Automatic emergency safety protection method and system

Abstract

The invention provides an automatic emergency safety protection method and system, and belongs to the technical field of automatic emergency safety protection, and the system comprises an interaction module, a sensing module, an information processing module, an information storage and transmission module, a protection measure execution module and the like. According to the method, the whole emergency process can be automatically completed without manual starting and intervention, various possible dangerous situations can be coped with by combining various sensing and information processing methods, and various appropriate emergency measure combinations are adopted according to the confidence and the critical degree of the identified dangerous situations, so that the safety of the system is improved. And the applicability and the success rate of emergency protection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic emergency safety protection, and in particular to a method and system for automatic emergency safety protection. Background Art

[0002] In daily life, people sometimes encounter emergencies that threaten their personal safety or health, requiring them to respond, seek help, or call the police. However, the person may be unconscious, incapacitated, or otherwise hindered by external forces, preventing them from taking appropriate safety precautions. For example, someone may suffer a stroke, fall, or become unconscious, requiring immediate emergency assistance. However, no one nearby recognizes the emergency situation in time. Someone may experience a heart attack or other unexpected illness and be aware and partially able to dial 120, but due to physical weakness or unconsciousness, they may be unable to complete the call and provide their exact location. Someone may be kidnapped, threatened, or harmed by a robber and want to call 110, but fear or coercion prevents them from completing the process. Someone may be in a dangerous situation, such as electric shock, drowning, or traffic accident, requiring assistance, but they may be unable to effectively call for help and no one nearby may recognize the situation or provide effective assistance. When faced with any of these situations, currently available methods and systems are insufficient to fully and effectively address them. There is an urgent need in the art for a system that can be carried or worn, automatically monitors and identifies various dangerous situations, and automatically initiates and implements appropriate emergency safety measures. The present invention aims to provide such a method and system. Summary of the Invention

[0003] The purpose of the present invention is to provide a system and method that can be carried or worn, can be applied to a variety of dangerous situations, and can automatically identify, automatically activate and automatically complete appropriate emergency safety protection measures.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] An automatic emergency safety protection system, comprising:

[0006] Interaction module, used for users to interact with the system;

[0007] Perception module, including visual perception unit, audio perception unit, lidar, motion inertial perception unit, physiological indicator perception unit and position perception unit;

[0008] An information processing module is used to process the information sensed by each sensing unit to determine whether a dangerous situation has occurred and to determine its confidence level and criticality. The information processing module includes:

[0009] Visual information processing unit, audio information processing unit, physiological index information processing unit, motion inertia information processing unit and comprehensive information processing unit;

[0010] An information storage and transmission module, for storing and transmitting a variety of information perceived, interacted, and generated, including a storage unit and a transmission unit; the storage unit includes: one or more combinations of electrical storage subunits, magnetic storage subunits, optical storage subunits, electromagnetic storage subunits, and semiconductor storage subunits; the transmission unit includes a wired transmission subunit and / or a wireless transmission subunit; the wired transmission subunit includes but is not limited to: any one or more combinations of transmission units such as USB1.0 / 2.0 / 3.x, MicroUSB, MiniUSB, serial interface, parallel interface, etc.; the wireless transmission subunit includes but is not limited to: any one or more combinations of 2G / 3G / 4G / 5G, Bluetooth, Star Flash, infrared, hotspot, NB-IoT, Rola, Zigbee, MavLink, WIFI, NFC, GPRS, GSM, and Ethernet;

[0011] The protective measures execution module includes two or more of a communication unit, an automatic alarm or automatic emergency call unit, a voice broadcast unit, and an audio-visual and electrical unit.

[0012] Optionally, the visual information processing unit combines visual perception with image recognition and deep learning technology to process and identify visual information to detect whether a dangerous situation occurs and assign confidence and criticality values ​​to the dangerous situation. The specific steps are as follows:

[0013] Visual data preprocessing: preprocessing the original images and video streams obtained by the visual perception unit, including image noise reduction, dynamic range adjustment, ROI cropping, using a deblurring algorithm to handle motion blur, and improving image quality in low-light environments through histogram equalization;

[0014] Object detection, feature extraction, and scene recognition, including:

[0015] Dangerous goods identification, based on the improved model of YOLOv8, performs real-time target detection of dangerous goods;

[0016] Human posture and behavior analysis: using OpenPose and MediaPipe algorithms to extract the coordinates of 135 key points on the human body, construct skeletal motion trajectories, analyze human posture and facial expressions, and identify abnormal movements;

[0017] Time series analysis and recognition, combined with LSTM / 3D-CNN models, analyzes action sequence features in consecutive frames and detects falls based on posture angles;

[0018] Dangerous scene identification, including:

[0019] Scene semantic segmentation, using U-Net and DeepLab models to segment flames, smoke, water ripples, or facial expressions;

[0020] Drowning detection uses a water surface segmentation algorithm to delineate dangerous areas; optical flow method analyzes abnormal trajectories of limb struggles;

[0021] Fire identification uses the YOLO-Fire model to detect flames and smoke in real time, and thermal imaging data to assist in identifying high-temperature areas;

[0022] Violent conflict detection uses a posture estimation model to detect changes in the angle between the head and torso; analyzes facial expressions and acute illnesses through changes in facial microvascular color;

[0023] To assign values ​​to dangers, the visual information processing unit will assign values ​​to the identified dangerous situations, that is, the confidence and criticality of a certain danger will be assigned respectively. The assignment range is 0-100. According to the density and intensity of the dangerous features in the identification of the above dangers, the confidence and criticality will be assigned respectively according to the preset rules.

[0024] Optionally, the audio information processing unit combines automatic speech recognition technology, deep learning technology, and natural language processing technology to process and identify audio information to detect whether a dangerous situation has occurred and assign confidence and criticality values ​​to the dangerous situation, as follows:

[0025] Speech signal processing and text conversion: First, the speech data signal obtained by the audio perception unit is preprocessed. The signal is converted into a multidimensional vector through noise reduction, echo cancellation, speech enhancement, framing, and Fourier transform, and the Mel-frequency cepstral coefficient features are extracted. Then, factor and text decoding is performed. The features are mapped to phonemes through a deep neural network acoustic model. The factors are then combined with the language model to form words and sentences, and then the text is output.

[0026] Hazard identification includes:

[0027] Acoustic feature extraction, anomaly detection, and sentiment analysis: MFCC algorithms are used to extract the spectral characteristics of sound. The audio information processing unit can identify acoustic features that exceed the normal range. Natural Language Processing (NLP) models are used to analyze the emotions and feelings in information, and abnormal fluctuations in acoustic features are also used to determine emotional states.

[0028] Keyword extraction, preset sensitive word library, and rapid screening of high-risk content through regular expressions or string matching;

[0029] The system establishes a multi-level keyword library, including primary keywords and secondary keywords;

[0030] The system uses dynamic time warping algorithm to achieve fuzzy matching and supports dialect and accent recognition;

[0031] Contextual association detection, combined with the conversation process, identifies contradictory statements and threatening logic, and builds a context model through the LSTM neural network to identify typical conversation patterns in dangerous scenarios;

[0032] Model recognition and machine learning: training classification models based on historical dangerous conversation data to identify potential threat patterns;

[0033] Use online learning framework to continuously optimize the model;

[0034] Assign values ​​to dangers. The audio information processing unit will assign values ​​to the identified dangerous situations, that is, assign values ​​to the confidence and urgency of a certain danger. The assignment range is 0-100. The system will assign values ​​to the confidence and urgency according to the density and intensity of the dangerous features in the identification of the above dangers and the preset rules.

[0035] Optionally, the motion inertia information processing unit combines sensor data feature analysis with intelligent algorithm processing to process and identify motion inertia information to detect whether a dangerous situation has occurred and assign confidence and criticality values ​​to the dangerous situation. The logical steps are as follows:

[0036] Preprocessing optimization, including:

[0037] Noise reduction: Kalman filtering eliminates sensor drift and is combined with a sliding window to segment short-term behavior data;

[0038] Coordinate calibration: The magnetometer helps establish a global coordinate system to reduce the impact of placement differences;

[0039] Time domain / frequency domain fusion: Extract acceleration mean, variance, and zero-crossing rate, combined with wavelet packet energy analysis;

[0040] Motion feature extraction and algorithm recognition of dangerous scenes, including:

[0041] Faint / Fall Detection:

[0042] Feature extraction: vertical acceleration drops sharply combined with horizontal acceleration approaching zero, supplemented by attitude angle mutation;

[0043] Algorithm optimization: Using threshold judgment and decision tree fusion model, combined with environmental data calibration, to improve accuracy;

[0044] Seizure Identification:

[0045] Feature extraction: 5-10 Hz high-frequency periodic jitter in the acceleration signal is combined with the gyroscope to capture limb symmetry abnormalities;

[0046] Algorithm logic: FFT frequency domain analysis extracts the main frequency component, compares it with the pre-stored epilepsy waveform library, and introduces LSTM to capture the temporal characteristics of pre-ictal auras;

[0047] Foreign body asphyxiation detection:

[0048] Movement characteristics: severe neck shaking, abnormal periodic chest and abdominal movements caused by respiratory muscle spasms;

[0049] Algorithm strategy: Analyze acceleration spectrum characteristics and use a bidirectional LSTM model to capture the temporal characteristics of presymptomatic asphyxia.

