Anomaly behavior analysis method based on artificial intelligence
By using multimodal data collection and personalized behavioral baseline models for abnormal behavior analysis, the problem of false alarms and missed alarms in traditional methods has been solved, enabling accurate identification and early warning of individual behaviors, and improving the initiative and accuracy of health monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGSHI CLOUD TECHNOLOGY DEVELOPMENT (ANHUI) CO LTD
- Filing Date
- 2026-01-29
- Publication Date
- 2026-06-05
AI Technical Summary
Traditional abnormal behavior monitoring solutions rely on fixed rules and single sensors, which cannot adapt to individual differences, leading to false alarms and missed alarms. Furthermore, they lack the ability to analyze long-term behavioral patterns, making it impossible to achieve proactive early warning and early intervention.
By employing multimodal data acquisition, personalized behavioral baseline construction, real-time anomaly identification, and deep AI analysis, data is collected through the deployment of various heterogeneous sensors to establish a personalized behavioral baseline model. Statistical bias and AI models are used to identify abnormal behaviors and generate early warning information.
It improves the targeting and accuracy of early warnings, enabling the early detection of subtle changes in behavioral patterns and shifting from post-event alarms to pre-event warnings, thus providing a window of opportunity for early intervention.
Smart Images
Figure CN122140231A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence and health IoT technology, specifically to an abnormal behavior analysis method based on artificial intelligence. Background Technology
[0002] Traditional abnormal behavior monitoring solutions often rely on fixed rules or single types of sensors. For example, in home care scenarios, infrared motion detectors, emergency buttons, or camera-based fall detection algorithms are commonly used. These solutions typically pre-set uniform, non-personalized thresholds (such as triggering an alarm if a user stays in the bathroom for more than 30 minutes), and their core flaw lies in failing to consider the differences in individual daily behavioral patterns. A healthy user with specific lifestyle habits may frequently trigger false alarms, while a user whose behavioral patterns are progressively abnormal may be missed because the changes do not reach the fixed threshold. Furthermore, relying on a single data source (such as using only cameras) makes it difficult to reliably interpret complex behaviors and poses a significant risk of privacy breaches.
[0003] Furthermore, existing technologies have limited intelligence, mostly limited to providing "post-event" alerts for already occurred, severe, discrete events (such as falls or prolonged periods of inactivity), which is a passive response mode. They lack the ability to model and analyze users' long-term, multi-dimensional historical behavioral data, and therefore cannot identify subtle behavioral deviations that indicate early deterioration of health (such as a gradual decrease in activity levels or an increase in nighttime urination). This limitation of "seeing only the event, not the behavior" makes it difficult for existing systems to achieve truly proactive early warning and intervention.
[0004] To address the aforementioned issues, we propose an improvement by developing an artificial intelligence-based method for analyzing abnormal behavior. Summary of the Invention
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an artificial intelligence-based method for analyzing abnormal behavior, comprising the following steps: S101. Multimodal data acquisition steps: Through multiple heterogeneous sensors deployed in the monitoring space, real-time dynamic data of the target user is continuously collected, and historical behavioral data of the user is obtained. The sensors include at least one or more combinations of millimeter-wave radar, AI smart camera, AI smart mattress, and smart water and electricity meter. S102. Steps for constructing a personalized behavior baseline: Based on the historical behavior data, a personalized behavior baseline model covering multiple behavioral dimensions is established for the target user through time series analysis. The behavioral dimensions include at least activity patterns, sleep patterns, bathroom usage patterns, and water and electricity usage patterns. S103. Real-time anomaly identification step: The collected real-time dynamic data is compared with the personalized behavior baseline model in real time, and the statistical deviation analysis algorithm is used to identify abnormal data points in at least one behavioral dimension. S104. Artificial intelligence analysis and confirmation steps: Input the abnormal data points into a pre-trained abnormal behavior classification AI model for in-depth analysis and classification to confirm whether a predefined abnormal behavior event has occurred. The event includes at least abnormal toilet use, abnormal home activities, abnormal sleep, and abnormal water and electricity use. S105. Warning Generation and Output Steps: When an abnormal behavior event is confirmed, an early warning message containing the event type, severity level, and evidence data is automatically generated and sent to the designated recipient.
