Human body safety monitoring and early warning glasses based on AI analysis

Abstract

The invention relates to the technical field of safety monitoring, in particular to a pair of human body safety monitoring and early warning glasses based on AI analysis, which comprises a signal acquisition module for acquiring three-axis acceleration values and three-axis angular velocity values of the glasses, recording instantaneous readings of acceleration and angular velocity in a segmented manner and aligning according to a time sequence, recording a sound sampling sequence, marking a sampling moment, and generating multi-source original information. According to the invention, the real-time performance and accuracy of safety monitoring are improved by integrating multi-source information processing. Specifically, three-axis acceleration numerical values and angular velocity numerical values of the glasses are collected, a sound sampling sequence is recorded, data are recorded in a segmented mode, time sequence alignment is carried out, and therefore capturing of the impact state and positioning of the sound source position are more accurate. Through combination of motion change element screening and acoustic identification, the early warning capability of potential dangerous events in a complex environment is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of safety monitoring technology, and in particular to a pair of human safety monitoring and early warning glasses based on AI analysis. Background Art

[0002] Human safety monitoring and early warning glasses are a wearable device integrated with artificial intelligence technology, designed to monitor the wearer's physiological and environmental status in real time to prevent potential health and safety risks.

[0003] Existing technologies rely on basic sensor data and lack the integration and in-depth analysis of complex data types, which limits their dynamic recognition and immediate response capabilities. Conventional safety monitoring equipment struggles to accurately identify rapidly occurring physical impacts or complex sound patterns, potentially rendering early warnings ineffective at critical moments. This technical limitation is particularly pronounced in high-risk environments, such as industrial sites. Traditional technologies' failure to promptly identify and address falls or collisions can lead to safety incidents and compromise the wearer's health and safety. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a human safety monitoring and early warning glasses based on AI analysis.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A pair of human safety monitoring and early warning glasses based on AI analysis includes:

[0006] The signal acquisition module collects the three-axis acceleration and three-axis angular velocity values of the glasses, records the instantaneous acceleration and angular velocity readings in segments and aligns them in time series, records the sound sampling sequence and marks the sampling time, and generates multi-source original information;

[0007] The inertial analysis module, based on the multi-source raw information, screens the three-axis acceleration values and three-axis angular velocity values, compares the instantaneous rate of change, captures the impact state, and combines the lens offset amplitude to mark the corresponding time period as a suspected fall interval. It summarizes the peak value differences and summarizes the impact moment to generate motion change factors;

[0008] An acoustic recognition module, based on the motion change elements, retrieves the sound sampling sequence and extracts the sound pressure amplitude changes and high-frequency energy proportions. It then analyzes the sound pressure waveform in blocks and detects peak segments. It then correlates the frame vibration characteristics with the acoustic excursion rate, compares the peak intervals, locates the suspected sound source, and generates a sudden sound signature.

[0009] The AI judgment module matches the corresponding time periods of the falling impact and the shattering sound based on the motion change elements and the sudden sound markers and compares the durations. It lists each falling interval and the sound marker interval item by item, and performs correlation comparison on the angular velocity peak interval and the high-frequency sound pressure segment. At the same time, it combines infrared to identify whether there are obstacles ahead, judge whether there is a dangerous event, and generate a sudden state instruction.

[0010] Preferably, the system further comprises: an early warning interaction module, which triggers the sound and light warning device of the glasses based on the sudden change state instruction and presents a reminder icon and voice prompt on the lens display interface, verifies whether the user needs external help, and generates a warning interaction message;

[0011] The linkage output module connects to the external communication channel and sends a distress message or a cancellation command based on the warning interaction information. After verifying that the communication connection is successful, it retrieves the remaining power in the frame and sends a battery life warning message to the server to generate remote linkage data.

[0012] Preferably, the signal acquisition module includes:

[0013] The data acquisition submodule records the acceleration and angular velocity readings in segments based on the three-axis acceleration and angular velocity values of the glasses, divides the readings into blocks according to the reading time, aggregates the continuous measurement sequences and marks the indexes to generate the initial segmented records;

[0014] A sequence marking submodule, based on the initial segmented records, performs time axis index marking and maps acceleration readings and angular velocity readings to each index, arranges the order of different sampling moments and maps position reference points to generate a calibration alignment sequence;

[0015] The information integration submodule, based on the calibration alignment sequence, corresponds the acceleration readings and angular velocity readings with the sound sampling sequence and summarizes them into a unified matrix, adds the lens position information parameters and combines all discrete data to generate multi-source original information.

[0016] Preferably, the inertia analysis module includes:

[0017] A threshold comparison submodule extracts acceleration values and angular velocity values based on the multi-source original information, compares the rate of change point by point, marks the positions above or below the target range, and generates a preliminary threshold segment;

[0018] The fall capture submodule identifies continuous segments of acceleration values based on the preliminary threshold segment and marks the start and end positions, tracks the drop phase and locks the adjacent rise phase to generate a fall period set;

[0019] The impact summarization submodule verifies the angular velocity value based on the falling period set and captures the impact peak value with reference to the lens offset amplitude, classifies the peak value records into a unified index and sorts out the time differences to generate motion change elements.

