Multi-dimensional early warning system for violent risks of drug addicts

Abstract

The invention relates to the technical field of stability monitoring, and discloses a multi-dimensional early warning system for violent risks of drug addicts. The system comprises wearable electrocardiogram monitoring equipment, a multi-modal data acquisition device, an administrator mobile terminal and a central processing computer, the wearable electrocardio monitoring equipment is used for monitoring and collecting physiological electrocardio characteristic data of the drug addict in real time; the multi-modal data acquisition device comprises the following modules. The system comprises a face recognition module, an emotion recognition module, a voice acquisition module and a motion capture module. The administrator mobile terminal is connected with the computer and collects data in real time. And the central processing computer receives and processes the collected data to judge the violent risk level. The system improves the violent risk prevention and intervention efficiency, guarantees the social safety and public order, and reduces the harm of violent behaviors of drug addicts to the society.

Description

Technical Field

[0001] The present application relates to the field of information technology, and in particular to a multidimensional early warning system for the violence risk of drug addicts. Background Art

[0002] Currently, monitoring the violence risk of drug addicts mostly relies on traditional single-data analysis methods, such as physiological testing or behavioral observation. These methods have significant limitations: first, a single data dimension makes it difficult to comprehensively assess risk; second, a lack of real-time performance leads to delayed early warnings; and third, intervention measures lack targetedness, making it difficult to effectively prevent the occurrence of violence risks. While intelligent monitoring technology offers high efficiency and real-time performance in data collection and analysis, it still suffers from data blind spots, the risk of misjudgment, and insufficient understanding of complex situations. Therefore, combining intelligent monitoring with human assistance is crucial. Human assistance can address areas not covered by intelligent monitoring equipment, reduce misjudgments, enhance the system's flexibility and adaptability, and increase management confidence in the system. In summary, there is an urgent need for a system that integrates multidimensional data, such as electrocardiogram, emotions, voice, daily life, and movement data, for real-time analysis. By combining intelligent monitoring with human assistance, the accuracy of violence risk prediction and the efficiency of intervention can be improved, thereby effectively safeguarding social security. Summary of the Invention

[0003] In order to solve the problem of accurately assessing and providing real-time warnings for drug addicts' violence risk based on their physiological electrocardiogram (ECG) data, emotional data, voice data, motion data, and daily life data, the present invention provides a multi-dimensional warning system for drug addicts' violence risk, which is characterized by comprising a wearable ECG monitoring device, a multimodal data acquisition device, an administrator's mobile terminal, and a central processing computer;

[0004] The wearable ECG monitoring device is used to monitor and collect physiological ECG characteristic data of drug addicts in real time, including heart rate and ECG waveform physiological data;

[0005] The multimodal data acquisition device includes the following modules:

[0006] A facial recognition module for real-time monitoring of facial expressions of drug addicts;

[0007] Emotion recognition module, used to identify the emotional trends of drug addicts;

[0008] Voice collection module, used to collect and analyze the voice content of drug addicts;

[0009] Motion capture module, used to monitor the behavior of drug addicts;

[0010] The administrator's mobile terminal is connected to the computer to collect data in real time;

[0011] The central processing computer receives and processes the collected data to determine the violence risk level.

[0012] The central processing computer determines whether the final violence risk score of the drug addict is greater than a preset value and sends an alarm to the administrator's mobile terminal based on the score classification.

[0013] Furthermore, the computer performs the following steps:

[0014] S1: The electrocardiogram (ECG) data, emotion data, voice data, and motion data of drug addicts are acquired in real time through a wearable ECG monitoring device worn by the drug addicts and a multimodal data acquisition device within the drug addicts' range of motion;

[0015] S2: Calculate the violence risk score of drug addicts through ECG data, emotion data, voice data, and motion data;

[0016] S3: Assess the violence risk level of drug addicts by analyzing their violence risk scores.

[0017] Furthermore, the violence risk level includes three levels: low risk, medium risk, and high risk.