[0006] As a preferred technical solution of the present invention, the personalized behavior baseline model is a dynamic model, which sets a reference benchmark for each behavior dimension, including mean, standard deviation and normal fluctuation range, through statistical analysis.
[0007] As a preferred technical solution of the present invention, the confirmation of abnormal toilet use is based on the real-time monitoring of the entry frequency and the duration of a single stay, which is compared with the corresponding toilet use pattern in the baseline model. When the deviation exceeds a preset statistical threshold, the AI model determines it to be abnormal.
[0008] As a preferred technical solution of the present invention, the confirmation of abnormal home activities is achieved by integrating millimeter-wave radar and AI smart camera data, analyzing the user's activity level, activity range and static duration within a set time period, and comparing them with the individual's behavioral baseline. When the indicators deviate significantly, they are judged as abnormal.
[0009] As a preferred technical solution of the present invention, the confirmation of sleep abnormality is based on one or more of the following data collected by the AI smart mattress or vital sign monitoring radar: sleep duration, deep sleep ratio, turning frequency, heart rate and respiratory rate. When the data deviates continuously from the individual's sleep baseline, it is determined to be abnormal.
[0010] As a preferred technical solution of the present invention, the abnormal behavior classification AI model is a multi-classifier system, which includes one or more combinations of decision trees, support vector machines (SVM), and deep neural networks (DNN).
[0011] As a preferred embodiment of the present invention, the warning information further includes actionable suggestions generated by an AI model from a preset knowledge base based on the anomaly type and severity.
[0012] As a preferred embodiment of the present invention, the method is executed on an edge computing device to achieve local data processing and low-latency early warning.
[0013] As a preferred technical solution of the present invention, it also includes a baseline update step: periodically integrating verified normal behavior data into historical data, and iteratively optimizing the personalized behavior baseline model to adapt to changes in user behavior.
[0014] The beneficial effects of this invention are: An AI-based abnormal behavior analysis method completely breaks away from the one-size-fits-all fixed threshold mode by establishing a personalized behavioral baseline model for each user based on their own historical data. The system identifies significant deviations from an individual's normal state rather than absolute values exceeding the limit, thereby greatly improving the targeting and accuracy of early warnings and effectively reducing false alarms caused by individual habit differences.
[0015] An AI-based method for analyzing abnormal behavior continuously compares real-time data with personalized baselines and uses AI models to perform in-depth analysis of abnormal patterns. The system can capture subtle behavioral pattern changes that represent potential risks before serious accidents occur, thus achieving a leap from post-event alarms to pre-event warnings and creating a valuable time window for early intervention. Attached Figure Description
[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an abnormal behavior analysis method based on artificial intelligence according to the present invention; Figure 2 This is an architecture diagram of an abnormal behavior analysis method based on artificial intelligence according to the present invention. Detailed Implementation
[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0018] Example: Figures 1-2 As shown, an artificial intelligence-based method for analyzing abnormal behavior includes the following steps: S101. Multimodal data acquisition steps: Through multiple heterogeneous sensors deployed in the monitoring space, real-time dynamic data of the target user is continuously collected, and historical behavioral data of the user is obtained. The sensors include at least one or more combinations of millimeter-wave radar, AI smart camera, AI smart mattress, and smart water and electricity meter. S102. Steps for constructing a personalized behavior baseline: Based on the historical behavior data, a personalized behavior baseline model covering multiple behavioral dimensions is established for the target user through time series analysis. The behavioral dimensions include at least activity patterns, sleep patterns, bathroom usage patterns, and water and electricity usage patterns. S103. Real-time anomaly identification step: The collected real-time dynamic data is compared with the personalized behavior baseline model in real time, and the statistical deviation analysis algorithm is used to identify abnormal data points in at least one behavioral dimension. S104. Artificial intelligence analysis and confirmation steps: Input the abnormal data points into a pre-trained abnormal behavior classification AI model for in-depth analysis and classification to confirm whether a predefined abnormal behavior event has occurred. The event includes at least abnormal toilet use, abnormal home activities, abnormal sleep, and abnormal water and electricity use. S105. Warning Generation and Output Steps: When an abnormal behavior event is confirmed, an early warning message containing the event type, severity level, and evidence data is automatically generated and sent to the designated recipient.