[0020] Preferably, the acoustic recognition module includes:

[0021] The amplitude extraction submodule sequentially scans the sound sampling sequence and detects the change in sound pressure amplitude based on the motion change factor, locates the continuous segments with obvious sudden increases and marks the peak moments, summarizes the results and outputs an index group to generate an initial amplitude mark;

[0022] The frequency domain calculation submodule, based on the initial amplitude mark, splits the sound pressure waveform into multiple blocks and calculates the high-frequency energy ratio, identifies the peak range of prominent energy and numbers it, summarizes the energy distribution parameters corresponding to each block, and generates a high-frequency energy distribution;

[0023] The position positioning submodule defines the area based on the high-frequency energy distribution, combined with the vibration characteristics of the frame and the acoustic offset rate, compares the peak range and marks the corresponding sound source position, compiles the recognition results and stores them in a tag list, and generates a sudden sound tag.

[0024] Preferably, the AI determination module includes:

[0025] The interval matching submodule lists the start and end times of the fall impact segment and the suspicious sound source segment based on the motion change elements and the sudden sound markers, compares the overlapping parts, summarizes the common intervals, and generates a segment corresponding set;

[0026] The peak comparison submodule compares the angular velocity peak moments with the high-frequency sound pressure peak moments one by one based on the segment corresponding set, integrates adjacent time indexes into parallel information and saves them in a comparison list to generate multi-peak correlation information;

[0027] The danger judgment submodule checks whether the impact and high-frequency energy appear synchronously in the same section based on the multi-peak correlation information, marks the suspected dangerous moment and records the correlation mark, and generates a sudden change state instruction.

[0028] Preferably, the early warning interaction module includes:

[0029] The warning trigger submodule turns on the sound and light warning device based on the sudden change state instruction and records the start time, saves and archives the triggering conditions in sequence, and generates a warning trigger record;

[0030] An interface prompt submodule, based on the warning trigger record, displays prompt icons and voice guidance on the lens display interface, arranges interactive options into a list and waits for user operation, and generates an interface interaction map;

[0031] The user response submodule collects user feedback based on the interface interaction mapping and indicates whether to request assistance, stores the selected action together with the record, and generates warning interaction information.

[0032] Preferably, the linkage output module includes:

[0033] The communication connection submodule connects to the external communication channel and monitors the establishment process based on the warning interaction information, merges the connection status records after successful verification, and generates a connection confirmation result;

[0034] The message sending submodule transmits a distress message or cancels the command based on the connection confirmation result and marks the sending time, writes the corresponding data into the monitoring platform directory, and generates an external message summary;

[0035] The power retrieval submodule checks the remaining power in the frame based on the external message summary and outputs a battery life warning message to the server, records the current status of the monitoring terminal device and puts it into the maintenance list, and generates remote linkage data.

[0036] Compared with the prior art, the advantages and positive effects of the present invention are:

[0037] In this invention, the real-time and accuracy of safety monitoring are improved through the integration of multi-source information processing. Specifically, the three-axis acceleration and angular velocity values of the glasses are collected and the sound sampling sequence is recorded. The data is recorded in segments and the time series is aligned, making it possible to more accurately capture the impact state and locate the sound source. By combining the screening of motion change elements with acoustic recognition, the early warning capability for potential dangerous events in complex environments is enhanced, especially the ability to identify sudden sounds and physical impacts. This attention to detail makes the early warning system superior in real-time monitoring and accurate warning, and improves the comprehensiveness and response speed of safety monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is the glasses flow chart of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] See also Figure 1 The present invention provides a technical solution: a pair of human safety monitoring and early warning glasses based on AI analysis, comprising:

[0041] The signal acquisition module collects the three-axis acceleration and three-axis angular velocity values of the glasses, records the instantaneous acceleration and angular velocity readings in segments and aligns them in time series, records the sound sampling sequence and marks the sampling time, and generates multi-source original information;

[0042] The inertial analysis module, based on multi-source raw information, screens triaxial acceleration and triaxial angular velocity values, compares instantaneous rates of change, captures the impact state, and, combined with the lens offset amplitude, labels the corresponding time period as a suspected fall interval. It then summarizes peak differences and summarizes the impact moment to generate motion change factors.

[0043] The acoustic recognition module, based on motion change factors, retrieves the sound sampling sequence and extracts the sound pressure amplitude changes and high-frequency energy proportion. It analyzes the sound pressure waveform in blocks and detects the peak segments. It correlates the vibration characteristics of the frame with the acoustic excursion rate, compares the peak intervals, locates the suspected sound source, and generates a sudden sound marker.

[0044] The AI judgment module, based on motion change factors and sudden sound markers, matches the corresponding time periods of falling impact and cracking sounds and compares their durations. It then lists each falling interval and the sound marker interval one by one, and compares the angular velocity peak interval and high-frequency sound pressure segment. It also uses infrared to identify whether there are obstacles ahead, determines whether there is a dangerous event, and generates a sudden state instruction.

[0045] The early warning interaction module triggers the glasses' sound and light warning device based on the sudden change state instruction and displays a reminder icon and voice prompt on the lens display interface to verify whether the user needs external assistance and generate warning interaction information;

[0046] The linkage output module connects to the external communication channel and sends a distress message or cancels the command based on the warning interaction information. After verifying that the communication connection is successful, it retrieves the remaining power in the frame and sends a battery life warning message to the server to generate remote linkage data.

