Intelligent security intercom system based on PIR and AI behavior analysis

Abstract

This invention discloses an intelligent security intercom system based on PIR and AI behavior analysis, belonging to the field of intelligent security technology. The system includes: a video acquisition module for acquiring video data from the monitored area in real time; an AI behavior analysis module for performing three-level model inference, cross-camera trajectory tracking, and skeletal key point anomaly calculation on the video data, performing dynamic behavior recognition and generating analysis results; an intercom communication module for dynamically configuring communication strategies through a behavior risk assessment algorithm, establishing a multimodal response mechanism, and establishing a two-way video communication channel based on the multimodal response mechanism; and a system control module for coordinating the video data transmission link based on a four-level pipeline architecture, a dynamic resource scheduling mechanism, and an energy efficiency optimization strategy, and triggering intercom when preset risk behaviors are identified. This invention can achieve an optimal balance between security protection and energy efficiency.

Description

Technical Field

[0001] The present invention relates to the field of intelligent security technology, and in particular to an intelligent security intercom system based on PIR and AI behavior analysis. Background Art

[0002] Traditional security systems are automated protection systems that monitor target areas in real time through sensor networks and trigger alarms when anomalies are detected. Their core structure consists of three parts: (1) a data acquisition layer, which collects data through sensors such as cameras, PIR sensors, and radars; (2) a data processing layer, which performs signal conversion and feature extraction; and (3) a response execution layer, which responds to the analysis results using alarms or intercom systems. Their basic operating principle is a linear "perception-analysis-response" process, where traditional PIR systems rely on changes in infrared thermal radiation to detect biological movement, while pure visual AI systems use computer vision algorithms to analyze video content.

[0003] Traditional PIR systems, which only detect the movement of heat sources, cannot distinguish between human behavior and intention (such as maintenance workers and thieves) and are easily affected by ambient temperature. Purely visual AI systems, while capable of recognizing complex behaviors, require continuous deep learning model execution, resulting in high energy consumption and a sharp performance degradation in low-light scenarios. Both systems share common pain points: the limitations of a single sensing modality, a mismatch between resource allocation and scenario requirements, and inefficient multi-source data collaboration.

[0004] Traditional hierarchical analysis uses fixed cascade models, such as detection-classification-tracking. This leads to a rigid computational pipeline: simple scenarios are over-computed (e.g., performing full-model analysis on stationary objects), while complex events may be missed due to insufficient resources. This architecture lacks dynamic adaptability and struggles to balance real-time performance with accuracy. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an intelligent security intercom system based on PIR and AI behavior analysis, which can achieve an optimal balance between security protection and energy efficiency.

[0006] On the one hand, an embodiment of the present invention provides an intelligent security intercom system based on PIR and AI behavior analysis, including: a video acquisition module for acquiring video data of a monitored area in real time; the video acquisition module includes a low-power standby unit for activating a high-definition video acquisition function through an external trigger signal; the external trigger signal comes from a PIR sensor, which generates a wake-up command after detecting a human body heat source and transmits it to the video acquisition module; an AI behavior analysis module is used to perform three-level model inference, cross-camera trajectory tracking, and skeletal key point abnormality calculation on video data, perform dynamic behavior recognition, and generate analysis results; the three-level model includes a YOLO-Nano target detection model, a ResNet18 behavior classification model, and a 3D-CNN spatiotemporal modeling model; the AI ​​behavior analysis module includes a dual-mode verification mechanism unit for cross-validating the PIR signal and the AI ​​preliminary analysis result, and skipping the ResNet18 behavior classification model when the two are consistent. and the in-depth analysis steps of the 3D-CNN spatiotemporal modeling model to directly trigger a response; an intercom communication module is used to dynamically configure communication strategies through a behavioral risk assessment algorithm, establish a multimodal response mechanism, and establish a video bidirectional communication channel based on the multimodal response mechanism; a system control module is used to coordinate the transmission link of video data based on a four-stage pipeline architecture, a dynamic resource scheduling mechanism, and an energy efficiency optimization strategy, and trigger intercom when a preset risk behavior is identified; the four-stage pipeline architecture includes: an acquisition layer, which is used to write data directly into the FPGA preprocessing unit through the DMA channel when the video acquisition module outputs RAW data; a preprocessing layer, which is used to integrate an H.265 hardware encoder to implement dynamic bit rate control on the video stream; a transport layer, which is used to directly map the video stream to the network protocol stack memory space after encoding by deploying zero-copy transmission technology; an analysis layer, which is used to establish a priority queue, and the AI ​​behavior analysis module extracts data from the shared memory pool through RDMA.