[0018] Furthermore, the hierarchical warning includes the following steps:

[0019] The beneficial effects of the present invention are: by real-time monitoring of the physiological electrocardiogram data, emotional data, voice data, motion data, and life data of drug addicts, the drug addicts' signs of violence risk and violence risk status assessment data are automatically obtained, and these data are sent to the central processing computer system for graded early warning. When the drug addict is in an unstable state at the set violence risk level, and the violence risk status score is large and there are signs of preparation for violent behavior, the computer system can promptly identify and issue an alarm. After receiving the alarm, the management personnel can quickly take intervention measures, thereby effectively curbing or reducing the occurrence of violent incidents, improving the efficiency of prevention and intervention of violence risks, ensuring social security and public order, and reducing the harm caused to society by the violent behavior of drug addicts. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of the present invention.

[0021] Figure 2 This is a diagram of the architecture of the present invention. DETAILED DESCRIPTION

[0022] like Figure 1The present invention provides a multi-dimensional early warning system for drug addicts' violence risk, which aims to monitor and analyze drug addicts' electrocardiogram (ECG) data, emotional data, daily life data, voice data, and motion data in real time. The system includes the following components: a wearable ECG monitoring device, a multimodal data acquisition device, an administrator's mobile terminal, and a central processing computer.

[0023] The wearable ECG monitoring device (smart micro-ECG patch) can monitor and collect real-time physiological ECG data from drug addicts, including heart rate, ECG waveforms, and other physiological data, to reflect their health status and mood swings. This data can provide a basis for assessing the state of the drug addict's autonomic nervous system and, in turn, help determine whether they are in a state of high tension or excitement that could lead to violent behavior.

[0024] The multimodal data acquisition device is deployed in multiple environments such as drug addicts' residential communities, homes, and supervision sites, and includes the following modules:

[0025] Facial Recognition Module: Based on Convolutional Neural Network (CNN) technology, it is used to monitor the facial expressions of drug addicts in real time, analyze their emotional state, and determine whether they are showing any emotional changes such as anger and anxiety that may lead to violent behavior;

[0026] Emotion Recognition Module: Combining a convolutional neural network (CNN) and a long short-term memory network (LSTM), it is used to identify the emotional trends of drug addicts, capture their mood swings, and promptly detect the accumulation of emotions that may lead to violent behavior;

[0027] Voice Collection Module: This module uses beamforming technology and a noise reduction algorithm (NLA) to collect and analyze the voice content of drug addicts, detecting threatening or aggressive language, which can often be a precursor to violent behavior.

[0028] Motion capture module: Utilizes OpenPose technology to monitor the behavior of drug addicts and identify any abnormal behaviors such as aggressive movements, self-harm, or frequent movement, which may indicate an increased risk of violence.

[0029] In the implementation of the present invention, face recognition can adopt the technologies disclosed in the following patents:

[0030] CN202211449046.2 Face recognition model training method, face recognition method and device

[0031] CN202010020092.5 Face recognition model construction method, face recognition method and related device

[0032] CN202110918545.0 Face recognition model processing method, face recognition method and device

[0033] The following patent disclosures can be used in the implementation of the emotion collection module of the present invention:

[0034] CN202411207642.9 A multi-dimensional directional training system for AI dynamic emotion recognition

[0035] CN202311522880.4 Voice emotion recognition model training method, voice emotion recognition method and device

[0036] CN202310360529.3 Emotion recognition model training method and emotion recognition method, device and storage medium

[0037] During the implementation of the present invention, the voice collection module can adopt the technology disclosed in the following patents:

[0038] CN202122981400.3 A microphone voice system

[0039] CN201521081351.6 A long-distance voice acquisition device with combined sound and image positioning

[0040] CN202010905748.1 An image and voice acquisition terminal device and its working method

[0041] During the implementation of the present invention, the motion acquisition module can adopt the following patent disclosure technology

[0042] CN202110036317.0A human motion capture and analysis system based on RGBD camera

[0043] CN202210382446.X A human motion acquisition device

[0044] CN201420382187.1 A Stereoscopic Vision-Based Motion Capture and Feedback System. The administrator's mobile terminal, equipped with a manual monitoring module, can connect to a computer to collect and input real-time data from blind spots in intelligent sensing devices, such as the behavior and interactions of drug addicts in areas not covered by the device. Administrators can also receive early warnings about the risk of violence from drug addicts through the terminal, enabling them to take timely intervention measures.