[0047] The signal acquisition module includes:

[0048] The data acquisition submodule records the acceleration and angular velocity readings in segments based on the three-axis acceleration and angular velocity values of the glasses, divides the readings into blocks according to the reading time, aggregates the continuous measurement sequences and marks the indexes to generate the initial segmented records;

[0049] The sequence marking submodule, based on the initial segmented records, performs time axis index marking and maps the acceleration and angular velocity readings to each index, arranges the order of different sampling moments and maps the position reference points to generate a calibration alignment sequence;

[0050] The information integration submodule, based on the calibration alignment sequence, corresponds the acceleration readings and angular velocity readings with the sound sampling sequence and summarizes them into a unified matrix, adds the lens position information parameters and combines all discrete data to generate multi-source raw information.

[0051] Specifically, based on the acquired three-axis acceleration and three-axis angular velocity data, the values of each data point are numbered and sorted in chronological order. First, combined with the results of multiple wearing tests, the acceleration is determined to be 0m / s 2 Up to 30m / s 2 The angular velocity range is between 0 rad / s and 20 rad / s as a preliminary reference range. This range is set based on the statistical results of more than 50 daily activities and the benchmark value is obtained by calculating the average peak value and standard deviation. Then δ = 1 m / s is selected. 2 and δ = 1 rad / s as the adjustable step size. If the peak acceleration exceeds 25 m / s for a long time in some tests, 2 Or if the angular velocity peak exceeds 18 rad / s, the corresponding threshold is gradually increased and the actual data is compared again to ensure that each record is arranged in order according to the collection time. During this process, the sampling frequency of 200 times per second is set to a fixed rate to capture subtle changes. The collected values are summarized into a continuous measurement sequence and indexed. Whenever a value is detected to be out of the reference range, the mark information is recorded for distinguishing extreme cases in subsequent analysis. After accumulating a sufficient amount of data, all records are divided into blocks according to time segments for subsequent subdivision and index association. Finally, they are summarized to form a stable and complete structured sequence to obtain the initial segmented record.

[0052] Based on the initial segmented records obtained above, entries containing complete sampling time and numerical information are selected for processing. First, the timestamps of all entries are aligned with the zero point. By comparing the data of the first 100 sampling moments, it is confirmed whether the initial delay needs to be corrected. If it is observed that the timestamp of the earliest sample deviates from the expected zero point by more than 0.01 seconds, the corresponding time compensation amount is inserted into the record and the time labels of each entry are reallocated. In this process, multiple verifications are performed in combination with the pre-established reference timetable. The deviation of the samples is statistically compared to determine whether the correction amount is reasonable. If the error of most entries with the reference timetable after correction is less than 0.005 seconds, the correction value is maintained and the operation is continued. Otherwise, fine-tune the compensation amount again, and the step size of each adjustment can be set to 0.001 seconds. This setting is derived from multiple actual measurements and takes into account practical feasibility. Afterwards, the acceleration and angular velocity data are interpolated on the time axis respectively, and the interval between adjacent recording points is controlled within a fixed interval. For example, 0.005 seconds is used as a resolution unit, and missing or overlapping time segments are filled in sequentially to ensure that the time positioning of each record is accurate and continuous. After all entries have completed time correction and interpolation, the acceleration and angular velocity are mapped in a unified order, and the corresponding position reference information is identified by an index. The two types of data can be directly compared at the same time node, and finally summarized to generate a calibration alignment sequence.

[0053] Based on the calibration alignment sequence obtained above, the corresponding sound sampling information is introduced at each time node. This sound sampling information comes from the continuous recording data of the micro-microphone configured in the early stage. The sampling rate of 200 times per second is used to ensure that it matches the data volume of acceleration and angular velocity. The corresponding moment can be located by comparing the timestamps. If the audio samples at certain moments are missing, linear interpolation is used to compensate according to the waveform characteristics of the surrounding adjacent samples. At the same time, the lens position parameters are integrated with the same time resolution. The left, right, up, down, and front and back coordinates of the lens during wearing are numbered and recorded, and when integrated, they are combined with the calibration alignment sequence obtained above. The speed and angular velocity are matched one by one. When comparing these coordinates with the time period information of the sound sampling, the change in the relative position of the sound source and the lens can be confirmed. In order to ensure that all discrete data can be tightly combined in matrix form, it is necessary to ensure that the four types of data entries, acceleration, angular velocity, sound and lens position, correspond precisely to the same index. If there are too many missing values in a certain category, they will be marked and observed in subsequent processing. These marked information is used to distinguish between faulty sampling and normal sampling. After multiple mergers and tests, all records are constructed into a four-dimensional matrix structure, and the multiple information corresponding to each index is packaged and summarized, and finally summarized to generate multi-source original information.

[0054] The inertial analysis module includes:

[0055] The threshold comparison submodule extracts acceleration and angular velocity values based on multi-source raw information, compares the rate of change point by point, marks the positions above or below the target range, and generates preliminary threshold segments.

[0056] The fall capture submodule, based on the preliminary threshold segment, identifies the continuous segments of acceleration values and marks the starting and ending positions, tracks the drop phase and locks the adjacent rising phase to generate the fall period set;

[0057] The impact induction submodule verifies the angular velocity value based on the fall period set and captures the impact peak with reference to the lens offset amplitude. It classifies the peak records into a unified index and organizes the time differences to generate motion change factors.