[0007] According to some embodiments of the present invention, the AI ​​behavior analysis module includes: a dynamic sampling and control unit for obtaining target movement speed parameters in real time through a PIR sensor and establishing a frame rate-speed mapping model; the frame rate-speed mapping model includes: when the target movement speed is lower than a first threshold, using a first frame rate interval sampling, and cooperating with a motion compensation algorithm to reconstruct the picture; when the target movement speed is in a range greater than the first threshold and less than the second threshold, enabling a second frame rate benchmark sampling, and cooperating with a motion compensation algorithm to reconstruct the picture; when the target movement speed exceeds the second threshold, switching to a third frame rate full frame rate mode.

[0008] According to some embodiments of the present invention, the AI ​​behavior analysis module includes: a regional interest focusing unit, which is used to establish a dynamic area of ​​interest based on the heat source coordinates provided by the PIR, and generate three concentric analysis areas with the heat source as the center: a core area, a buffer area and a peripheral area; perform pixel-level feature analysis on the core area, downsample the buffer area, and only perform motion detection on the peripheral area; use an adaptive ROI segmentation algorithm to automatically merge the analysis areas when the distance between multiple heat sources is less than the first distance.

[0009] According to some embodiments of the present invention, the AI ​​behavior analysis module includes: a multi-level analysis pipeline unit, which is used to construct a three-level cascade analysis architecture to realize three-level model reasoning: primary pipeline, intermediate pipeline and advanced pipeline; the primary pipeline runs a lightweight YOLO-Nano model to complete basic target detection; the intermediate pipeline activates ResNet18 for behavior classification when an anomaly is detected; the advanced pipeline only loads the 3D CNN spatiotemporal model for high-risk events; and confidence valves are set between each level of pipeline.

[0010] According to some embodiments of the present invention, the AI ​​behavior analysis module includes: a timing association optimization unit, which is used to align PIR pulses and video frame sequences through a hardware timestamp synchronizer to establish an event timeline database; an improved LSTM network is used to analyze continuous events, and when a wandering-climbing-intrusion pattern sequence is detected, the event danger level is automatically increased; the startup time of heat sources that have been stationary for more than a first time interval is compressed and stored, and only the first and last key frames and the middle differential frames are retained.

[0011] According to some embodiments of the present invention, the AI ​​behavior analysis module includes: an energy efficiency perception model unit, a built-in environmental status evaluator, which is used to dynamically adjust the model complexity based on the comprehensive light intensity and PIR activation frequency; when the PIR does not detect activity for a second consecutive time interval, the behavior recognition model is automatically uninstalled and only the motion detection function is retained; the chip operating condition is monitored through a temperature sensor, and a computing resource reallocation strategy is triggered when overheating occurs.

[0012] According to some embodiments of the present invention, the dual-mode verification mechanism unit is used to design a PIR-AI confidence fusion algorithm. When the PIR is continuously triggered and the AI ​​detects a valid target within a third time interval, a pre-stored response strategy is called to skip the ResNet18 behavior classification model and the 3D-CNN spatiotemporal modeling model. For verification conflict events, a delayed review mechanism is initiated: a secondary judgment is performed after continuously collecting data for a fourth time interval. If a conflict still occurs, it is marked as a device failure. The confidence of the PIR-AI confidence fusion algorithm is calculated as follows: PIR confidence = 1-0.5^(number of consecutive triggers), AI confidence = detection box IOU × classification probability, PIR-AI fusion confidence = A*PIR confidence + B*AI confidence, where A and B are coefficients. The verification conflict events include: the deviation between the PIR trigger time and the AI ​​detection time window exceeds a set threshold, the PIR is continuously triggered but the AI ​​does not detect, the AI ​​recognizes an anomaly but the PIR does not provide feedback, and the difference in confidence between the two parties is greater than a set threshold.

