering submodule, an NPU accurate recognition submodule, a confidence verification submodule, and a timing consistency verification submodule. The AOV fast filtering submodule includes solidified inter-frame difference, edge detection, and region proportion triple filtering logic. The NPU accurate recognition submodule deploys a lightweight YOLO model and supports multi-objective parallel inference. The verification submodule performs confidence threshold filtering and continuous frame result comparison on the recognition results.

10. The dual-layer intelligent discrimination system based on ultra-low power AOV mode and event triggering as described in claim 9, characterized in that, The MCU power management unit integrates a system-level power management module, a low-battery monitoring module, an extreme value protection control module, and a multi-level power supply control module. The system-level power management module is responsible for module sleep / wake-up timing, power on / off control, and dynamic voltage regulation. The low-battery monitoring module collects lithium battery SOC, voltage, and current in real time with an accuracy of ≤1%. When triggered, the extreme value protection control module cuts off power to the image sensor, radar, NPU, and main communication module, retaining only the MCU and basic communication monitoring circuitry. The multi-level power supply control module supports multiple voltage outputs of 1.2V / 1.8V / 3.3V / 5V.

Images