[0058] Specifically, based on the multi-source original information obtained above, the point-by-point change rate of acceleration and angular velocity is checked, and a segmented threshold is set in the judgment process to identify situations where the acceleration and angular velocity exceed or fall below the target range. In order to more accurately select the appropriate upper and lower limits of acceleration and angular velocity, at least three hundred groups of acceleration and angular velocity samples from daily wear can be collected and labeled. These samples are divided into a normal range group and an out-of-limit range group. Then, a three-layer fully connected neural network is established. The input layer receives the vector composed of the acceleration and angular velocity at the current moment. The hidden layer contains sixty-four neurons and uses the ReLU activation function. The output layer gives a binary classification result to indicate whether the limit is exceeded. During the training process, the weights ω={ω1,ω2,…,ω n}, using cross entropy as the loss function, forward propagation and back propagation are iterated for 100 rounds for each batch of training sets containing 32 data, and the weights are continuously updated. When the accuracy of the verification set is stable and converged, the trained network is used to make a binary judgment on the real-time acceleration and angular velocity data to assist in setting the threshold. When the network determines that the limit is exceeded, it is marked as an abnormal section. If the proportion of the data exceeding the standard range exceeds 10% within a certain period of time, it is further compared with the set fixed threshold. For example, the acceleration is greater than 25m / s 2 Or an angular velocity greater than 18 rad / s is used as the warning boundary. The prediction results of this neural network are cross-checked with the fixed threshold, and the positions above or below the range are finally recorded as preliminary threshold segments.

[0059] After completing the identification of the initial threshold segment, further check the continuous distribution of acceleration values for each time period marked as abnormal. By comparing the acceleration differences between adjacent sampling points, we can determine whether a continuous downward trend occurs. For example, when the acceleration of five consecutive sampling points changes from 10m / s to 10m / s, 2 Gradually reduce to 2m / s 2If the angular velocity changes during this period are synchronized with the acceleration changes, and if there are three or more sampling points where the angular velocity rises rapidly from 5 rad / s to 15 rad / s and then drops sharply, it is considered that the spatial position is likely to have shifted significantly. To refine this process, all abnormal sections can be numbered and recorded according to the starting and ending positions, and then the acceleration can be compared section by section to see if it has been continuously lower than 5 m / s. 2 The angular velocity changes significantly to determine whether the segment has entered the drop area, and the drop area is included in the candidate fall period set. If a large drop is maintained within thirty sampling points, the segment is marked as a high-risk fall segment, otherwise it is maintained as an ordinary abnormal segment. As the candidate time periods are compared with each other, the adjacent or overlapping parts can be merged to form a more complete time period, and finally a fall period set is obtained.

[0060] After confirming the falling period set, it is necessary to detect whether there is obvious impact behavior in each period. First, find the acceleration records corresponding to the falling period set that exceed 20m / s. 2 The instantaneous peak value is detected, and it is determined whether the instantaneous peak value occurs at the same time as a significant jump in angular velocity. If the angular velocity suddenly increases from 2 rad / s to 15 rad / s within a certain time index and then drops rapidly in the next sampling, this moment is regarded as a suspected impact point. Then, combined with the offset amplitude of the lens, a comparison is performed, requiring to query whether the lens position changes by more than 5 mm from the original reference coordinate on the same time axis. If such an offset is observed and meets the peak condition together with the acceleration and angular velocity, the record at this moment and the surrounding samples are marked as the impact peak. In the case of multiple peaks, the time intervals, numerical differences and position offsets of these peaks are counted one by one. By comparing whether the intervals between multiple peaks are less than 0.1 seconds, it is determined whether there are continuous impacts. The peak information after statistics is combined with the unified index established earlier, the required parameters are recorded one by one and the time difference is verified. Finally, all impact information is integrated into motion change elements.

[0061] The acoustic recognition module includes:

[0062] The amplitude extraction submodule, based on motion change factors, sequentially scans the sound sampling sequence and detects changes in sound pressure amplitude, locates continuous segments with obvious sudden increases and marks the peak moments, summarizes the results and outputs an index group to generate an initial amplitude mark;

[0063] The frequency domain measurement submodule, based on the initial amplitude mark, splits the sound pressure waveform into multiple blocks and calculates the proportion of high-frequency energy. It identifies the peak range of prominent energy and numbers it, summarizes the energy distribution parameters corresponding to each block, and generates a high-frequency energy distribution.

[0064] The position positioning submodule defines the area based on high-frequency energy distribution, combined with the vibration characteristics of the frame and the acoustic offset rate, compares the peak range and marks the corresponding sound source position, compiles the recognition results and stores them in a tag list, and generates a sudden sound tag.