[0013] According to some embodiments of the present invention, the intercom communication module includes: a behavior risk assessment unit, which is used to construct a spatiotemporal feature vector based on the behavior type, duration and trajectory coordinates, and output the risk level using an XGBoost classifier.

[0014] According to some embodiments of the present invention, the intercom communication module includes: a multimodal response mechanism unit, which is used to determine a combined response mode based on the risk level, and the combined response mode includes: a voice inquiry and log recording combination, a video intercom and AI bypass monitoring combination, and an alarm linkage and drone tracking combination.

[0015] According to some embodiments of the present invention, the system control module includes: a dynamic resource scheduling unit, configured to assign a dynamic weight to each video channel based on an improved weighted fair queueing algorithm, and perform computing resource scheduling based on the dynamic weight; the dynamic weight allocation algorithm is as follows: ;in, is the dynamic weight of the i-th video channel, PIR trigger strength = PIR signal amplitude × duration. The risk level output by AI behavior analysis, Score the computational complexity of the current analysis model, is the coefficient.

[0016] An intelligent security intercom system based on PIR and AI behavioral analysis according to an embodiment of the present invention includes at least the following beneficial effects: the embodiment of the present invention realizes complementary advantages of sensors through PIR-AI dual-mode verification, utilizes dynamic sampling control and three-level concentric analysis areas to achieve precise allocation of computing resources, and combines time series correlation optimization to solve the problem of multi-source data synchronization. The analysis depth is dynamically adjusted according to the environmental status through the energy efficiency perception model, and the multi-level analysis pipeline realizes intelligent reasoning degradation through the confidence valve. Through the synergistic effect of the above technical features, it is possible to break through the performance bottlenecks of traditional systems in terms of false alarm rate, energy consumption, response speed, etc., and achieve the optimal balance between security protection and energy efficiency.

[0017] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0019] Figure 1 This is a schematic block diagram of modules of a system according to an embodiment of the present invention.

[0020] Reference numerals:

[0021] Video acquisition module 100, AI behavior analysis module 200, intercom communication module 300, and system control module 400. DETAILED DESCRIPTION

[0022] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0023] In the description of the present invention, "several" means one or more, "more" means two or more, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The use of "first" and "second" in the description is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0024] Reference Figure 1 The embodiment of the present invention proposes an intelligent security intercom system based on PIR and AI behavior analysis, including:

[0025] The video acquisition module 100 is used to acquire video data of the monitored area in real time. The video acquisition module includes a low-power standby unit for activating the high-definition video acquisition function through an external trigger signal. The external trigger signal comes from the PIR sensor, which generates a wake-up command after detecting the human body heat source and transmits it to the video acquisition module.

[0026] The AI ​​behavior analysis module 200 is used to perform three-level model inference, cross-camera trajectory tracking, and skeletal key point anomaly calculation on video data, perform dynamic behavior recognition and generate analysis results; the three-level model includes a YOLO-Nano target detection model, a ResNet18 behavior classification model, and a 3D-CNN spatiotemporal modeling model; the AI ​​behavior analysis module includes a dual-mode verification mechanism unit for cross-validating the PIR signal and the AI ​​preliminary analysis results. When the two are consistent, the deep analysis steps of the ResNet18 behavior classification model and the 3D-CNN spatiotemporal modeling model are skipped, and a response is directly triggered.

[0027] The intercom communication module 300 is used to dynamically configure communication strategies through a behavioral risk assessment algorithm, establish a multimodal response mechanism, and establish a two-way video communication channel based on the multimodal response mechanism.

[0028] The system control module 400 is used to coordinate the transmission link of video data based on a four-stage pipeline architecture, a dynamic resource scheduling mechanism and an energy efficiency optimization strategy, and trigger intercom when a preset risk behavior is identified.

[0029] The four-stage pipeline architecture of the embodiment of the present invention includes:

[0030] The acquisition layer is used to write the data directly into the FPGA pre-processing unit through the DMA channel when the video acquisition module outputs RAW data.

[0031] The pre-processing layer is used to integrate the H.265 hardware encoder and implement dynamic bit rate control on the video stream.