[0045] The central processing computer, serving as the system's core processing unit, connects to the aforementioned components and is responsible for receiving and processing collected data. Using pre-set intelligent algorithms, it analyzes and evaluates multi-dimensional data, generates a violence risk score, and determines the violence risk level based on pre-set thresholds.

[0046] The central processing computer determines that when the final violence risk score of the drug addict is greater than the minimum preset value and classifies the scores based on the scores, it sends an alarm to the administrator's mobile terminal so that the administrator can take targeted intervention measures in time to effectively prevent the occurrence of violent incidents.

[0047] The wearable ECG monitoring device can be a dynamic ECG recorder VV-330 (Weiling (Hangzhou) Information Technology Co., Ltd.), the ECG data can be the long-term change trend of HRV, the emotional data can be the 60-second non-contact video analysis data of the Cloud Heart Test Physical and Mental Health AI Rapid Detector (HQ1.0 version), the multimodal data acquisition device can be a Hikvision camera, the violence warning control personnel can be compulsory drug rehabilitation personnel or community drug rehabilitation personnel, the administrator's mobile terminal can be a smartphone, and the administrator can be a community member, guardian, or family member.

[0048] When a drug addict is in an unstable state, has a high score for violence risk, and takes measures to prepare for violent behavior, the central processing computer promptly sends an alarm to the drug addict's manager, who then quickly takes emergency measures to curb or reduce the harm posed by the drug addict's violence risk.

[0049] Furthermore, the computer performs the following steps:

[0050] S1: Through wearable ECG monitoring devices worn by drug addicts, multimodal data collection devices within the drug addict's range of motion, and manual auxiliary monitoring modules, real-time ECG data, emotional data, daily life data, voice data, and movement data of drug addicts are collected. All collected data serves as the basis for risk assessment.

[0051] S2: Compare the collected ECG data, emotion data, voice data, movement data, and daily life data with a multidimensional training sample database. This database contains a large number of characteristic samples of drug addicts at different violence risk levels. Through this comparison, the system can intelligently identify the similarities between the current data and the sample data, and intelligently assign scores based on this information to calculate the drug addict's violence risk score.

[0052] S3: By analyzing the drug addict's violence risk score, the system ultimately assesses the drug addict's violence risk level. Based on pre-set grading criteria, the risk level is divided into three levels: low, medium, and high, providing managers with clear risk warnings.

[0053] During the implementation of this invention, we extracted "early warning factors" based on the abnormal emotions and behaviors prior to hundreds of violent incidents caused by drug addicts. According to the following tables, we categorized emotional, speech, daily life, movement, electrocardiogram, and daily life observation data into three groups of data features: indicator features 1, 2, and 3, in order of increasing risk. We also established a manual monitoring dimension and supplemented the blind spot data of intelligent sensing devices to form a multidimensional training sample database. Note: The indicator features of each dimension are arranged in tiers 1, 2, and 3, based on increasing risk.

[0054]

[0055]

[0056]

[0057]

[0058] During the implementation of the present invention, the life observation dimension indicators need to be manually input into the administrator's mobile terminal, and the other specific manifestations mentioned above are automatically obtained by automatically analyzing the emotional data, voice data, movement data, and electrocardiogram data of the measured person by the computer.

[0059] Experimental study on expert assessment of violence risk among drug addicts (validation and optimization)

[0060] 1. Experimental Purpose

[0061] Through expert evaluation, the basic weights, correction coefficients, and low, medium, and high-risk level thresholds in the dynamic assessment algorithm for violence risk of drug addicts are scientifically and reasonably determined to improve the accuracy and reliability of the assessment algorithm.

[0062] 2. Selection of experimental subjects

[0063] We have established a diverse team of experts, whose members cover the following areas:

[0064] 4 clinical medicine experts: They have extensive experience in drug rehabilitation treatment and physiological health assessment of drug addicts, and have conducted in-depth research on physiological changes caused by drug addiction, such as electrocardiogram abnormalities.