[0065] Specifically, based on the motion change factors obtained above, we first read the sound sampling sequence in the corresponding time segment one by one and compare the sound pressure value of each record with the previous record. In order to determine whether there is a significant amplitude increase, the sound pressure in the range of 30dB to 120dB can be used as a monitoring benchmark. The background noise mean of 40dB obtained through indoor and outdoor environment tests and the variance analysis results are used to set the threshold. When the cumulative sound pressure difference of three consecutive sampling points exceeds 10dB, it is regarded as an amplitude surge. This period of time is marked with an index and included in the list to be processed. The start time and end time of each surge point are counted in the list, and the distance between these time indexes can be calculated to obtain the amplitude surge. Each continuous segment is included, and then the difference between the maximum and minimum sound pressure jumps within each segment is compared to see if it is still greater than 10dB. If the fluctuation of the values of adjacent sampling points is less than 5dB, the abnormal mark of the segment is removed to ensure that only the segments that truly meet the sudden rise conditions are retained. The 5dB and 10dB mentioned here are determined by adding twice the standard deviation to the average value obtained in multiple actual environment tests. For some extremely quiet environments, this threshold range can be recalibrated according to the lower limit of the background noise. After all continuous segments and peak information are marked, the results are integrated with the time index into a list to indicate the time when the suspicious peak occurs. Finally, the initial amplitude mark is obtained by summing up.

[0066] Based on the initial amplitude mark obtained above, the sound pressure waveform with obvious sudden rise is divided into several non-overlapping time blocks in sequence. In order to refine the energy characteristics of each time block in the high-frequency region, a discrete Fourier transform can be performed in each time block. Let s(t) represent the discrete sound pressure signal in the time block. The transformation result at frequency f can be obtained by It is calculated, where T represents the number of samples contained in the time block, and t represents the discrete time index. After the calculation is completed, the amplitude values in different frequency intervals are counted, and the range of 2kHz to 10kHz is taken as the high-frequency interval and its energy proportion is recorded. If the energy proportion in this interval is greater than 30% and the duration exceeds half of the length of the time block, the time block is regarded as a paragraph with prominent high-frequency energy. In order to clarify the attribution of this paragraph in subsequent processing, these prominent blocks can be numbered and the corresponding peak frequency and amplitude mean can be recorded. If the peak frequency of each block exceeds 8kHz and the amplitude exceeds the threshold, which is obtained by conventional environmental testing and determined by taking the background noise plus 12dB, it is regarded as a strong high-frequency component marker. After all time blocks are processed, the high-frequency energy parameters of each block are summarized and the high-frequency energy distribution is uniformly generated.

[0067] Based on the high-frequency energy distribution obtained previously, each high-frequency energy block is divided into regions, combined with the frame vibration characteristics and acoustic excursion rate. To illustrate the frame vibration characteristics, an accelerometer can be pre-installed in the frame structure to obtain vibration peak records of at least 100 times per second. Through multiple wearing tests, amplitudes within the range of 0.1g to 2g are considered normal fluctuations, and those greater than 1.5g are considered strong vibration signals. The corresponding time period can be compared with the sound pressure moments. If strong vibration is detected within the same time index and the high-frequency energy accounts for more than 30%, the acoustic excursion rate of this interval is compared. The acoustic excursion rate can be determined by observing the rate of change of the incident direction of the sound source during the previously recorded moments. For example, the sound intensity centers of adjacent sampling points are differentiated in the spatial coordinate system. If the difference results for three consecutive times exceed a set azimuth change threshold (this threshold can be set to 5 degrees and adjusted according to the application), this time period and the corresponding spatial direction are marked and recorded as a suspected sound source region. Finally, the indices of these marked suspicious sound source regions are summarized along with the vibration characteristics and listed one by one in a list to generate a burst sound marker.

[0068] The AI judgment module includes:

[0069] The interval matching submodule, based on motion change elements and sudden sound markers, lists the start and end times of the fall impact segment and the suspicious sound source segment, compares the overlapping parts, summarizes the common intervals, and generates a segment correspondence set;

[0070] The peak comparison submodule compares the angular velocity peak moments with the high-frequency sound pressure peak moments one by one based on the segment correspondence set, integrates the adjacent time indexes into parallel information and saves them in a comparison list to generate multi-peak correlation information;

[0071] The danger judgment submodule, based on multi-peak correlation information, checks whether the impact and high-frequency energy appear synchronously in the same section, marks the suspected dangerous moments and records the correlation marks, and generates mutation state instructions.

[0072] Specifically, based on the motion change elements and sudden sound marks obtained above, the start and end times of the identified falling impact segments and suspicious sound source segments are arranged one by one. First, the start and end indexes of all falling impact segments are listed on the time axis, and the start and end indexes of the suspicious sound source segments are placed in the same time dimension for comparison. For each impact segment, try to find the sound source segment that intersects or overlaps with its start and end range. For the intersecting parts, sort them according to the length of the overlapping time. If the overlapping time exceeds 10% of the total time of the segment, the time zone is marked as For potential correlation intervals, in order to provide a more detailed judgment process, the segments with an overlap ratio of less than 10% can be regarded as unrelated segments, and the segments with an overlap ratio between 10% and 30% can be classified as low-correlation segments. The segments with an overlap ratio exceeding 30% can be regarded as high-correlation segments and their time offsets can be further recorded. In this process, matching can be performed according to the start and end time indexes. If the start indexes of two segments differ by less than 10 sampling cycles, they are considered to be almost synchronized. Finally, all the compared segment information is summarized and classified according to the degree of overlap and time offset to form a segment corresponding set as a whole.