[0050] Electric shock identification:

[0051] Motor characteristics: high-frequency tremors and sudden limb stiffness caused by tonic muscle contraction;

[0052] Algorithm optimization: Wavelet packet decomposition is used to extract myoelectric interference features, and cross-validation is performed with environmental electric field sensor data;

[0053] Brute force attack detection:

[0054] Feature enhancement: multi-axis impact peak, limb collision frequency, and fusion of environmental sound wave characteristics;

[0055] Model optimization: Improve the dynamic time warping algorithm and introduce an attention mechanism to improve template matching accuracy;

[0056] Assigning values ​​to dangers: The motion inertia information processing unit will assign values ​​to identified dangerous situations, that is, the confidence and urgency of a certain danger will be assigned values ​​respectively. The assigned values ​​range from 0 to 100. The system will assign values ​​to the confidence and urgency according to the density and intensity of the dangerous features in the identification of the above dangers and the preset rules.

[0057] Optionally, the physiological indicator information processing unit is used to process the physiological indicator information to find out whether a dangerous situation occurs and assign a confidence level and a criticality level to the dangerous situation. The data processing flow is as follows:

[0058] Signal preprocessing: Kalman filtering is used to eliminate signal interference; Z-score normalization is used to process the dimensional differences of different sensors;

[0059] Identify and process time domain features and frequency domain features;

[0060] Hierarchical decision-making, including:

[0061] Primary screening: threshold judgment; Secondary verification: SVM classifier analysis of multi-parameter correlation;

[0062] Physiological feature extraction and algorithm recognition of dangerous scenarios, including:

[0063] Electric shock detection:

[0064] Feature extraction: muscle rigidity and tremor trigger high-frequency heart rate fluctuations; sudden drop in skin resistance combined with abnormal breathing rate;

[0065] Algorithm strategy: Wavelet packet decomposition is used to extract myoelectric interference features, and cross-validation is performed by integrating environmental electric field sensor data; real-time monitoring of skin resistance changes;

[0066] Myocardial infarction detection:

[0067] Feature extraction: ECG ST segment deviation + sudden drop in blood pressure; decreased blood oxygen saturation accompanied by cold sweat physiological indicators;

[0068] Algorithm strategy: 1D-CNN model: processes ECG signal waveform features and identifies ST segment morphological abnormalities; graph neural network: integrates ECG, blood pressure, and blood oxygen data to improve prediction specificity;

[0069] Seizure Identification:

[0070] Feature extraction: heart rate surge accompanied by respiratory disorder; blood oxygen saturation drops sharply;

[0071] Algorithm optimization: LSTM time series model captures the gradual change pattern of physiological signals in the 5 minutes before the onset of the attack; wavelet packet decomposition: extracts the 5-10Hz high-frequency components in the EEG signal and integrates them with the jitter characteristics of the inertial sensor;

[0072] Violence / robbery identification:

[0073] Feature extraction: Adrenaline surges cause heart rates > 120 bpm; skin conductivity surges; respiratory rate > 30 breaths / minute + rapid drop in blood oxygen saturation; multi-source verification, combined with inertial sensors to detect intense limb movements;

[0074] Model identification: Support vector machine, classifying normal and stress states; dynamic time warping, matching preset physiological response templates of violent scenes;

[0075] Faint / Loss of Consciousness Detection:

[0076] Core indicators: sudden drop in heart rate, sharp drop in blood pressure, abnormal respiratory rate;

[0077] Algorithm logic: Real-time monitoring of combined sudden changes in heart rate and blood pressure through a dynamic threshold model; multi-sensor fusion combined with acceleration sensors to determine sudden changes in body position;

[0078] Drowning detection:

[0079] Core indicators: sudden drop in respiratory rate, sharp drop in blood oxygen saturation, and loss of body temperature;

[0080] Algorithm optimization: Through the decision tree model, jointly judge the coordinated abnormalities of respiration, blood oxygen, and body temperature;

[0081] Assign a value to the danger:

[0082] The physiological indicator information processing unit will assign values ​​to the identified dangerous situations, that is, the confidence and criticality of a certain danger will be assigned respectively, and the assigned values ​​range from 0 to 100. The system will assign values ​​to the confidence and criticality respectively according to the density and intensity of the dangerous characteristics in the identification of the above dangers and the preset rules.

[0083] Optionally, the comprehensive information processing unit is used for comprehensive information processing. On the basis of the visual information processing, audio information processing, physiological indicator information processing and other information processing, the information and data are comprehensively processed to assign a comprehensive value to the overall confidence and urgency of the dangerous situation, so as to form a judgment on the comprehensive confidence and comprehensive urgency of the dangerous situation, and provide a basis for whether to initiate emergency measures and which emergency measures to initiate subsequently.

[0084] An automatic emergency safety protection method, comprising:

[0085] setup, monitoring, automatic startup and execution;

[0086] The settings include:

[0087] Through the interactive module, users input some important relevant information into the system in advance, set and select the dangerous situations to be focused on and a series of emergency plans;

[0088] Important relevant information to be entered includes the user's name, gender, age, height, weight, any medical conditions, contraindications, mobile phone number, address, information of relatives and friends, emergency contact information of relatives and friends, as well as the user's personal image, voice and other information to facilitate identity confirmation and authentication in emergency situations;

[0089] Setting and selecting a series of emergency plans, including selecting which dangerous situations to guard against, pre-selecting a priority emergency plan from the system's built-in series of plans for certain dangerous situations, and setting parameter combinations of multiple element options in the emergency plan to suit the user's own situation or personal preferences. If the user does not set and select in advance, the system will automatically select the system default settings;

[0090] The monitoring includes:

[0091] The automatic emergency safety protection system continuously monitors the user's condition and the surrounding environment in standby mode, and detects whether there is a dangerous situation through perception, analysis and judgment;

[0092] The perception process is to perceive and measure the images, dynamics, sounds and physiological indicators of the user and the surrounding environment through the system's visual perception unit, audio perception unit, physiological indicator perception unit, position perception unit and motion inertia perception unit; each perception unit then transmits the perceived data information to the information processing unit for processing. The information processing unit uses models and algorithms to analyze various data information and form a judgment, thereby discovering whether a security incident has occurred, such as personal injury, kidnapping, robbery, animal attack, fire, drowning, electric shock, car accident, fainting, myocardial infarction, epilepsy, foreign body suffocation and other dangerous situations;

[0093] The four information processing units, namely the visual information processing unit, the audio information processing unit, the motion inertia information processing unit, and the physiological index information processing unit, respectively assign values ​​to the confidence and criticality of 12 dangerous situations. In this way, the system is simultaneously monitoring 4*12*2=96 values. The comprehensive information processing unit then calculates the comprehensive confidence and comprehensive criticality of the 12 dangerous situations, thus obtaining 24 comprehensive values. If any of these 24 comprehensive values ​​reaches a preset threshold, the corresponding emergency procedure will be automatically initiated.

[0094] The automatic startup includes:

[0095] Once the information processing unit determines with a high degree of confidence that a dangerous situation has occurred and the situation is critical, it will automatically issue instructions to the corresponding execution unit according to the degree of criticality to activate the preset corresponding emergency procedures;

[0096] The subsequent response measures that are automatically activated are divided into different dangerous situations, different criticality levels, and different confidence levels. The system has pre-set default settings, which can also be modified by the user in the settings. For certain dangerous situations, certain criticality levels, and certain confidence levels, certain response measures can be set to be automatically activated immediately; for certain dangerous situations, certain criticality levels, and certain confidence levels, certain response measures can be set to be automatically activated after inquiry.

[0097] The execution includes:

[0098] After the system starts the emergency measures combination corresponding to a specific state, the corresponding emergency protection measures execution unit will automatically execute the series of emergency measures combinations.

[0099] Compared with the prior art, the present invention has the following beneficial effects:

[0100] 1. "Automatically start" the emergency procedure. The emergency method and portable device described in this embodiment use AI algorithms to monitor the status of the parties and their environment at any time, automatically identify and determine dangerous situations, and then automatically start the emergency safety protection program without manual activation. Currently available emergency alarm methods and equipment basically require manual activation. Even the so-called "one-button alarm" still requires manual triggering of "one button" to start the alarm program. This type of emergency method and equipment that requires manual operation to start cannot effectively function when encountering coma, disability, or obstruction by external forces. Methods and equipment that automatically identify and determine dangerous situations and automatically start a suitable series of emergency procedures can better cope with multiple types of dangerous situations.