[0019] Specifically, the personalized behavior baseline model is a dynamic model that uses statistical analysis to set a reference benchmark for each behavioral dimension, including the mean, standard deviation, and normal fluctuation range.
[0020] Specifically, the confirmation of abnormal toilet use is based on the real-time monitoring of entry frequency and single stay duration, which is compared with the corresponding toilet use patterns in the baseline model. When the deviation exceeds the preset statistical threshold, the AI model determines it to be abnormal.
[0021] Furthermore, the confirmation of abnormal home activities is achieved by integrating data from millimeter-wave radar and AI smart cameras to analyze the user's activity level, activity range, and static duration within a set time period, and comparing them with the individual's behavioral baseline. When the indicators deviate significantly, they are judged as abnormal.
[0022] Furthermore, the confirmation of sleep abnormalities is based on one or more of the following data collected by the AI smart mattress or vital sign monitoring radar: sleep duration, ratio of deep sleep to light sleep, frequency of turning over, heart rate, and respiratory rate. When these data deviate continuously from the individual's sleep baseline, it is considered abnormal.
[0023] Furthermore, the abnormal behavior classification AI model is a multi-classifier system, which includes one or more combinations of decision trees, support vector machines (SVM), and deep neural networks (DNN).
[0024] Specifically, the warning information also includes actionable suggestions generated by an AI model based on the anomaly type and severity from a pre-set knowledge base.
[0025] Preferably, the method is executed on an edge computing device to achieve local data processing and low-latency alerts.
[0026] Furthermore, it also includes a baseline update step: regularly incorporating validated normal behavior data into historical data and iteratively optimizing the personalized behavior baseline model to adapt to changes in user behavior.
[0027] like Figure 2 As shown, the system in this embodiment adopts a three-layer architecture of cloud-edge-device.
[0028] Multimodal sensing layer edge: The following sensor nodes are deployed within the user's residence; Millimeter-wave radar sensor: Installed in the center of the ceiling in the living room and bedroom, this sensor operates in the 65GHz band and can continuously detect the presence of people in the room, as well as subtle movements such as breathing, heartbeat, movement trajectory, movement speed, and stationary state in a non-contact and privacy-free manner.
[0029] AI smart camera: Installed at the entrance of the living room, the field of view covers the main activity area, but follows the privacy design principle and does not store or upload video streams by default. Its built-in edge AI chip can run lightweight human pose estimation and behavior recognition models such as OpenPose algorithm variants in real time, output structured data, and discard the original image immediately after local processing.
[0030] The AI smart mattress is placed under the user's mattress and has a built-in high-precision pressure sensor array that can monitor the user's status in and out of bed, the frequency of turning over during sleep, breathing rhythm, and estimate heart rate.
[0031] Smart water and electricity meters: Replace existing water and electricity meters, or add smart modules. Monitor the total water and electricity consumption and instantaneous flow rate / current of the residence.
[0032] Edge computing gateway: Deploy a high-performance edge computing gateway in the user's home, such as an industrial-grade embedded device based on ARM or x86 architecture; This gateway has the following core functions. Data aggregation: Real-time data streams from all sensors are received via Zigbee, Wi-Fi, or wired connections. The core algorithm runs and is equipped with an AI early warning engine.
[0033] Personalized Behavior Baseline Model: This model is not a simple database, but a set of parameters dynamically maintained by the algorithm.
[0034] Abnormal Behavior Classification AI Model: This is a lightweight neural network model that is pre-trained in the cloud and periodically distributed to the edge, such as a combination of a one-dimensional convolutional neural network and an attention mechanism.