[0032] The transport layer is used to deploy zero-copy transmission technology, and the video stream is directly mapped to the network protocol stack memory space after encoding.

[0033] The analysis layer is used to establish a priority queue. The AI ​​behavior analysis module extracts data from the shared memory pool through RDMA.

[0034] In some embodiments, the AI ​​behavior analysis module 200 includes: a dynamic sampling control unit, which is used to obtain the target movement speed parameters in real time through a PIR sensor and establish a frame rate-speed mapping model; the frame rate-speed mapping model includes: when the target movement speed is lower than a first threshold, using a first frame rate interval sampling, and cooperating with a motion compensation algorithm to reconstruct the picture; when the target movement speed is in a range greater than the first threshold and less than the second threshold, enabling a second frame rate baseline sampling, and cooperating with a motion compensation algorithm to reconstruct the picture; when the target movement speed exceeds the second threshold, switching to a third frame rate full frame rate mode.

[0035] In some embodiments, the AI ​​behavior analysis module 200 includes: a regional interest focusing unit, which is used to establish a dynamic focus area based on the heat source coordinates provided by the PIR, and generate three concentric analysis areas with the heat source as the center: a core area, a buffer area and a peripheral area; perform pixel-level feature analysis on the core area, downsample the buffer area, and only perform motion detection on the peripheral area; use an adaptive ROI segmentation algorithm to automatically merge analysis areas when the distance between multiple heat sources is less than the first distance.

[0036] In some embodiments, the AI ​​behavior analysis module 200 includes: a multi-level analysis pipeline unit, which is used to build a three-level cascade analysis architecture to implement three-level model reasoning: a primary pipeline, an intermediate pipeline, and a high-level pipeline; the primary pipeline runs a lightweight YOLO-Nano model to complete basic target detection; the intermediate pipeline activates ResNet18 for behavior classification when an anomaly is detected; the high-level pipeline only loads a 3D CNN spatiotemporal model for high-risk events; and confidence valves are set between each level of pipeline.

[0037] In some embodiments, the AI ​​behavior analysis module 200 includes: a timing correlation optimization unit for aligning PIR pulses and video frame sequences through a hardware timestamp synchronizer to establish an event timeline database; using an improved LSTM network to analyze continuous events, and automatically raising the event danger level when a wandering-climbing-intrusion pattern sequence is detected; compressing and storing the startup time of heat sources that have been stationary for more than a first time interval, retaining only the first and last key frames and the middle differential frames.

[0038] In some embodiments, the AI ​​behavior analysis module 200 includes: an energy efficiency perception model unit, a built-in environmental status evaluator, which is used to dynamically adjust the model complexity based on the comprehensive light intensity and PIR activation frequency; when the PIR does not detect activity for a second consecutive time interval, the behavior recognition model is automatically uninstalled and only the motion detection function is retained; the chip operating condition is monitored through a temperature sensor, and a computing resource reallocation strategy is triggered when overheating occurs.

[0039] In some embodiments, the dual-mode verification mechanism unit of the embodiments of the present invention is used to design a PIR-AI confidence fusion algorithm. When the PIR is continuously triggered and the AI ​​detects a valid target within a third time interval, a pre-stored response strategy is called to skip the ResNet18 behavior classification model and the 3D-CNN spatiotemporal modeling model. For verification conflict events, a delayed review mechanism is initiated: a secondary determination is performed after continuously collecting data for a fourth time interval. If a conflict still occurs, it is marked as a device failure. The confidence of the PIR-AI confidence fusion algorithm is calculated as follows: PIR confidence = 1-0.5^(number of consecutive triggers), AI confidence = detection box IOU × classification probability, PIR-AI fusion confidence = A*PIR confidence + B*AI confidence, where A and B are coefficients. The verification conflict events include: the deviation between the PIR trigger time and the AI ​​detection time window exceeds a set threshold, the PIR is continuously triggered but the AI ​​does not detect, the AI ​​recognizes an anomaly but the PIR does not provide feedback, and the difference in confidence between the two parties is greater than a set threshold.