[0101] 2. Automatically complete the "whole process" of emergency rescue. The current degree of automation of calling 110 for emergency help or 120 for emergency medical help is not very high. Even if some calls can be made relatively automatically, after the call is connected, human communication is still required to inform the police situation, medical condition, or specific people, places, events and other information to complete the whole process of calling the police for help. The emergency method and portable device described in this embodiment can automatically complete the communication and transmission of information through AI and automation programs. The "whole process" of emergency rescue can be completed without the human effort of the parties involved. In the event of an emergency, when the parties involved are in a coma, disabled, panicked, or have limited behavioral ability, the problem can be solved more effectively.

[0102] 3. It can adapt to a variety of situations intelligently. Traditional one-button alarms and emergency rescues are usually only applicable to a single situation. For example, if you encounter bad people and bad things, you need to call 110, if you suddenly get sick, you need to call 120 for emergency treatment, if you encounter a fire, you need to call 119 for emergency treatment, and if you encounter a drowning risk, you need to seek help from people around you. These different single situations require different corresponding emergency protection measures. Previous equipment and methods are difficult to automatically apply to a variety of possible situations that may occur in reality without human intervention. The emergency method and portable device described in this embodiment can intelligently identify different dangerous situations through AI and automated programs, and intelligently and automatically take a suitable series of emergency safety protection measures, which can automatically respond to a variety of dangerous situations.

[0103] 4. The innovative use of a method that integrates multiple sensing units and information processing models has improved the accuracy of comprehensive judgments on the occurrence of dangerous situations. It can more accurately judge the confidence and urgency of dangerous situations, and innovatively initiate corresponding more appropriate emergency measures based on the different comprehensive confidence and comprehensive urgency of dangerous situations.

[0104] 5. Simultaneous implementation of multiple emergency measures improves the success rate of emergency protection. Traditional emergency methods and equipment typically only implement a single emergency protection measure for a given dangerous situation. However, the emergency method and portable device described in this embodiment automatically implement multiple emergency measures simultaneously, such as remote calling, notifying friends and family, and seeking help from nearby personnel, for a given dangerous situation. This means that a suitable combination of emergency measures is implemented, and as long as one of the measures is effective, it is considered a success, effectively improving the success rate of the entire emergency process. BRIEF DESCRIPTION OF THE DRAWINGS

[0105] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0106] Figure 1 This is a schematic diagram of the structure of the automatic emergency safety protection system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0107] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0108] The purpose of the present invention is to provide a system that can be carried or worn, can automatically monitor and identify a variety of dangerous situations, and can automatically activate and complete a series of appropriate emergency safety protection measures.

[0109] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0110] Example 1:

[0111] This embodiment provides a method and system for automatic emergency safety protection, such as Figure 1 As shown in the figure, the composition of the automatic emergency safety protection system is as follows:

[0112] The automatic emergency safety protection system provided in this embodiment is a portable device that can exist as a standalone device or be fully or partially integrated into one or more other portable devices, such as smart glasses, smart watches, smart bracelets, smartphones, necklaces, pendants, hats, helmets, belts, and other portable or wearable devices, to protect the user's safety. The device consists of an interaction module, a perception module, an information processing module, a storage and transmission module, a protective measure execution module, a control module, and other modules. Each module is composed of one or more software and hardware components.

[0113] 1. Interaction module.

[0114] This unit includes input and output subunits, as well as a multimedia display. It's used for user interaction with the device, such as setting up various emergency plans, selecting elements within those plans, and entering personal information and contact information, identity verification information such as personal image, voice, and fingerprint, as well as the contact information of emergency contacts or friends and family. The interaction unit can reside in a separate device or be integrated into another portable or wearable device, completing the interaction process through a communication connection with the device.

[0115] 2. Perception module.

[0116] It includes categories such as visual perception unit, audio perception unit, lidar, motion inertial perception unit, physiological indicator perception unit, position perception unit, etc., which are selected and configured according to the specific functions and grade requirements of the equipment.

[0117] (1) Visual perception unit. This is mainly a recording unit that can automatically and quickly record videos and images. The recording device is equipped with one or more high-definition shooting lenses (such as a camera that supports 1080P / 4K high-resolution), and can also be equipped with an automatic fill light device that can automatically focus or track the target. In low-light environments, even if the person, object, and surrounding environment to be recorded are not accurately recorded, various visual information can be obtained well.

[0118] (2) Audio perception unit. This is mainly a recording device that can effectively record various sounds from the surrounding area and can also effectively identify the voiceprint characteristics of various sounds.

[0119] (3) Physiological indicator sensing unit. It can measure one or more physiological indicators of the user, such as heart rate, blood oxygen, blood pressure, body temperature, and respiratory rate.

[0120] (4) Position sensing unit. It can accurately locate the device and its user, such as Beidou positioning, GPS positioning, base station positioning, electronic map positioning, etc., to accurately detect the user's location.

[0121] (5) Motion inertia sensing unit. It can sense the user's motion state and is used to determine whether the user has fallen, had a convulsion, etc.

[0122] (6) Other sensing units. Some high-end equipment can also be equipped with sensing units such as lidar, thermal imaging cameras, odor sensing, toxic gas detection, high temperature detection, humidity detection, and PM2.5 detection.

[0123] 3. Information processing module.

[0124] It is an intelligent processing unit that contains multiple models, algorithms, indicators, parameters and multiple response procedures. It processes the information perceived by each sensing unit to determine whether a dangerous situation has occurred and its confidence and degree of criticality.

[0125] The various models, algorithms, indicators, parameters and response procedures contained in the information processing unit have undergone extensive training and verification in advance. Their sensitivity and confidence meet the requirements of actual applications and are adaptable to various application environments.

[0126] The information processing units include visual information processing units, audio information processing units, motion inertia information processing units, physiological index information processing units, comprehensive information processing units, etc. Each information processing unit exists independently or is integrated into one or several large processing units.

[0127] (1) Visual information processing unit

[0128] The visual information processing unit described in this embodiment is used to process and identify visual information to detect whether a dangerous situation occurs and assign a confidence level and a criticality level to the dangerous situation.

[0129] It mainly combines visual perception, image recognition and deep learning technology to judge the occurrence of dangerous situations. Its core principle and logical steps are as follows:

[0130] 1) Visual data preprocessing:

[0131] The original images and video streams obtained by the visual perception unit are preprocessed, including: image denoising, dynamic range adjustment, ROI (region of interest) cropping, motion blur processing using the deblurring algorithm (DeBlurGAN), and improving image quality in low-light environments through histogram equalization.

[0132] 2) Object detection, feature extraction and scene recognition:

[0133] ①Identification of dangerous goods

[0134] An improved model based on YOLOv8 performs real-time target detection on dangerous objects such as knives and guns.

[0135] ②Human posture and behavior analysis:

[0136] Use OpenPose, MediaPipe and other algorithms to extract the coordinates of 135 key points of the human body, construct skeletal motion trajectories, analyze human posture and facial expressions, and identify abnormal movements (such as punching and pushing in violent conflicts).

[0137] Time series analysis and recognition: Combined with the LSTM / 3D-CNN model, it analyzes action sequence features in consecutive frames (such as the sudden fall during fainting) and detects falls based on posture angles (torso-ground angle <25° for 8 seconds) (with an accuracy rate of 93% for distinguishing falls from squats).

[0138] ③ Dangerous scene identification solution:

[0139] Scene semantic segmentation: Use models such as U-Net and DeepLab to segment features such as flames, smoke (fire), water ripples (drowning), or facial expressions.

[0140] Drowning Detection: Features include violent water surface fluctuations, floating figures, and the absence of voluntary movement. Technology: A water surface segmentation algorithm identifies dangerous areas (such as deep water in a swimming pool); optical flow analysis analyzes abnormal trajectories of struggling limbs.

[0141] Fire Identification: Features include dynamic flame texture, smoke diffusion, and temperature anomalies. Technology: The YOLO-Fire model detects flames and smoke in real time; thermal imaging data assists in identifying high-temperature areas.

[0142] Violent conflict detection: Features include crowded areas, physical contact, and rapid movement. Technology: A skeletal keypoint-based action classification model identifies punches, kicks, and other movements; optical flow analysis identifies abnormal group movement directions (e.g., running in all directions).

[0143] Faint / coma detection: Characteristics: A sudden change in posture from upright to lying down, or prolonged inactivity. Technology: A posture estimation model detects changes in head-to-torso angles; facial microvascular color changes are used to analyze facial expressions and acute illness.