[0065] Psychology Expert 2: Focuses on the psychological health of drug addicts and the psychology of violent behavior, and is good at emotion assessment and psychological warning analysis.

[0066] Social Work Expert 2: Familiar with the community rehabilitation, living environment and social support system of drug addicts, and experienced in life observation and analysis.

[0067] Artificial Intelligence and Data Analysis Expert 2: Proficient in intelligent analysis algorithms, data mining and processing, and able to provide professional insights into dimensions such as action and voice intelligent analysis from a technical perspective.

[0068] Frontline staff of drug rehabilitation institutions 5: Directly contact drug addicts, understand their daily behavior, risk characteristics and intervention needs, and have practical experience.

[0069] 15 experts were invited to participate in the experiment to ensure the comprehensiveness and professionalism of the evaluation results.

[0070] 3. Experimental Procedure

[0071] (1) First round of evaluation

[0072] Data distribution: Provide experts with detailed descriptions of the dimensions of violence risk assessment for drug addicts (including the specific content and assessment methods of dimensions such as movement, voice, emotion, electrocardiogram, and daily life observation), existing basic weights, correction coefficients, and level threshold setting schemes, as well as case data on some typical drug addicts.

[0073] Independent assessment: Experts will independently assess the basic weights, correction factors and grade thresholds based on their professional knowledge and experience, and provide suggestions and opinions on modifications, with reasons.

[0074] Data collection: Collect expert evaluation forms and organize experts' comments and suggestions for revisions.

[0075] (2) Feedback and discussion

[0076] Summary of opinions: Systematically organize and analyze the experts' evaluation opinions to extract representative views and suggestions.

[0077] Feedback communication: Feedback the summarized opinions to the experts so that they can understand the views and suggestions of other colleagues.

[0078] Online / offline discussions: Organize experts to hold online meetings or offline seminars to conduct in-depth discussions on controversial issues, promote exchanges and exchanges of ideas among experts, and further improve the assessment plan.

[0079] (3) Second round of evaluation

[0080] Re-evaluation: Based on the first round of feedback and discussion results, experts will independently re-evaluate the basic weights, correction factors and level thresholds and adjust their judgment.

[0081] Data collection and analysis: The expert evaluation forms were collected again, and statistical methods were used to analyze the data. The mean value, standard deviation and other indicators of each parameter were calculated to evaluate the degree of concentration and dispersion of expert opinions.

[0082] Determine whether consensus has been reached: If the degree of dispersion of expert opinions is within an acceptable range (for example, the standard deviation of the basic weight and correction coefficient is less than the preset threshold, and the expert opinions on the level threshold are concentrated in a certain interval), the final basic weight, correction coefficient and level threshold are determined; if consensus is not reached, repeat the feedback and discussion and re-evaluation steps until the expert opinions tend to be consistent.

[0083] 4. Application of experimental methods

[0084] Delphi method: Through multiple rounds of anonymous evaluation and feedback, expert opinions are gradually concentrated, mutual interference between experts is reduced, the professional advantages of each expert are fully utilized, and the objectivity and scientific nature of the evaluation results are ensured.

[0085] Analytic Hierarchy Process (AHP): When determining the basic weights, each evaluation dimension is constructed into a hierarchical model. The weights of each dimension are calculated based on the experts' judgment of the relative importance of each dimension, making the weight determination process more systematic and reasonable.

[0086] 5. Experimental Results Analysis and Conclusion

[0087] Results Analysis: By conducting a rational analysis of the basic weights, correction coefficients, and level thresholds, combined with the actual needs and characteristics of violence risk assessment for drug addicts, the basis and advantages of each parameter setting are explained, and the following conclusions are drawn:

[0088] The present invention has finally determined the level correction value adjustment scheme based on fixed basic weights (action 25%, voice 20%, emotion 15%, electrocardiogram 20%, and daily life 20%) through preliminary experiments. The actual weight of each dimension in different risk levels is dynamically adjusted through the correction coefficient, and accurate assessment is achieved by combining the trigger threshold and time window:

[0089] Correction coefficient table for each level (core adjustment)

[0090]

[0091]