[0035] The cloud service platform is deployed on a secure cloud server, including: Historical Data Lake: Long-term, secure archiving of all users' de-identified historical data, with personally identifiable information removed; Model training and optimization platform: Regularly uses massive amounts of anonymized data from the data lake to incrementally train and iterate the general-purpose abnormal behavior classification AI model, learning a wider range of abnormal patterns and improving the model's generalization ability and accuracy. The updated model is then securely distributed to each edge gateway. Early Warning Notification Center: Receives abnormal events confirmed by the edge gateway, manages the user-guardian relationship mapping, and distributes early warning information to designated guardians through various channels such as APP push, SMS, and telephone robots; it also provides an event management backend for guardians to view historical early warnings, evidence charts, and processing progress.
[0036] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An artificial intelligence-based method for analyzing abnormal behavior, characterized in that, Includes the following steps: S101. Multimodal data acquisition steps: Through multiple heterogeneous sensors deployed in the monitoring space, real-time dynamic data of the target user is continuously collected, and historical behavioral data of the user is obtained. The sensors include at least one or more combinations of millimeter-wave radar, AI smart camera, AI smart mattress, and smart water and electricity meter. S102. Steps for constructing a personalized behavior baseline: Based on the historical behavior data, a personalized behavior baseline model covering multiple behavioral dimensions is established for the target user through time series analysis. The behavioral dimensions include at least activity patterns, sleep patterns, bathroom usage patterns, and water and electricity usage patterns. S103. Real-time anomaly identification step: The collected real-time dynamic data is compared with the personalized behavior baseline model in real time, and the statistical deviation analysis algorithm is used to identify abnormal data points in at least one behavioral dimension. S104. Artificial intelligence analysis and confirmation steps: Input the abnormal data points into a pre-trained abnormal behavior classification AI model for in-depth analysis and classification to confirm whether a predefined abnormal behavior event has occurred. The event includes at least abnormal toilet use, abnormal home activities, abnormal sleep, and abnormal water and electricity use. S105. Warning Generation and Output Steps: When an abnormal behavior event is confirmed, an early warning message containing the event type, severity level, and evidence data is automatically generated and sent to the designated recipient.
2. The abnormal behavior analysis method based on artificial intelligence according to claim 1, characterized in that, The personalized behavior baseline model is a dynamic model that uses statistical analysis to set a reference benchmark for each behavioral dimension, including the mean, standard deviation, and normal fluctuation range.
3. The abnormal behavior analysis method based on artificial intelligence according to claim 1, characterized in that, The confirmation of abnormal toilet use is based on the real-time monitoring of entry frequency and single stay duration, which is compared with the corresponding toilet use pattern in the baseline model. When the deviation exceeds a preset statistical threshold, the AI model determines it to be abnormal.
4. The abnormal behavior analysis method based on artificial intelligence according to claim 1, characterized in that, The confirmation of abnormal home activities is achieved by integrating data from millimeter-wave radar and AI smart cameras to analyze the user's activity level, activity range, and static duration within a set time period, and comparing them with the individual's behavioral baseline. When the indicators deviate significantly, they are judged as abnormal.
5. The abnormal behavior analysis method based on artificial intelligence according to claim 1, characterized in that, The confirmation of sleep abnormalities is based on one or more of the following data collected by the AI smart mattress or vital sign monitoring radar: sleep duration, ratio of deep sleep to light sleep, frequency of turning over, heart rate, and respiratory rate. When these data deviate continuously from the individual's sleep baseline, it is considered abnormal.
6. The abnormal behavior analysis method based on artificial intelligence according to claim 1, characterized in that, The abnormal behavior classification AI model is a multi-classifier system, which includes one or more combinations of decision trees, support vector machines (SVM), and deep neural networks (DNN).
7. The abnormal behavior analysis method based on artificial intelligence according to claim 1, characterized in that, The warning information also includes actionable suggestions generated by an AI model based on the anomaly type and severity from a preset knowledge base.
8. The abnormal behavior analysis method based on artificial intelligence according to claim 1, characterized in that, The method is executed on an edge computing device to achieve local data processing and low-latency early warning.
9. The abnormal behavior analysis method based on artificial intelligence according to claim 1, characterized in that, It also includes a baseline update step: periodically incorporating verified normal behavior data into historical data and iteratively optimizing the personalized behavior baseline model to adapt to changes in user behavior.