[0144] ④Multimodal data fusion and online learning mechanism:

[0145] The information fusion of multiple sensing units and multimodal data fusion can improve the accuracy and confidence of identifying dangerous situations. For example, in addition to identifying water surface fluctuations, human floating posture, and no autonomous movement, drowning detection also combines audio sensing unit data to synchronously detect the splashing sound and cry for help in the drowning scene, which can improve the confidence of detection; violent behavior detection logic, in addition to detecting continuous punching (inter-frame displacement>30 pixels), sudden increase in bone acceleration (>5m / s 2) can also detect impact sounds and cries of pain, improving judgment accuracy. Multi-sensory and multimodal fusion will be discussed in detail in the section on integrated information processing. Here, we will only discuss the information processing of a single sensory unit.

[0146] The online learning mechanism is designed to improve model capabilities, such as incremental updates to the dangerous scenario library (adding multiple samples every month) and a closed-loop feedback loop for false alarm cases (reducing the false alarm rate from 6.7% to 1.2%).

[0147] 3) Assign a value to the risk:

[0148] The visual information processing unit assigns values ​​to identified dangerous situations, specifically the confidence and criticality of the danger, ranging from 0 to 100. The system assigns these values ​​based on the density and intensity of the dangerous features detected, according to pre-set rules. For example, if a possible "fainting" danger is detected, the system assigns a confidence value of syz = 75 and a criticality value of syj = 80, meaning the confidence is 75% (high probability) and the criticality is 80% (relatively critical).

[0149] The purpose of assigning values ​​to hazards is to conduct comprehensive processing of subsequent information data and trigger corresponding emergency measures based on the size of the values.

[0150] (2) Audio information processing unit:

[0151] The audio information processing unit described in this embodiment is used to process and identify audio information to find out whether a dangerous situation occurs and assign a confidence level and a criticality level to the dangerous situation.

[0152] It mainly combines automatic speech recognition (ASR) technology, deep learning technology and natural language processing (NLP) technology. Its core principle logic is as follows:

[0153] 1) Speech signal processing and text conversion:

[0154] The voice data signal obtained by the audio perception unit is first preprocessed, mainly through noise reduction, echo cancellation, voice enhancement, framing (millisecond-level cutting), Fourier transform and other steps to convert the signal into a multidimensional vector, and extract features such as Mel-frequency cepstral coefficients (MFCC).

[0155] Then, phoneme and text decoding is performed, and the features are mapped to phonemes through the deep neural network acoustic model. The phonemes are then combined into words and sentences in combination with the language model (Transformers, RNN), and then the text is output.

[0156] 2) Hazard identification:

[0157] ①Acoustic feature extraction, anomaly detection and sentiment analysis:

[0158] By extracting the spectral characteristics of sound using algorithms such as MFCC (Mel-Frequency Cepstral Coefficients), the audio information processing unit can identify acoustic features that exceed normal ranges. Examples include high-frequency screams (sudden energy surges above 2000Hz) and abnormal breathing sounds (wheezing characteristic of asthma attacks). Natural Language Processing (NLP) models (LSTM and BERT) are used to analyze the emotions and feelings in information. Abnormal fluctuations in acoustic features (such as pitch and speech rate) can also be used to determine emotional state. High-intensity emotions such as anger and fear are often associated with conflict and danger.

[0159] ②Keyword extraction mechanism:

[0160] The system has a pre-set sensitive word library (such as violence, threat, alarm, and help-related terms) and uses regular expressions or string matching to quickly filter out high-risk content. For example, words such as "murder," "call 110," and "drowning" will significantly increase the confidence and severity of dangerous situations.

[0161] The system establishes a multi-level keyword library, including:

[0162] First-level keywords: help, fire, alarm (dangerous high confidence, high urgency);

[0163] Secondary keywords: Don’t be impulsive, Run fast (medium confidence, medium criticality);

[0164] The system uses the dynamic time warping (DTW) algorithm to achieve fuzzy matching and supports dialect and accent recognition.

[0165] ③Contextual association detection:

[0166] Based on the conversation process, the system identifies contradictory statements (e.g., semantic conflicts between preceding and following sentences) and threatening logic (e.g., "If you don't give me money, then..."). A contextual model is built using an LSTM neural network to identify typical conversation patterns in dangerous scenarios. For example, if the phrase "help" is detected and background noise contains "crying" and lasts for more than 6 seconds, it is considered a serious personal injury incident.

[0167] ④Pattern recognition and machine learning

[0168] Classification models are trained based on historical dangerous conversation data to identify potential threat patterns. For example, supervised learning can be used to distinguish between normal conversations and those involving robbery or violence.

[0169] Use an online learning framework to continuously optimize the model: for example, update the regional dialect library weekly (supporting 50+ dialect variants); expand the acoustic features of new dangerous scenarios every quarter; and so on.

[0170] 3) Assign a value to the risk:

[0171] The audio information processing unit assigns values ​​to identified dangerous situations, specifically the confidence and urgency of a particular danger, ranging from 0 to 100. The system assigns these values ​​based on the density and intensity of the dangerous features detected during the identification process, according to pre-set rules. For example, if a dangerous situation with the potential for personal injury is detected and words like "kill," "help," and "call the police" appear multiple times, the system assigns a confidence value of yrz = 90 and an urgency value of yrj = 95. This means the confidence level of personal injury is 90%, indicating a high probability of injury, and the urgency is 95%, indicating an extremely serious situation.

[0172] (3) Motion inertia information processing unit:

[0173] The motion inertia information processing unit described in this embodiment is used to process and identify motion inertia information to detect whether a dangerous situation has occurred and assign confidence and criticality values ​​to the dangerous situation. It combines sensor data feature analysis with intelligent algorithm processing. The logical steps are as follows:

[0174] 1) Preprocessing optimization:

[0175] Noise reduction: Kalman filtering eliminates sensor drift and is combined with a sliding window to segment short-term behavior data.

[0176] Coordinate calibration: The magnetometer helps establish a global coordinate system to reduce the impact of placement differences.

[0177] Time domain / frequency domain fusion: Extract acceleration mean, variance, and zero-crossing rate, combined with wavelet packet energy analysis.

[0178] 2) Motion feature extraction and algorithm recognition of dangerous scenes:

[0179] Typical examples include:

[0180] ① Faint / Fall Detection:

[0181] Feature extraction: A sudden drop in vertical acceleration (such as a free fall effect > 3g) combined with horizontal acceleration approaching zero, supplemented by a sudden change in attitude angle (such as an angle with the horizontal < 25° for 3 seconds).

[0182] Algorithm optimization: adopt threshold judgment and decision tree fusion model, combined with environmental data calibration to improve accuracy.

[0183] ②Epileptic seizure identification:

[0184] Feature extraction: 5-10 Hz high-frequency periodic jitter in the acceleration signal, combined with the gyroscope, can capture limb symmetry abnormalities.

[0185] Algorithm logic: FFT frequency domain analysis is used to extract the main frequency component, which is then compared with the pre-stored epilepsy waveform library. LSTM is introduced to capture the temporal characteristics of pre-ictal signs.

[0186] ③Foreign body asphyxiation detection:

[0187] Movement characteristics: severe shaking of the neck (acceleration > 4g) and abnormal periodic chest and abdominal movements caused by respiratory muscle spasms.

[0188] Algorithm strategy: Analyze the acceleration spectrum characteristics and use a bidirectional LSTM model to capture the temporal characteristics of pre-asphyxia signs.

[0189] ④Electric shock identification:

[0190] Movement characteristics: high-frequency tremor (20-50Hz) caused by tonic muscle contraction, sudden limb rigidity (sudden drop in angular velocity).

[0191] Algorithm optimization: Wavelet packet decomposition is used to extract myoelectric interference features, and cross-validation is performed with environmental electric field sensor data.

[0192] ⑤Brute force attack detection:

[0193] Feature enhancement: multi-axis impact peak (>8g), limb collision frequency (3-5 times / second), and integration of environmental sound wave characteristics.

[0194] Model optimization: Improve the dynamic time warping (DTW) algorithm and introduce an attention mechanism to improve template matching accuracy.

[0195] 3) Assign a value to the risk:

[0196] The motion inertia information processing unit assigns values ​​to identified dangerous situations, specifically the confidence and urgency of a particular hazard. The values ​​range from 0 to 100. The system assigns these values ​​based on the density and intensity of the hazard signatures identified during the hazard identification process, according to pre-set rules. For example, if a possible epilepsy risk is identified, the system assigns a confidence value of gdz = 70 and an urgency value of gdj = 85. This means the confidence level is 70%, indicating a high probability of an epilepsy, and the urgency is 85%, indicating a high probability of an epilepsy.

[0197] (4) Physiological indicator information processing unit:

[0198] The physiological indicator information processing unit described in this embodiment is used to process physiological indicator information to find out whether a dangerous situation occurs and assign a confidence level and a criticality level to the dangerous situation.