[0040] In some embodiments, the intercom communication module 300 includes: a behavior risk assessment unit for constructing a spatiotemporal feature vector based on the behavior type, duration, and trajectory coordinates, and outputting a risk level using an XGBoost classifier.

[0041] In some embodiments, the intercom communication module 300 includes: a multimodal response mechanism unit, which is used to determine a combined response mode based on the risk level, and the combined response mode includes: a voice inquiry and log recording combination, a video intercom and AI bypass monitoring combination, and an alarm linkage and drone tracking combination.

[0042] In some embodiments, the system control module 400 includes a dynamic resource scheduling unit configured to assign a dynamic weight to each video channel based on an improved weighted fair queueing algorithm and perform computing resource scheduling based on the dynamic weight. The dynamic weight allocation algorithm is as follows:

[0043] ;

[0044] in, is the dynamic weight of the i-th video channel, PIR trigger strength = PIR signal amplitude × duration. The risk level output by AI behavior analysis, Score the computational complexity of the current analysis model, is the coefficient.

[0045] Although specific embodiments are described herein, those skilled in the art will recognize that many other modifications or alternative embodiments are also within the scope of the present disclosure. For example, any of the functions and / or processing capabilities described in conjunction with a particular device or component may be performed by any other device or component. In addition, although various exemplary implementations and architectures have been described in accordance with the embodiments of the present disclosure, those skilled in the art will recognize that many other modifications to the exemplary implementations and architectures described herein are also within the scope of the present disclosure.

[0046] Those skilled in the art will appreciate that all or some of the steps and systems disclosed above can be implemented as software, firmware, hardware, or any suitable combination thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is well known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVDs) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0047] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the above implementation mode. Technical personnel familiar with the field can also make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. An intelligent security intercom system based on PIR and AI behavior analysis, characterized by: include: Video acquisition module, used to obtain video data of the monitored area in real time; The video acquisition module includes a low-power standby unit for activating the high-definition video acquisition function through an external trigger signal; the external trigger signal comes from the PIR sensor, which generates a wake-up command after detecting the human body heat source and transmits it to the video acquisition module; The AI ​​behavior analysis module is used to perform three-level model inference, cross-camera trajectory tracking, and skeletal key point anomaly calculation on video data to perform dynamic behavior recognition and generate analysis results. The three-level model includes a YOLO-Nano target detection model, a ResNet18 behavior classification model, and a 3D-CNN spatiotemporal modeling model. The AI ​​behavior analysis module includes a dual-mode verification mechanism unit for cross-validating the PIR signal and the AI ​​preliminary analysis results. If the two are consistent, the in-depth analysis steps of the ResNet18 behavior classification model and the 3D-CNN spatiotemporal modeling model are skipped and a response is directly triggered. Intercom communication module, which is used to dynamically configure communication strategies through behavioral risk assessment algorithms, establish a multimodal response mechanism, and establish a two-way video communication channel based on the multimodal response mechanism; The system control module coordinates the video data transmission link based on a four-stage pipeline architecture, dynamic resource scheduling mechanism, and energy efficiency optimization strategy, and triggers intercom when pre-set risk behaviors are identified; The four-stage pipeline architecture includes: The acquisition layer is used to write the data directly into the FPGA pre-processing unit through the DMA channel when the video acquisition module outputs RAW data; The pre-processing layer is used to integrate the H.265 hardware encoder and implement dynamic bit rate control on the video stream; The transport layer is used to deploy zero-copy transmission technology to directly map the encoded video stream into the network protocol stack memory space; The analysis layer is used to establish a priority queue. The AI ​​behavior analysis module extracts data from the shared memory pool through RDMA.

2. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1 is characterized in that: The AI ​​behavior analysis module includes: The dynamic sampling control unit is used to obtain the target movement speed parameters in real time through the PIR sensor and establish a frame rate-speed mapping model; the frame rate-speed mapping model includes: when the target movement speed is lower than a first threshold, adopting a first frame rate interval sampling, and reconstructing the picture in conjunction with a motion compensation algorithm; when the target movement speed is between greater than the first threshold and less than a second threshold, enabling a second frame rate baseline sampling, and reconstructing the picture in conjunction with a motion compensation algorithm; when the target movement speed exceeds the second threshold, switching to a third frame rate full frame rate mode.

3. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1 is characterized in that: The AI ​​behavior analysis module includes: The regional interest focusing unit is used to establish a dynamic region of interest based on the heat source coordinates provided by the PIR, and generate three concentric analysis areas with the heat source as the center: the core area, the buffer area, and the peripheral area; pixel-level feature analysis is performed on the core area, downsampling is performed on the buffer area, and only motion detection is performed on the peripheral area; an adaptive ROI segmentation algorithm is used to automatically merge analysis areas when the distance between multiple heat sources is less than the first distance.

4. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1 is characterized in that: The AI ​​behavior analysis module includes: The multi-level analysis pipeline unit is used to build a three-level cascade analysis architecture to implement three-level model inference: primary pipeline, intermediate pipeline, and advanced pipeline. The primary pipeline runs a lightweight YOLO-Nano model to complete basic target detection. The intermediate pipeline activates ResNet18 for behavior classification when an anomaly is detected. The advanced pipeline only loads the 3D CNN spatiotemporal model for high-risk events. Confidence valves are set between each level of the pipeline.

5. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1 is characterized in that: The AI ​​behavior analysis module includes: The timing association optimization unit is used to align PIR pulses and video frame sequences through a hardware timestamp synchronizer to establish an event timeline database. An improved LSTM network is used to analyze continuous events. When a wandering-over-intrusion pattern sequence is detected, the event danger level is automatically increased. Heat sources that have been stationary for more than a first time interval are compressed and stored, retaining only the first and last key frames and the intermediate differential frames.

6. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1 is characterized in that: The AI ​​behavior analysis module includes: The energy-efficiency perception model unit has a built-in environmental state evaluator that dynamically adjusts model complexity based on light intensity and PIR activation frequency. When the PIR detects no activity for the second consecutive time interval, the behavior recognition model is automatically uninstalled, retaining only the motion detection function. The chip's operating condition is monitored through a temperature sensor, triggering a computing resource reallocation strategy in the event of overheating.

7. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1 is characterized in that: The dual-mode verification mechanism unit is used to design a PIR-AI confidence fusion algorithm. When the PIR is continuously triggered and the AI ​​detects a valid target within a third time interval, a pre-stored response strategy is invoked to skip the ResNet18 behavior classification model and the 3D-CNN spatiotemporal modeling model. For verification conflict events, a delayed review mechanism is initiated: a secondary determination is performed after continuously collecting data for the fourth time interval. If a conflict still occurs, the device is marked as a fault. The confidence of the PIR-AI confidence fusion algorithm is calculated as follows: PIR confidence = 1-0.5^(number of consecutive triggers), AI confidence = detection box IOU × classification probability, and PIR-AI fusion confidence = A*PIR confidence + B*AI confidence, where A and B are coefficients. Verification conflict events include: the deviation between the PIR trigger time and the AI ​​detection time window exceeds a set threshold, the PIR is continuously triggered but the AI ​​does not detect, the AI ​​recognizes an anomaly but the PIR does not provide feedback, and the difference in confidence between the two parties is greater than a set threshold.

8. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1 is characterized in that: The intercom communication module includes: The behavioral risk assessment unit is used to construct a spatiotemporal feature vector based on the behavior type, duration, and trajectory coordinates, and uses the XGBoost classifier to output the risk level.

9. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1 is characterized in that: The intercom communication module includes: The multimodal response mechanism unit is used to determine a combination response mode based on the risk level. The combination response mode includes: a combination of voice inquiry and log recording, a combination of video intercom and AI bypass monitoring, and a combination of alarm linkage and drone tracking.

10. The intelligent security intercom system based on PIR and AI behavior analysis according to claim 1, characterized in that: The system control module includes: The dynamic resource scheduling unit is used to assign a dynamic weight to each video channel based on an improved weighted fair queueing algorithm and to schedule computing resources based on the dynamic weight. The dynamic weight allocation algorithm is as follows: ； in, is the dynamic weight of the i-th video channel, PIR trigger strength = PIR signal amplitude × duration. The risk level output by AI behavior analysis, Score the computational complexity of the current analysis model, is the coefficient.

Images