[0073] Based on the corresponding set of segments obtained above, the angular velocity peak moments are compared with the high-frequency sound pressure peak moments one by one to determine whether there are extreme value points that are relatively consistent in time. First, in each highly correlated segment, the peak with an angular velocity greater than 15 rad / s is found and the sampling index of its occurrence is recorded. Then, the high-frequency sound pressure is retrieved from the same segment to see whether it exceeds the standard of the noise mean plus 20 dB. If the above conditions are met simultaneously within the adjacent index range (±5 sampling periods can be taken), the group of peaks is considered to be correlated. In this process, all correlated peak combinations can be compiled into a record. Each record contains the time index of the peak, the specific value and the offset difference between the two. If multiple peak groups that meet the conditions are found in a highly correlated segment, they are marked one by one and collected into a comparison list in the order of appearance. Finally, all the sorted comparison data are sorted to generate multi-peak correlation information.

[0074] Based on the multi-peak correlation information obtained previously, check segment by segment to see whether the impact peak and high-frequency energy amplitude peak in the same time range form a synchronous rise and appear in a short period of time. In order to more accurately confirm possible dangerous situations, further comparison is required according to the established numerical standards. If, in the same high-correlation segment, the time difference between the angular velocity peak and the sound pressure amplitude peak is less than 0.02 seconds and the amplitudes both exceed the set benchmark threshold, for example, the angular velocity is higher than 15 rad / s and the sound pressure exceeds the background average plus 20 dB, then this segment is marked as a suspected dangerous segment. If there are at least two groups of peaks that meet the conditions in the same segment, it will be marked as a high-level warning segment in the record. Finally, all suspected dangerous segments are summarized, and their corresponding time indexes and annotation information are combined to form a command output list to generate a mutation state command as a whole.

[0075] The early warning interaction module includes:

[0076] The warning trigger submodule turns on the sound and light warning device based on the sudden change state instruction and records the start time, saves and archives the triggering situation in sequence, and generates a warning trigger record;

[0077] The interface prompt submodule displays prompt icons and voice guidance on the lens display interface based on the warning trigger record, arranges interactive options into a list and waits for user operation, and generates an interface interaction map;

[0078] The user response submodule collects user feedback based on the interface interaction mapping and indicates whether to request assistance, stores the selected actions together with the records, and generates warning interaction information.

[0079] Specifically, based on the sudden change state instruction, the triggering mode of the sound and light warning device is first set and the start time is prepared to be recorded. To determine the opening condition of the sound and light warning device, the corresponding danger level label and time index can be found in the sudden change state instruction generated previously. If the danger level label reaches or exceeds the preset level obtained through multiple field tests (for example, based on fifty wearing tests, the statistical finding of the acceleration peak value is greater than 25m / s 2 And when the angular velocity peak is greater than 15rad / s, the danger level is judged to be medium to high), then the corresponding level of sound and light output is activated. Here, the sound intensity can be set to the range of 70dB to 90dB and the light warning is presented at a flashing frequency of 3 times / second to 5 times / second. The specific threshold is obtained from indoor and outdoor experiments. If the background noise in a quiet environment is lower than 40dB, 70dB is selected as the minimum warning value. If the background noise in a noisy environment is greater than 60dB, the warning value is considered to be increased to 80dB or higher, and the light warning frequency is fine-tuned in steps of 3 times / second based on crowd tests. The opening moment is synchronized and compared with the time index of the mutation state instruction with millisecond accuracy. If the time index deviates by more than 5ms, compensation is performed. All trigger actions are written in a continuous sequence and appended to a list. At the same time, it is recorded whether the operating temperature and power supply voltage of the relevant equipment are within the predetermined range (for example, temperature 0℃ to 60℃, power supply voltage 3.7V to 4.2V). If the temperature is detected to be too high or the voltage is too low, the corresponding prompt code is inserted. After the whole process is completed, all trigger situations are summarized and arranged in sequence, and finally a warning trigger record is generated.

[0080] Based on the warning trigger record, first select the start time and the predetermined warning method corresponding to each trigger situation from the record, and prepare to present the prompt icon and voice guidance on the lens display interface. In order to clarify the conditions for the appearance of the prompt icon, the danger level label and the device activation status listed in the warning trigger record can be combined. If the current record is read to show a medium or high danger level, the corresponding icon style is called (for example, a red flashing pattern or an orange jumping pattern), and the stored voice prompt material is matched. In order to ensure that the voice guidance can still be heard in a noisy environment, the prompt volume can be controlled in the range of 70dB to 80dB and the actual output sampling amplitude can be checked. If the voice file is out of range during the test, the prompt volume can be controlled in the range of 70dB to 80dB. If peak distortion is detected, the volume is lowered by 2dB and monitored again. Indoor and outdoor scenes are distinguished based on the background noise level. If the background noise of the outdoor scene is higher than 60dB, the voice prompt duration is extended to 1.5 times to allow the user sufficient reaction time. Several interactive options for users to operate are uniformly arranged on the lens display screen. These options may include dialing emergency contacts or sending location sharing. Each option has a fixed index and graphic label, and is selected by touch or voice command. Each step of the user's operation is added to the corresponding list, and after all the required prompt content is broadcast, the next step of processing is entered, and the interface interaction mapping is finally obtained.