[0199] 1) Data processing technology process:

[0200] ①Signal preprocessing:

[0201] Noise reduction: Kalman filtering removes signal interference.

[0202] Standardization: Z-score normalization handles the differences in sensor dimensions.

[0203] ② Feature Engineering Strategy:

[0204] Time domain features: heart rate variability (HRV), standard deviation of breathing interval.

[0205] Frequency domain features: FFT extracts the main frequency component of the ECG signal, and wavelet packets decompose the blood oxygen signal.

[0206] ③Hierarchical decision-making:

[0207] Primary screening: threshold judgment (such as heart rate continuously >120bpm for more than 30 seconds).

[0208] Secondary validation: SVM classifier analysis of multi-parameter correlation (such as heart rate-blood oxygen coupling abnormalities).

[0209] 2) Physiological feature extraction and algorithm recognition of dangerous scenarios:

[0210] ①Electric shock detection:

[0211] Feature extraction:

[0212] Muscle rigidity and tremor trigger high-frequency heart rate fluctuations (20-50 Hz);

[0213] A sudden drop in skin resistance (when current passes through it) combined with an abnormal breathing rate.

[0214] Algorithm strategy:

[0215] Wavelet packet decomposition is used to extract myoelectric interference features, which are then fused with environmental electric field sensor data for cross-validation.

[0216] Real-time monitoring of skin resistance change rate (>50% mutation trigger threshold)

[0217] ②Myocardial infarction detection:

[0218] Feature extraction:

[0219] ECG ST segment deviation (a sign of myocardial ischemia) + sudden drop in blood pressure (systolic blood pressure <90 mmHg)

[0220] Decreased blood oxygen saturation (<90%) accompanied by cold sweats

[0221] Algorithmic strategy:

[0222] 1D-CNN model: processes ECG signal waveform features and identifies ST segment morphological abnormalities;

[0223] Graph Neural Network: Integrates ECG, blood pressure, and blood oxygen data to improve prediction specificity.

[0224] ③Epileptic seizure identification:

[0225] Feature extraction:

[0226] A rapid heart rate (>150 bpm) accompanied by respiratory disturbances (irregular, shallow, rapid breathing, periodic apnea);

[0227] A sudden drop in blood oxygen saturation (due to muscle stiffness affecting breathing).

[0228] Algorithm optimization:

[0229] LSTM time series model: captures the gradual change pattern of physiological signals 5 minutes before the onset;

[0230] Wavelet packet decomposition: Extracts the 5-10 Hz high-frequency component from the EEG signal and fuses it with the jitter characteristics of the inertial sensor.

[0231] ④Violence / robbery identification:

[0232] Feature extraction:

[0233] Adrenaline surge causing heart rate >120 bpm

[0234] A surge in skin conductivity (due to adrenaline);

[0235] Respiratory rate > 30 breaths / minute + rapid drop in blood oxygen saturation (traumatic asphyxia)

[0236] Multi-source verification, combined with inertial sensors to detect violent limb movements (acceleration > 8g)

[0237] Model Identification:

[0238] Support vector machine (SVM), classification of normal and stress states;

[0239] Dynamic Time Warping (DTW) matches preset physiological response templates of violent scenes.

[0240] ⑤ Fainting / loss of consciousness detection:

[0241] Core indicators: sudden drop in heart rate (<40 beats / min), sharp drop in blood pressure (systolic blood pressure <70 mmHg), and abnormal respiratory rate (such as apnea or hyperventilation).

[0242] Algorithm logic:

[0243] Through the dynamic threshold model, the joint mutation of heart rate and blood pressure (such as a 40% drop in heart rate within 8 seconds) is monitored in real time.

[0244] Multi-sensor fusion, combined with acceleration sensors, determines sudden changes in body position (such as a sudden drop in vertical acceleration >3g).

[0245] ⑥Drowning detection:

[0246] Core indicators:

[0247] A sudden drop in respiratory rate (<8 breaths / minute);

[0248] A sharp drop in blood oxygen saturation (<85%);

[0249] Loss of body temperature (low temperature in water resulting in <35°C).

[0250] Algorithm optimization:

[0251] Through the decision tree model, the coordinated abnormalities of respiration, blood oxygen and body temperature are jointly judged.

[0252] 3) Assign a value to the risk:

[0253] The physiological indicator information processing unit assigns values ​​to identified dangerous situations, specifically the confidence and criticality of a particular danger, ranging from 0 to 100. The system assigns these values ​​based on the density and intensity of the dangerous features identified during the above risk identification, according to pre-set rules. For example, if a possible "heart attack" is identified, the system assigns a confidence value of lxz = 75 and a criticality value of lxj = 95. This means the confidence level of a heart attack is 75%, indicating a high probability, and the criticality is 95%, indicating a very critical situation.

[0254] (5) Comprehensive information processing unit:

[0255] The comprehensive information processing unit, based on the aforementioned visual information processing, audio information processing, physiological indicator information processing and other information processing, further processes the information and data, and assigns comprehensive values ​​to the confidence and urgency of the dangerous situation, so as to form a judgment on the comprehensive confidence and comprehensive urgency of the dangerous situation, and provide a basis for whether to initiate emergency measures and which appropriate emergency measures to initiate.

[0256] The following describes the logical process by taking the integrated information processing unit that integrates four types of perception information, including visual perception, audio perception, motion inertia perception, and physiological indicator perception, as an example.

[0257] For a certain danger, such as "personal injury," as mentioned above, after information perception and information processing, the visual information processing unit, audio information processing unit, motion inertia information processing unit, and physiological indicator information processing unit respectively assign values ​​to the confidence level and criticality level of the dangerous situation of "personal injury." Assume that the values ​​are as follows:

[0258] The visual information processing unit assigns the confidence and criticality values ​​of "personal injury" as follows: srz = 90, srj = 85;

[0259] The audio information processing unit assigns the confidence and criticality values ​​of "personal injury" as follows: yrz = 80, yrj = 90;

[0260] The motion inertia information processing unit assigns the confidence and criticality values ​​of "personal injury" as: grz = 75, grj = 80;

[0261] The physiological index information processing unit assigns the confidence and criticality of "personal injury" as: Lrz = 85, Lrj = 85;

[0262] The comprehensive values ​​of confidence and criticality for “personal injury” are:

[0263] Overall confidence level for “Personal injury”

[0264] rz=30%*srz+25%*yrz+15%*grz+30%*Lrz

[0265] =30%*90+25%*80+15%*75+30%*85

[0266] =83.75

[0267] The 30%, 25%, 15%, and 30% in the formula are the weights of various information processing units.

[0268] The overall severity of “personal injury”

[0269] rj=25%*srj+20%*yrj+20%*grj+35%*Lrj

[0270] =25%*85+20%*90+20%*80+35%*85

[0271] =85

[0272] The result of comprehensive information processing shows that the comprehensive confidence level of "personal injury" is 83.75% and the comprehensive criticality level is 85%.

[0273] The weights of various information processing units are built into the system in advance based on objective conditions and remain relatively stable, but can be optimized through subsequent learning and verification.

[0274] The weight settings of different information processing units will be different during comprehensive calculations; for the same type of information processing units, the weight settings when calculating the comprehensive criticality may be different from the weight settings when calculating the comprehensive confidence; for different dangerous situations, the weight settings of the same type of information processing units may also be different during comprehensive calculations.

[0275] Furthermore, after weighting, the integration of multiple types of information and multi-party verification generally improves the accuracy of assigned judgments. However, there are exceptions. In some cases, if a small number of one or two types of information processing can have a high degree of confidence, weighting them with more information processing can sometimes reduce sensitivity. To address these rare exceptions, the system also sets an exception principle when setting weights. That is, under normal circumstances, the weights are set according to the system's pre-set plan, but if several special circumstances arise, the weights will be adjusted. For example, if the user's physiological indicators are found to be seriously abnormal during physiological indicator information processing, the weight of the physiological indicator information processing unit will be temporarily set to 100% when calculating the comprehensive criticality level; if the user's verified voice contains the voice of "Call 110 or help" during audio information processing, the weight of the audio information processing unit will be temporarily set to 95% when calculating the comprehensive confidence level.

[0276] The system currently senses, identifies, and calculates 12 dangerous situations: personal injury, kidnapping, robbery, animal attack, fire, drowning, electric shock, car accident, fainting, myocardial infarction, epilepsy, and asphyxiation. Four information processing units—the visual information processing unit, the audio information processing unit, the motion inertia information processing unit, and the physiological indicator information processing unit—assign confidence and criticality values ​​to each of the 12 dangerous situations. This simultaneously monitors 12*4*2=96 values. The comprehensive information processing unit then calculates the combined confidence and criticality values ​​for each of the 12 dangerous situations, generating 24 comprehensive values. If any of these 24 comprehensive values ​​reaches a preset threshold, the corresponding emergency procedure is initiated.