[0092] Level threshold

[0093] Low risk: total score ≤ 30 points

[0094] Intermediate risk: 31 points ≤ total score ≤ 60 points

[0095] High risk: total score ≥ 61 points

[0096] The present invention uses the above methods such as risk level feature library construction based on hundreds of cases, expert evaluation method, and practical case retrospective experimental verification to finally determine the following risk assessment algorithm:

[0097] The risk assessment algorithm used in the present invention is as follows:

[0098] 1. Quantitative definition of indicators:

[0099] 1. Sentiment Indicators

[0100] Emotion=1:Balance<40 or Energy>40 or Depression>50

[0101] Emotion=2:Anxiety>40 or Neuroticism>50 or Self-regulation<50 or Doubt>50

[0102] Emotion=3: Aggression>50 or Inhibition<15 or Stress>40

[0103] 2. Voice indicators

[0104] Voice = 1: Complaints about environmental pressures or the behavior of others more than twice a week; or repeated illogical monologues more than twice a week; or self-deprecation and annoyance when talking to others more than twice a week; or threatening innuendos in conversations without explicit violent content or actual actions.

[0105] Voice = 2: Frequently provoking or escalating conflicts with others, more than twice a week; verbally attacking or threatening others without obvious provocation, more than twice a week; or announcing one's behavior without regard for the consequences, more than twice a week;

[0106] Voice=3: Intense verbal attacks against others (including death threats); or explicit expressions of indiscriminate violent intent; or explicit expressions of suicidal thoughts; or disclosures of violent intentions or plans.

[0107] 3. Action indicators

[0108] Action = 1: Slightly disturbed sleep and daytime routine: reversing morning and evening rhythms, or frequently getting up at night for more than 2 hours (at least 2 times a week); or unconsciously knocking on the table or dropping small objects more than 2 times a week; or intentionally pushing down or destroying objects more than 2 times a week.

[0109] Action = 2: Frequent fist clenching, glaring, and aggressive gestures, more than twice a week; or aimlessly waving handheld instruments, more than twice a week; self-injury or self-harm with superficial scratches; or smashing objects or banging against walls, doors, windows, etc., or physical conflicts with others that can be stopped, more than twice a week.

[0110] Action = 3: Acts of self-harm or actions that seriously threaten life or health; damage to public safety facilities or intentional creation of public safety hazards; active attacks on others, seriously endangering their safety; or the collection and accumulation of flammable materials and the preparation of burning tools.

[0111] 4. ECG indicators

[0112] ECG=1: F value between 50-65 or HF value between 35-50 or LF / HF value between 1.0-2.0

[0113] ECG = 2: LF value between 65-80 or HF value between 20-35 or LF / HF between 2.0-3.5

[0114] ECG = 3: HRV relative power is severely increased or LF is greater than 80 or HF is less than 20 or LF / HF ratio is greater than 3.5

[0115] 5. Living Indicators

[0116] Life = 1: The individual has a stable family, work, or social support network, but has decreased interaction with family and friends, and is indifferent or perfunctory towards them; or is unwilling to actively cooperate with drug rehabilitation treatment more than twice a month; or has disorganized personal belongings (e.g., clothes thrown around, daily necessities placed in disorder, for more than two days); or is unusually vigilant about ordinary things, with mild persecution delusions or obsessive-compulsive symptoms (e.g., repeatedly checking doors and windows to see if they are closed, more than twice a day).

[0117] Life=2: Irritable and irritable for more than two consecutive days, headaches, tremors, and loss of impulse control during withdrawal; or causing panic and disrupting order in various ways during rehabilitation or public activities for more than two times; or talking on the phone with a drug addict and refusing to explain the content of the contact or at least twice a week; or refusing to cooperate with treatment for seven consecutive days.

[0118] Life=3: Severe hallucinations and delusions leading to uncontrolled behavior (such as chasing fantasy objects); or purchasing dangerous items (such as knives, gasoline, ropes, etc.), or single consumption exceeding 2,000 yuan without reasonable explanation; or secret meeting with known drug addicts or violent criminals.