[0081] Based on the interface interaction mapping, it is necessary to collect user feedback on each interactive option and sort out whether to request assistance. To this end, the user's touch actions or voice commands are first monitored through the option index arranged in the previous stage. If the user does not make a selection within five seconds, the feedback is considered as no operation. If the user explicitly chooses to request assistance, a specific request note is attached to the record (such as calling a pre-designated emergency contact or sending location information). If the user chooses to cancel assistance, the cancellation type is noted in the record. To prevent accidental touches, a threshold range can be defined for touch operations. For example, a touch coordinate offset of more than 5 pixels is required to confirm a press action. During voice recognition, the recognition confidence level must be set to no less than 85%. If the recognition result is lower than this value, a prompt is given to record again. All feedback content is appended to the same list in chronological order, distinguishing between successful feedback and timeout without feedback or recognition failure, and marking whether it has entered the next step of the process. After confirming that all necessary information has been collected, the data of the selected action is packaged and integrated with the record to finally generate a warning interaction information.

[0082] The linkage output module includes:

[0083] The communication connection submodule connects to external communication channels and monitors the establishment process based on the alert interaction information. After successful verification, it merges the connection status records and generates a connection confirmation result.

[0084] The message sending submodule transmits a distress message or cancels the command based on the connection confirmation result and marks the sending time, writes the corresponding data into the monitoring platform directory, and generates an external message summary;

[0085] The power retrieval submodule checks the remaining power in the frame based on the external message summary and outputs the battery life warning information to the server, records the current status of the monitoring device and puts it into the maintenance list, and generates remote linkage data.

[0086] Specifically, based on the alert interaction information, the system initiates the connection to an external communication channel and monitors the connection process. The system first reads the operation type in the interaction information to distinguish between several methods, such as making a call, sending a short message, or a network request. If the operation type is a call, the system checks whether the current network signal strength is within a specified range (for example, a mobile network signal strength greater than -90dBm or a Wi-Fi signal strength greater than -75dBm). If network conditions are insufficient, the system attempts to switch to an alternative connection method. After successful establishment, the connection status is recorded. If the operation type is a network request, the system checks the connectivity of the server address and sends a handshake verification packet. After confirming that the other party has a correct response, the connection success time is recorded. If no response is received within 5 seconds, the system retries again. If the response still fails, the system adds a failure mark and terminates the connection process. All actions are executed in a fixed order, and the time of each network switching or sending operation is recorded. All connection success and failure events are aggregated and corresponding annotations are generated. If the connection is abnormal due to insufficient device power consumption or signal fluctuations during the connection process, an additional attempt can be made based on the previously determined signal range threshold. Finally, all connection status and verification results are summarized to generate a connection confirmation result.

[0087] Based on the connection confirmation result, if the connection is confirmed to be successful, a distress message is sent or the command is canceled and the sending time is marked. In order to clarify different message types, the specific request content attached to the warning interaction information can be parsed before transmission. If the request content is to call an emergency contact, the contact number and timestamp information are included in the message and the sending process is divided into three stages: dialing, connecting, and sending coordinate information. If the request content is location sharing, the previously acquired wearable device positioning data (for example, longitude and latitude are accurate to five decimal places) is carried and the current time index is added. If the request content is to cancel assistance, an interrupt message is sent and the operation record is retained. All transmission actions are arranged in sequence in the queue. Each time a transmission is completed, the sending status and success mark are written in the corresponding list. If the transmission fails, the error type is marked for subsequent investigation. After all transmissions are completed, all sending situations and time indexes are summarized, and the success or failure of making calls and data messages is distinguished in the monitoring platform directory, and finally integrated into an external message summary.

[0088] Based on the external message summary, the executed action type and corresponding result status are read from each record, and the remaining battery power in the frame is checked. To determine the current remaining battery power, the built-in voltage measurement component is used to capture the battery voltage and compare it with temperature and load conditions. If the voltage is less than 3.7V or the estimated remaining capacity is less than 20%, this status is recorded in the table. If the temperature exceeds 0°C to 60°C, a note is also added. The subsequent battery life is estimated based on the previously determined usage time and historical power consumption rate. This calculated time is compared with the device monitoring database established by the server. If the monitoring database shows that the average power consumption rate of the device has increased by more than 15% over three consecutive measurements, the battery warning level is raised and a battery life warning message is added to the output. The current time and reported status are uniformly included in the maintenance list, and the parts requiring inspection are classified. The classification standard here can be based on empirical data of current consumption weighted values of approximately 10% to 20% when operating in high-temperature outdoor environments. Once all the information is sorted, it can be summarized into the remote linkage data set and a complete record is generated, ultimately obtaining the remote linkage data.

Claims

1. A human safety monitoring and early warning glasses based on AI analysis, characterized in that: The glasses include: The signal acquisition module collects the three-axis acceleration and three-axis angular velocity values of the glasses, records the instantaneous acceleration and angular velocity readings in segments and aligns them in time series, records the sound sampling sequence and marks the sampling time, and generates multi-source original information; The inertial analysis module, based on the multi-source raw information, screens the three-axis acceleration values and three-axis angular velocity values, compares the instantaneous rate of change, captures the impact state, and combines the lens offset amplitude to mark the corresponding time period as a suspected fall interval. It summarizes the peak value differences and summarizes the impact moment to generate motion change factors; An acoustic recognition module, based on the motion change elements, retrieves the sound sampling sequence and extracts the sound pressure amplitude changes and high-frequency energy proportions. It then analyzes the sound pressure waveform in blocks and detects peak segments. It then correlates the frame vibration characteristics with the acoustic excursion rate, compares the peak intervals, locates the suspected sound source, and generates a sudden sound signature. The AI judgment module matches the corresponding time periods of the falling impact and the shattering sound based on the motion change elements and the sudden sound markers and compares the durations. It lists each falling interval and the sound marker interval item by item, and performs correlation comparison on the angular velocity peak interval and the high-frequency sound pressure segment. At the same time, it combines infrared to identify whether there are obstacles ahead, judge whether there is a dangerous event, and generate a sudden state instruction.