[0277] With deep learning and continuous improvement, the system will expand to more perception units, more information processing models, and more dangerous situations.

[0278] 4. Information storage and transmission module

[0279] A module for storing and transmitting various information perceived, interacted and generated. The storage module includes an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, other storage devices or any suitable combination of the above. The transmission module includes a wired transmission unit and / or a wireless transmission unit; the wired transmission unit includes but is not limited to: any one or more combinations of transmission units such as USB1.0 / 2.0 / 3.x, MicroUSB, MiniUSB, serial interface, parallel interface, etc.; the wireless transmission unit includes but is not limited to: any one or more combinations of 2G / 3G / 4G / 5G, Bluetooth, Star Flash, Infrared, Hotspot, NB-IoT, Rola, Zigbee, MavLink, WIFI, NFC, GPRS, GSM, and Ethernet.

[0280] 5. Protective measures execution module

[0281] Mainly includes two or more of the following units:

[0282] (1) Communication unit: A unit that informs relatives and friends of danger through telephone, text messages, WeChat, etc.

[0283] (2) Automatic alarm or emergency call unit. This unit automatically dials 110 or 120 and communicates through automatic voice to complete the 110 alarm or 120 emergency call.

[0284] (3) Voice broadcast unit: A unit that automatically warns bad guys at the scene of an incident or broadcasts a call for help to nearby people.

[0285] (4) Sound, light, and electrical units. Some equipment may also be equipped with sound, light, and electrical units, such as units that can emit a special loud sound in an emergency to attract attention or deter bad guys; units that can flash continuously in a dark environment to attract attention or guide rescuers; units that can use electric shocks for self-defense; units that can release special sprays, etc.

[0286] Once automatically activated, the equipment system will execute the corresponding safety protection emergency measures. There are many types of emergency measures. In order to increase the success rate of response, two or more safety protection emergency measures suitable for a certain dangerous situation are usually activated. The main types of emergency measures are as follows:

[0287] (1) Evidence collection, storage and transmission:

[0288] If a dangerous situation is detected, evidence collection, storage, and transmission procedures can be initiated. For example, visual, audio, and location sensing units can be activated to automatically capture images or videos of the relevant people and environment, record the corresponding audio, obtain the specific location, detect the user's physiological indicators and other data and information, and store and transmit this information to a predetermined medium or contact for subsequent use in alarm, rescue, accountability, and investigation. Sensitive and private information must be filtered, encrypted, or processed only after permission is obtained.

[0289] (2) Warning reminder:

[0290] If it is determined that a dangerous situation has occurred, but the confidence level is not very high or the situation is not urgent, the warning reminder program can be activated to inform the preset relatives and friends of the user's dangerous situation through the communication unit by phone, text message, WeChat reminder, etc., including the type of dangerous situation, the user's current specific location, some information about the surrounding environment, etc., so that relatives and friends can discover the dangerous situation in time, organize rescue or call the police to deal with the crisis.

[0291] (3) Warning:

[0292] If it is judged that a dangerous situation such as a personal attack has occurred, you can choose to start the warning program, that is, through the automatic voice broadcast unit and AI language, warn the bad guys, inform them that the evidence of their bad behavior has been recorded and transmitted, and persuade them to stop, which has a persuasive and deterrent effect; or through the sound and light electrical unit, emit a particularly loud sound or strong flash to attract the attention of the surrounding or scare off the bad guys; and so on.

[0293] (4) Alarm:

[0294] If it is determined that a dangerous situation such as a serious criminal incident or a serious traffic accident has occurred, you can choose to start the alarm program, automatically dial 110 or send a text message to 12110, and communicate through automatic voice or automatic text to inform the police of the specific location of the dangerous situation, the identity and physical characteristics of the parties involved, the telephone numbers and contact information of the parties and their relatives and friends, some important information and related evidence perceived by the sensing unit, etc., to complete the entire 110 alarm process.

[0295] (5) Emergency assistance

[0296] If it is determined that a dangerous situation such as acute illness, drowning, personal injury, etc. has occurred, you can choose to start the emergency assistance program, such as automatically dialing the 120 emergency number, and informing the specific condition, specific location and other information through automatic voice; or informing relatives, friends and emergency contacts of the location and general situation of the dangerous situation through communication methods such as telephone, WeChat or text messages, and asking for help; or through automatic voice broadcast, broadcast the situation of the dangerous situation to the surrounding area of ​​the scene, and seek help from surrounding people; or for specific dangerous situations, guide surrounding people to implement correct rescue measures in appropriate methods through interactive units or automatic voice.

[0297] For the various emergency measures mentioned above, the degree of automation of activation can be intelligently set based on the confidence level, the urgency of the situation, and the impact of the emergency measures. For example, if the confidence level of the judgment is very high and the situation is relatively urgent, the device system can fully and automatically activate the subsequent emergency procedures immediately. If the confidence level of the judgment is not particularly high, the situation is not very urgent, or the negative impact of the subsequent emergency measures to be activated is large, the device can ask the user's opinion before automatically activating. For example, when an emergency is detected, the device can first ask the user by voice or text whether to automatically call 120 emergency number immediately. If the user gives a clear opinion, the device can also identify it as the user's opinion through the voiceprint and follow the user's opinion. If the user does not give a negative opinion within the set time or the set number of inquiries, the subsequent emergency measures will still be automatically activated to prevent the user from being unable to express their opinions due to coma. This can relatively well take into account the possible inconsistency between the user's opinion and the automatic judgment activation.

[0298] Therefore, for each type of emergency measures, there are three specific emergency measures: immediate automatic start, automatic start after inquiry, and no action. For each type of emergency measures, according to the three specific situations, each specific emergency measure is given a code:

[0299] A1, immediately and automatically collect evidence, store and transmit; B1, collect evidence, store and transmit after inquiry; O1, no action.

[0300] A2: Immediately encrypt the evidence, store and transmit; B2: Encrypt the evidence, store and transmit after inquiry; O2: No action.

[0301] A3, immediate phone reminder; B3, phone reminder after inquiry; O3, no action.

[0302] A4, immediate SMS reminder; B4, SMS reminder after inquiry; O4, no action.

[0303] A5, immediate WeChat reminder; B3, WeChat reminder after inquiry; O5, no action.

[0304] A6, immediately give a voice warning to the bad guy; B6, give a voice warning to the bad guy after questioning; O6, do nothing.

[0305] A7, immediate sound and light warning; B7, sound and light warning after inquiry; O7, no action.

[0306] A8, call 120 emergency services immediately; B8, call 120 emergency services after inquiring; O8, do nothing.

[0307] A9, call 119 immediately; B9, call 119 after questioning; O9, do nothing.

[0308] A10, call 110 immediately; B10, call 110 after questioning; O10, do nothing.

[0309] A11, immediately send an SMS to 12110 to alert the police; B11, send an SMS to 12110 to alert the police after inquiry; O10, take no action.

[0310] A12, immediately seek emergency help from relatives and friends; B12, seek emergency help from relatives and friends after asking; O12, do nothing.

[0311] A13, immediately seek help from the surroundings; B13, seek help from the surroundings after asking; O13, do nothing. The above 13 types of emergency measures, each type has 3 specific options, after permutation and combination, theoretically there are

[0312] 3 13 =A combination of 1,594,323 specific measures.

[0313] 6. Control module

[0314] The control module is used to control and coordinate each module and unit to complete their respective functions according to a pre-set plan or in an automatic and intelligent manner.

[0315] 7. Other modules:

[0316] Including modules such as power supply, switches, structural parts, and decorative parts.

[0317] The following details the method and process of automatic emergency safety protection

[0318] The process includes steps such as setting, monitoring, automatic startup, and execution:

[0319] 1. Settings

[0320] Through the interactive module, the user inputs some important relevant information into the device in advance, and sets and selects the dangerous situations to be focused on and a series of emergency plans.

[0321] Important relevant information to be entered includes the user's name, gender, age, height, weight, any medical conditions, contraindications, mobile phone number, address and other basic personal information, information about relatives and friends, emergency contact information of relatives and friends, as well as the user's personal image, voice and other information to facilitate identity confirmation and authentication in emergency situations, etc.

[0322] Set up and select a series of emergency plans, including choosing which dangerous situations to guard against, pre-selecting a priority emergency plan from the series of plans built into the device itself for a certain type of dangerous situation, and setting parameter combinations of multiple element options in the emergency plan to suit the user's own situation or personal preferences. If the user does not set and select in advance, the device will automatically select the system default settings.

[0323] 2. Monitoring

[0324] The portable or wearable intelligent emergency device described in this embodiment continuously monitors the user's condition and the surrounding environment in standby mode, and detects whether a dangerous situation occurs through perception, analysis, and judgment.