[0119] II. Calculation Method for Violence Risk Assessment of Drug Addicts

[0120] 1. Function data screening

[0121] I {Emotion＝i} This function is used to mark whether the condition is met. If the event Emotion=i is met, its value is 1, if not, its value is 0, i.e.

[0122]

[0123] There are also:

[0124]

[0125] 2. Weight vector

[0126] Basic weight vector W = [0.15 0.20 0.25 0.20 0.20]

[0127] Characteristic index 1 correction vector A1 = [1.20 0.70 0.60 1.1 1.3]

[0128] Characteristic index 2 correction vector A2 = [1 1 1 1 1]

[0129] Characteristic index 3 correction vector A3 = [0.80 1.40 1.50 1.60 0.7]

[0130] Feature index 1 weight vector

[0131]

[0132] Feature index 2 weight vector

[0133]

[0134] Feature index 3 weight vector

[0135]

[0136] 3. Five-dimensional feature indicators

[0137] Indicator feature 1 is a matrix composed of the corresponding group data features

[0138] Indicator feature 2 is a matrix composed of the corresponding group data features

[0139]

[0140] Indicator feature 3 is a matrix composed of the data features of the corresponding group.

[0141]

[0142] 4. Risk score = 40W1I1+70W2I2+100W3I3

[0143] 5. Determine the violence risk level of drug addicts based on SCORE values:

[0144] Low risk: total score ≤ 30 points

[0145] Intermediate risk: 31 points ≤ total score ≤ 60 points

[0146] High risk: total score ≥ 61 points. For example: In one embodiment, Emotion = 1, Voice = 1, Action = 2, EGG = 2, life = 3 are substituted into the characteristic indicators to obtain:

[0147]

[0148] Substitute the risk score formula to obtain:

[0149]

[0150] =40*(0.189+0.147)+70*(0.25+0.20)+100*(0.111)

[0151] =13.44+31.5+11.1=56.04

[0152] The final score is 56.04, so the violence risk level of this embodiment is medium.

[0153] The weighted score is compared with the preset threshold score, and a three-level warning is output based on the score comparison result.

[0154] The present invention has the following beneficial effects: by monitoring drug addicts' physiological electrocardiogram (ECG) data, emotional data, voice data, movement data, and daily life data in real time, it automatically obtains data on their violence risk signs and violence risk status, and transmits this data to a central processing computer system for graded early warning. When a drug addict is unstable within a set violence risk level, with a high violence risk status score and signs of preparing for violent behavior, the computer system can promptly identify and issue an alarm.

Claims

1. A multi-dimensional early warning system for the risk of violence among drug addicts, characterized by: Including wearable ECG monitoring equipment, multimodal data acquisition device, administrator mobile terminal, central processing computer; The wearable ECG monitoring device is used to monitor and collect physiological ECG characteristic data of drug addicts in real time, including heart rate and ECG waveform physiological data; The multimodal data acquisition device includes the following modules: A facial recognition module for real-time monitoring of facial expressions of drug addicts; Emotion recognition module, used to identify the emotional trends of drug addicts; Voice collection module, used to collect and analyze the voice content of drug addicts; Motion capture module, used to monitor the behavior of drug addicts; The administrator's mobile terminal is connected to the computer to collect data in real time; The central processing computer receives and processes the collected data to determine the violence risk level; The central processing computer determines whether the final violence risk score of the drug addict is greater than a preset value and sends an alarm to the administrator's mobile terminal based on the score classification.

2. A multi-dimensional early warning system for the risk of violence among drug addicts as claimed in claim 1, characterized in that: The computer performs the following steps: S1: The electrocardiogram (ECG) data, emotion data, voice data, and motion data of drug addicts are acquired in real time through a wearable ECG monitoring device worn by the drug addicts and a multimodal data acquisition device within the drug addicts' range of motion; S2: Calculate the violence risk score of drug addicts through ECG data, emotion data, voice data, and motion data; S3: Assess the violence risk level of drug addicts by analyzing their violence risk scores.

3. A multi-dimensional early warning system for the risk of violence among drug addicts as claimed in claim 1, characterized in that: The violence risk levels include low risk, medium risk, and high risk.

Images