2. The human safety monitoring and early warning glasses based on AI analysis according to claim 1 are characterized in that: Also includes: The early warning interaction module triggers the sound and light warning device of the glasses based on the sudden change state instruction and presents a reminder icon and voice prompt on the lens display interface to verify whether the user needs external help and generate a warning interaction message; The linkage output module connects to the external communication channel and sends a distress message or a cancellation command based on the warning interaction information. After verifying that the communication connection is successful, it retrieves the remaining power in the frame and sends a battery life warning message to the server to generate remote linkage data.

3. The human safety monitoring and early warning glasses based on AI analysis according to claim 1 are characterized in that: The signal acquisition module includes: The data acquisition submodule records the acceleration and angular velocity readings in segments based on the three-axis acceleration and angular velocity values of the glasses, divides the readings into blocks according to the reading time, aggregates the continuous measurement sequences and marks the indexes to generate the initial segmented records; A sequence marking submodule, based on the initial segmented records, performs time axis index marking and maps acceleration readings and angular velocity readings to each index, arranges the order of different sampling moments and maps position reference points to generate a calibration alignment sequence; The information integration submodule, based on the calibration alignment sequence, corresponds the acceleration readings and angular velocity readings with the sound sampling sequence and summarizes them into a unified matrix, adds the lens position information parameters and combines all discrete data to generate multi-source original information.

4. The human safety monitoring and early warning glasses based on AI analysis according to claim 1 are characterized in that: The inertia analysis module includes: A threshold comparison submodule extracts acceleration values and angular velocity values based on the multi-source original information, compares the rate of change point by point, marks the positions above or below the target range, and generates a preliminary threshold segment; The fall capture submodule identifies continuous segments of acceleration values based on the preliminary threshold segment and marks the start and end positions, tracks the drop phase and locks the adjacent rise phase to generate a fall period set; The impact summarization submodule verifies the angular velocity value based on the falling period set and captures the impact peak value with reference to the lens offset amplitude, classifies the peak value records into a unified index and sorts out the time differences to generate motion change elements.

5. The human safety monitoring and early warning glasses based on AI analysis according to claim 1 are characterized in that: The acoustic recognition module includes: The amplitude extraction submodule sequentially scans the sound sampling sequence and detects the change in sound pressure amplitude based on the motion change factor, locates the continuous segments with obvious sudden increases and marks the peak moments, summarizes the results and outputs an index group to generate an initial amplitude mark; The frequency domain calculation submodule, based on the initial amplitude mark, splits the sound pressure waveform into multiple blocks and calculates the high-frequency energy ratio, identifies the peak range of prominent energy and numbers it, summarizes the energy distribution parameters corresponding to each block, and generates a high-frequency energy distribution; The position positioning submodule defines the area based on the high-frequency energy distribution, combined with the vibration characteristics of the frame and the acoustic offset rate, compares the peak range and marks the corresponding sound source position, compiles the recognition results and stores them in a tag list, and generates a sudden sound tag.

6. The human safety monitoring and early warning glasses based on AI analysis according to claim 1 are characterized in that: The AI determination module includes: The interval matching submodule lists the start and end times of the fall impact segment and the suspicious sound source segment based on the motion change elements and the sudden sound markers, compares the overlapping parts, summarizes the common intervals, and generates a segment corresponding set; The peak comparison submodule compares the angular velocity peak moments with the high-frequency sound pressure peak moments one by one based on the segment corresponding set, integrates adjacent time indexes into parallel information and saves them in a comparison list to generate multi-peak correlation information; The danger judgment submodule checks whether the impact and high-frequency energy appear synchronously in the same section based on the multi-peak correlation information, marks the suspected dangerous moment and records the correlation mark, and generates a sudden change state instruction.

7. The human safety monitoring and early warning glasses based on AI analysis according to claim 2 are characterized in that: The early warning interaction module includes: The warning trigger submodule turns on the sound and light warning device based on the sudden change state instruction and records the start time, saves and archives the triggering conditions in sequence, and generates a warning trigger record; An interface prompt submodule, based on the warning trigger record, displays prompt icons and voice guidance on the lens display interface, arranges interactive options into a list and waits for user operation, and generates an interface interaction map; The user response submodule collects user feedback based on the interface interaction mapping and indicates whether to request assistance, stores the selected action together with the record, and generates warning interaction information.

8. The human safety monitoring and early warning glasses based on AI analysis according to claim 2 are characterized in that: The linkage output module includes: The communication connection submodule connects to the external communication channel and monitors the establishment process based on the warning interaction information, merges the connection status records after successful verification, and generates a connection confirmation result; The message sending submodule transmits a distress message or cancels the command based on the connection confirmation result and marks the sending time, writes the corresponding data into the monitoring platform directory, and generates an external message summary; The power retrieval submodule checks the remaining power in the frame based on the external message summary and outputs a battery life warning message to the server, records the current status of the monitoring terminal device and puts it into the maintenance list, and generates remote linkage data.

Images