[0325] The perception process primarily involves sensing and measuring the images, movements, sounds, physiological indicators, and other data about the user and their surroundings through the device's visual perception unit, audio perception unit, physiological indicator perception unit, position perception unit, motion inertia perception unit, or other perception units. Each perception unit then transmits the perceived data to the information processing unit for processing. The information processing unit uses models and algorithms to analyze the various data and information, forming a judgment to determine whether a security incident has occurred, such as "personal injury, kidnapping, robbery, animal attack, fire, drowning, electric shock, car accident, fainting, myocardial infarction, epilepsy, or suffocation by a foreign object."

[0326] For example, as mentioned above, the four information processing units—the visual information processing unit, the audio information processing unit, the motion inertia information processing unit, and the physiological indicator information processing unit—each assigns a value to the confidence and criticality of 12 dangerous situations. This means the system is simultaneously monitoring 4*12*2=96 values. The comprehensive information processing unit then calculates the combined confidence and criticality of these 12 dangerous situations, resulting in 24 comprehensive values. If any of these 24 comprehensive values ​​reaches a preset threshold, the corresponding emergency procedure combination is automatically activated.

[0327] 3. Automatic startup

[0328] Once the information processing unit determines with a high degree of confidence that a dangerous situation has occurred and the situation is critical, it will automatically issue instructions to the corresponding execution unit based on the degree of criticality to initiate the preset corresponding emergency procedures.

[0329] The response measures that will be automatically activated subsequently are divided into different dangerous situations, different levels of criticality, and different confidence levels. The system has pre-set default settings, and users can also revise them in the settings. Certain response measures under certain dangerous situations, certain levels of criticality, and certain confidence levels can be set to be fully automatically activated immediately; certain response measures under certain dangerous situations, certain levels of criticality, and certain confidence levels can be set to be automatically activated after inquiry.

[0330] For example, for 12 dangerous situations, the integrated information processing unit continuously calculates and monitors their comprehensive confidence and comprehensive criticality. The system pre-sets three thresholds of 85, 70, and 55 for the comprehensive confidence, dividing it into four situations:

[0331] Comprehensive confidence:

[0332] z>85, very high;

[0333] 85≥z>70, high;

[0334] 70≥z>55, higher;

[0335] z≤55, general.

[0336] The system has pre-set three thresholds for comprehensive criticality: 90, 75, and 60, which are also divided into four situations:

[0337] Comprehensive criticality level:

[0338] j>90, critical;

[0339] 90≥j>75, urgent;

[0340] 75≥j>60, nervous;

[0341] j≤60, general.

[0342] By monitoring the comprehensive confidence and comprehensive criticality of 12 dangerous situations, there will be 12*4*4=192 specific states.

[0343] For each state, a corresponding combination of emergency measures is also pre-set. The following is an example of an emergency measure combination:

[0344] Table 1: Example of emergency measures combination

[0345]

[0346]

[0347] The above table only lists the recommended emergency response combinations for the two dangerous situations of "personal injury" and "fainting" under different conditions. In actual application, some revisions may be made appropriately. The same can be applied to other types of dangerous situations.

[0348] When the system monitoring detects a specific state, the emergency measures combination corresponding to the state is immediately activated.

[0349] 4. Execution

[0350] After the system starts the emergency measures combination corresponding to a specific state, the corresponding emergency protection measures execution unit will automatically execute the series of emergency measures combinations.

[0351] For example, if the state of "personal injury" is 85≥rz>70 and 75≥rj>60, the system will activate the emergency response combination corresponding to this state: B2, A5, B6, B7, B8, B11, A12, B13. At this time, the overall confidence level of "personal injury" is "high", and its overall criticality is "tense". The emergency response combination is:

[0352] B2: Encrypted storage and transmission after inquiry;

[0353] A5: Immediately alert relatives and friends;

[0354] B6: After questioning, give a voice warning to the bad guys;

[0355] B7: Sound and light warning after inquiry;

[0356] B8: Call 120 emergency number after inquiry;

[0357] B11: After making inquiries, send an SMS to 12110 to alert the police;

[0358] A12: Immediately seek emergency help from relatives and friends;

[0359] B13: After asking for help, he sought help from the surrounding area.

[0360] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0361] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. An automatic emergency safety protection system, characterized in that: include: Interaction module, used for users to interact with the system; Perception module, including visual perception unit, audio perception unit, motion inertia perception unit, physiological indicator perception unit and position perception unit; An information processing module is used to process the information sensed by each sensing unit to determine whether a dangerous situation has occurred and to determine its confidence level and criticality. The information processing module includes: Visual information processing unit, audio information processing unit, physiological index information processing unit, motion inertia information processing unit and comprehensive information processing unit; The information storage and transmission module is used to store and transmit various types of information perceived, interacted and generated, including a storage unit and a transmission unit; The protective measures execution module includes two or more of a communication unit, an automatic alarm or automatic emergency call unit, a voice broadcast unit, and an audio-visual and electrical unit.

2. The automatic emergency safety protection system according to claim 1, characterized in that: The visual information processing unit combines visual perception with image recognition and deep learning technology to process and identify visual information to detect whether a dangerous situation has occurred and assign confidence and criticality values ​​to the dangerous situation. The specific steps are as follows: Visual data preprocessing: preprocessing the original images and video streams obtained by the visual perception unit, including image noise reduction, dynamic range adjustment, ROI cropping, using deblurring algorithms to deal with motion blur, and improving image quality in low-light environments through histogram equalization; Object detection, feature extraction, and scene recognition, including: Dangerous goods identification, based on the improved model of YOLOv8, performs real-time target detection of dangerous goods; Human posture and behavior analysis: using OpenPose and MediaPipe algorithms to extract the coordinates of 135 key points on the human body, construct skeletal motion trajectories, analyze human posture and facial expressions, and identify abnormal movements; Time series analysis and recognition, combined with LSTM / 3D-CNN models, analyzes action sequence features in consecutive frames and detects falls based on posture angles; Dangerous scene identification, including: Scene semantic segmentation, using U-Net and DeepLab models to segment flames, smoke, water ripples, or facial expressions; Drowning detection uses a water surface segmentation algorithm to delineate dangerous areas and an optical flow method to analyze abnormal trajectories of struggling limbs. Fire identification uses the YOLO-Fire model to detect flames and smoke in real time, and thermal imaging data to assist in identifying high-temperature areas; Violent conflict detection uses a posture estimation model to detect changes in head and torso angles and analyzes facial expressions and acute illnesses through changes in facial microvascular color. To assign values ​​to dangers, the visual information processing unit will assign values ​​to the identified dangerous situations, that is, the confidence and criticality of a certain danger will be assigned respectively. The assignment range is 0-100. According to the density and intensity of the dangerous features in the identification of the above dangers, the confidence and criticality will be assigned respectively according to the preset rules.

3. The automatic emergency safety protection system according to claim 1, characterized in that: The audio information processing unit combines automatic speech recognition technology, deep learning technology, and natural language processing technology to process and identify audio information to detect whether a dangerous situation has occurred and assign confidence and criticality values ​​to the dangerous situation, as follows: Speech signal processing and text conversion: First, the speech data signal obtained by the audio perception unit is pre-processed. Through noise reduction, echo cancellation, speech enhancement, framing and Fourier transform, the signal is converted into a multidimensional vector and the Mel-frequency cepstral coefficient features are extracted; Then, the factors and text are decoded, and the features are mapped to phonemes through the deep neural network acoustic model. The factors are then combined with the language model to form words and sentences, and then the text is output. Hazard identification includes: Acoustic feature extraction, anomaly detection, and sentiment analysis: MFCC algorithms are used to extract the spectral characteristics of sound. The audio information processing unit can identify acoustic features that exceed the normal range. Natural Language Processing (NLP) models are used to analyze the emotions and feelings in information, and abnormal fluctuations in acoustic features are also used to determine emotional states. Keyword extraction, preset sensitive word library, and rapid screening of high-risk content through regular expressions or string matching; The system establishes a multi-level keyword library, including primary keywords and secondary keywords; The system uses dynamic time warping algorithm to achieve fuzzy matching and supports dialect and accent recognition; Contextual association detection, combined with the conversation process, identifies contradictory statements and threatening logic, and builds a context model through the LSTM neural network to identify typical conversation patterns in dangerous scenarios; Model recognition and machine learning: training classification models based on historical dangerous conversation data to identify potential threat patterns; Use online learning framework to continuously optimize the model; Assign values ​​to dangers. The audio information processing unit will assign values ​​to the identified dangerous situations, that is, assign values ​​to the confidence and urgency of a certain danger. The assignment range is 0-100. The system will assign values ​​to the confidence and urgency according to the density and intensity of the dangerous features in the identification of the above dangers and according to preset rules.

4. The automatic emergency safety protection system according to claim 1, characterized in that: The motion inertia information processing unit is used to process and identify motion inertia information to detect whether a dangerous situation has occurred and assign confidence and criticality values ​​to the dangerous situation. It combines sensor data feature analysis with intelligent algorithm processing. The logical steps are as follows: Preprocessing optimization, including: Noise reduction: Kalman filtering eliminates sensor drift and is combined with a sliding window to segment short-term behavior data; Coordinate calibration: The magnetometer helps establish a global coordinate system to reduce the impact of placement differences; Time domain / frequency domain fusion: Extract acceleration mean, variance, and zero-crossing rate, combined with wavelet packet energy analysis; Motion feature extraction and algorithm recognition of dangerous scenes, including: Faint / Fall Detection: Feature extraction: vertical acceleration drops sharply combined with horizontal acceleration approaching zero, supplemented by attitude angle mutation; Algorithm optimization: Using threshold judgment and decision tree fusion model, combined with environmental data calibration, to improve accuracy; Seizure Identification: Feature extraction: 5-10 Hz high-frequency periodic jitter in the acceleration signal is combined with the gyroscope to capture limb symmetry abnormalities; Algorithm logic: FFT frequency domain analysis extracts the main frequency component, compares it with the pre-stored epilepsy waveform library, and introduces LSTM to capture the temporal characteristics of pre-ictal auras; Foreign body asphyxiation detection: Movement characteristics: severe neck shaking, abnormal periodic chest and abdominal movements caused by respiratory muscle spasms; Algorithm strategy: Analyze acceleration spectrum characteristics and use a bidirectional LSTM model to capture the temporal characteristics of presymptomatic asphyxia. Electric shock identification: Motor characteristics: high-frequency tremors and sudden limb stiffness caused by tonic muscle contraction; Algorithm optimization: Wavelet packet decomposition is used to extract myoelectric interference features, and cross-validation is performed with environmental electric field sensor data; Brute force attack detection: Feature enhancement: multi-axis impact peak, limb collision frequency, and fusion of environmental sound wave characteristics; Model optimization: Improve the dynamic time warping algorithm and introduce an attention mechanism to improve template matching accuracy; Assigning values ​​to dangers: The motion inertia information processing unit will assign values ​​to identified dangerous situations, that is, the confidence and urgency of a certain danger will be assigned values ​​respectively. The assigned values ​​range from 0 to 100. The system will assign values ​​to the confidence and urgency according to the density and intensity of the dangerous features in the identification of the above dangers and the preset rules.

5. The automatic emergency safety protection system according to claim 1, characterized in that: The physiological indicator information processing unit is used to process physiological indicator information to detect whether a dangerous situation occurs and assign a confidence level and a criticality level to the dangerous situation. The data processing process is as follows: Signal preprocessing: Kalman filtering is used to eliminate signal interference; Z-score normalization is used to process the dimensional differences of different sensors; Identify and process time domain features and frequency domain features; Hierarchical decision-making, including: Primary screening: threshold judgment; Secondary validation: SVM classifier analysis of multi-parameter correlation; Physiological feature extraction and algorithm recognition of dangerous scenarios, including: Electric shock detection: Feature extraction: muscle rigidity and tremor trigger high-frequency heart rate fluctuations, sudden drop in skin resistance combined with abnormal breathing rate; Algorithm strategy: Wavelet packet decomposition is used to extract myoelectric interference features, which are then integrated with environmental electric field sensor data for cross-validation to monitor skin resistance changes in real time. Myocardial infarction detection: Feature extraction: ECG ST segment deviation + sudden drop in blood pressure, decreased blood oxygen saturation accompanied by cold sweat physiological indicators; Algorithm strategy: 1D-CNN model: processes ECG signal waveform features and identifies ST segment morphological abnormalities; graph neural network: integrates ECG, blood pressure, and blood oxygen data to improve prediction specificity; Seizure Identification: Feature extraction: A surge in heart rate accompanied by respiratory disturbances and a sudden drop in blood oxygen saturation; Algorithm optimization: LSTM time series model captures the gradual change pattern of physiological signals in the 5 minutes before the onset of the attack; wavelet packet decomposition: extracts the 5-10Hz high-frequency components in the EEG signal and integrates them with the jitter characteristics of the inertial sensor; Violence / robbery identification: Feature extraction: Adrenaline surges cause heart rates > 120 bpm; skin conductivity surges; respiratory rate > 30 breaths / minute + rapid drop in blood oxygen saturation; multi-source verification, combined with inertial sensors to detect intense limb movements; Model identification: Support vector machine, classifying normal and stress states; dynamic time warping, matching preset physiological response templates of violent scenes; Faint / Loss of Consciousness Detection: Core indicators: sudden drop in heart rate, sharp drop in blood pressure, abnormal respiratory rate; Algorithm logic: Real-time monitoring of combined sudden changes in heart rate and blood pressure through a dynamic threshold model; multi-sensor fusion combined with acceleration sensors to determine sudden changes in body position; Drowning detection: Core indicators: sudden drop in respiratory rate, sharp drop in blood oxygen saturation, and loss of body temperature; Algorithm optimization: Through the decision tree model, jointly judge the coordinated abnormalities of respiration, blood oxygen, and body temperature; Assign a value to the danger: The physiological indicator information processing unit will assign values ​​to the identified dangerous situations, that is, the confidence and criticality of a certain danger will be assigned respectively, and the assigned values ​​range from 0 to 100. The system will assign values ​​to the confidence and criticality respectively according to the density and intensity of the dangerous characteristics in the identification of the above dangers and the preset rules.

6. The automatic emergency safety protection system according to claim 1, characterized in that: The comprehensive information processing unit is used for comprehensive information processing, that is, on the basis of the visual information processing, audio information processing, physiological indicator information processing and other information processing, the information and data are comprehensively processed to give a comprehensive value to the overall confidence and urgency of the dangerous situation, so as to form a judgment on the comprehensive confidence and comprehensive urgency of the dangerous situation, and provide a basis for whether to initiate emergency measures and which emergency measures to initiate subsequently.

7. An automatic emergency safety protection method for an automatic emergency safety protection system according to any one of claims 1 to 6, characterized in that: include: setup, monitoring, automatic startup and execution; The settings include: The user inputs some important relevant information into the system in advance through the interactive unit, and sets and selects the dangerous situations to be focused on and a series of emergency plans; Important relevant information to be entered includes the user's name, gender, age, height, weight, any medical conditions, contraindications, mobile phone number, address, information of relatives and friends, emergency contact information of relatives and friends, as well as the user's personal image, voice and other information to facilitate identity confirmation and authentication in emergency situations; Setting and selecting a series of emergency plans, including selecting which dangerous situations to guard against, pre-selecting a priority emergency plan from the system's built-in series of plans for certain dangerous situations, and setting parameter combinations of multiple element options in the emergency plan to suit the user's own situation or personal preferences. If the user does not set and select in advance, the system will automatically select the system default settings; The monitoring includes: The automatic emergency safety protection system continuously monitors the user's condition and the surrounding environment in standby mode, and detects whether there is a dangerous situation through perception, analysis and judgment; The perception process is to perceive and measure the images, dynamics, sounds and physiological indicators of the user and the surrounding environment through the system's visual perception unit, audio perception unit, physiological indicator perception unit, position perception unit and motion inertia perception unit; each perception unit then transmits the perceived data information to the information processing unit for processing. The information processing unit uses models and algorithms to analyze various data information and form a judgment, thereby discovering whether a security incident has occurred, such as personal injury, kidnapping, robbery, animal attack, fire, drowning, electric shock, car accident, fainting, myocardial infarction, epilepsy, foreign body suffocation and other dangerous situations; The four information processing units, namely the visual information processing unit, the audio information processing unit, the motion inertia information processing unit, and the physiological index information processing unit, respectively assign values ​​to the confidence and criticality of 12 dangerous situations. In this way, the system is simultaneously monitoring 4*12*2=96 values. The comprehensive information processing unit then calculates the comprehensive confidence and comprehensive criticality of the 12 dangerous situations, thus obtaining 24 comprehensive values. If any of these 24 comprehensive values ​​reaches a preset threshold, the corresponding emergency procedure will be automatically initiated. The automatic startup includes: Once the information processing unit determines with a high degree of confidence that a dangerous situation has occurred and the situation is critical, it will automatically issue instructions to the corresponding execution unit according to the degree of criticality to activate the preset corresponding emergency procedures; The subsequent response measures that are automatically activated are divided into different dangerous situations, different criticality levels, and different confidence levels. The system has pre-set default settings, which can also be modified by the user in the settings. For certain dangerous situations, certain criticality levels, and certain confidence levels, certain response measures can be set to be automatically activated immediately; for certain dangerous situations, certain criticality levels, and certain confidence levels, certain response measures can be set to be automatically activated after inquiry. The execution includes: After the system starts the emergency measures combination corresponding to a specific state, the corresponding emergency protection measures execution unit will automatically execute the series of emergency measures combinations.

Images