A system for assessing the consciousness status of injured persons based on facial recognition

Abstract

The present invention provides a system for assessing the state of consciousness of injured persons based on facial recognition, which relates to the field of health assessment technology. It includes a facial information extraction module, a facial feature analysis module, a state of consciousness calculation module, an anomaly detection module, and a state of consciousness assessment module. In the present invention, an infrared camera is used to carefully capture the subtle movements of the muscles around the eyes, corners of the mouth, and forehead, thereby improving the accuracy and comprehensiveness of data collection. By evaluating the blinking frequency, displacement of the corners of the mouth, and stretching amplitude of the forehead muscles, the accuracy and dimension of the assessment are improved. By extracting deformation features and calculating the gradient of motion amplitude, combined with time series analysis, the assessment of the state of consciousness becomes more sensitive and accurate, and is able to detect potential changes in consciousness in a timely manner, thereby responding quickly. The density and distribution feature analysis of mutation points provide rich data interpretation for diagnosis and monitoring, so that the state of consciousness of the injured person can be quickly and accurately assessed without contacting the injured person.

Description

Technical Field

[0001] The present invention relates to the technical field of health assessment, and in particular to a system for assessing the state of consciousness of a wounded person based on facial recognition. Background Art

[0002] The field of health assessment technology encompasses a variety of methods and devices used to monitor and evaluate individual health status, particularly in clinical and remote healthcare monitoring. This field ranges from basic vital sign monitoring to complex pathological condition analysis, including the use of biosensors, medical imaging technologies, and various physiological parameter analysis methods. Furthermore, health assessment technology increasingly incorporates digital tools and algorithms to provide real-time, accurate health status information, which is crucial for disease prevention, early diagnosis, and treatment monitoring.

[0003] Among them, the facial recognition-based casualty consciousness status assessment system refers to a system that uses facial image processing technology to analyze the casualty's consciousness level, and assesses the consciousness status through facial expression recognition and eye movement analysis. Specifically, it includes using image capture devices to obtain facial images of the injured person, analyzing expression changes through facial expression recognition algorithms, and assessing the injured person's reaction ability and consciousness level through eye movement tracking. By integrating facial recognition technology with medical assessment needs, a non-invasive, rapid-response consciousness status monitoring tool is constructed.

[0004] Existing technologies rely on traditional biosensors and basic vital signs monitoring, and are unable to quickly provide information on the state of consciousness in dynamic and complex environments. Relying on basic physiological parameter analysis and medical imaging, it is difficult to interpret changes in facial expressions and their impact on the state of consciousness. In emergency medical rescue, the inability to quickly assess the precise level of consciousness of the injured person delays treatment and increases the risk to the injured. Especially when dealing with large-scale disaster responses, it cannot meet the needs of rapid deployment and efficient operation. Summary of the Invention

[0005] In order to solve the technical problems that the existing technology relies on traditional biosensors and basic vital signs monitoring, cannot quickly provide consciousness status information in dynamic and complex environments, relies on basic physiological parameter analysis and medical imaging, is difficult to interpret facial expression changes and their impact on consciousness status, cannot quickly assess the precise consciousness level of the injured in emergency medical rescue, delays treatment, increases the risk of the injured, and cannot meet the requirements of rapid deployment and efficient operation, especially when dealing with large-scale disaster response, the embodiment of the present invention provides a system for assessing the consciousness status of the injured based on facial recognition. The technical solution is as follows:

[0006] On the one hand, a system for assessing the consciousness status of a casualty based on facial recognition is provided, comprising:

[0007] The facial information extraction module uses an infrared camera to collect movement data of the eye area, mouth corners, and forehead muscles, calculates blink frequency, mouth corner displacement, and forehead muscle stretch amplitude, and obtains the key sequence of facial movement;

[0008] The facial feature analysis module extracts deformation features based on the facial motion key sequence, calculates the motion amplitude gradient of the facial area, analyzes the gradient difference, identifies feature points, and obtains facial dynamic features;

[0009] The consciousness state calculation module analyzes facial movement trends based on the facial dynamic features, calculates the movement change rate, identifies and counts the distribution of change rate mutation points, determines the impact of the injured area on the mutation points, and obtains movement trend deviation analysis results;

[0010] The anomaly detection module measures the density of mutation points based on the movement trend deviation analysis results, compares the distribution characteristics of mutation points, determines whether the number of mutation points exceeds the average value of mutation points, marks abnormal mutation points and classifies them, and generates mutation point marking data;

[0011] The consciousness state assessment module calculates the duration, distribution density and time stability of the mutation point based on the mutation point marking data, determines the type and pattern of the abnormal mutation, divides the consciousness level, and obtains the consciousness state assessment result.

[0012] Optionally, the key facial movement sequences include eyelid movement sequences, mouth corner movement sequences, and forehead expression sequences; the facial dynamic features include deformation rate, deformation period, and deformation amplitude; the movement trend deviation analysis results include mutation frequency, mutation persistence, and mutation intensity; the mutation point marking data includes anomaly timestamp, anomaly duration, and anomaly intensity level; and the consciousness state assessment results include consciousness state stability, consciousness state change frequency, and consciousness state classification level.

[0013] Optionally, the facial information extraction module includes:

[0014] The facial motion data acquisition submodule uses an infrared camera to collect movement change data of the eye area, mouth corners, and frontal muscles. It extracts the position information of facial monitoring points in adjacent time frames, calculates the contraction amplitude of the eye area, the movement trajectory of the mouth corners, and the stretch length of the frontal muscles, calibrates the movement range, and generates the facial motion trajectory.

[0015] The facial deformation analysis submodule calculates the deformation gradient of facial monitoring points in adjacent time frames based on the facial motion trajectory, selects motion points that meet the deformation criteria, analyzes the blinking and closing cycle, mouth corner offset, and frontalis muscle stretching degree in consecutive time frames, and obtains facial deformation features;

[0016] The key sequence screening submodule calls the facial deformation features, screens the facial motion sequence, determines the continuity of the motion sequence, screens stable motion points according to the time axis, calculates the facial motion pattern matching degree, determines the time period when the motion pattern is stable, and obtains the facial motion key sequence.

[0017] Optionally, the facial feature analysis module includes:

[0018] The deformation feature extraction submodule analyzes the motion trajectories of the eye area, mouth corners, and frontal muscle area based on the facial motion key sequence, identifies and marks key feature points, obtains the position change of each feature point in continuous time frames, and generates key feature motion data;

[0019] The gradient calculation submodule calls the key feature motion data, calculates the motion amplitude gradient values ​​of the eye area, mouth corners and frontalis muscle area, compares the motion point positions of consecutive time frames, analyzes the gradient change rate between adjacent time frames, filters the gradient data that exceeds the motion amplitude standard, and obtains the motion gradient feature value;

[0020] The trend analysis submodule analyzes the changing trend of facial movement based on the motion gradient eigenvalue, compares the facial dynamics in different time periods, tracks the motion state of feature points in local areas, determines the stability and change pattern of facial dynamics, and obtains the facial dynamic feature trend.

[0021] Optionally, the calculation of the motion amplitude gradient of the eye area, mouth corner and frontalis muscle area adopts the formula:

[0022]

[0023] Compare the positions of motion points in consecutive time frames, analyze the gradient change rate between adjacent time frames, filter out gradient data that exceeds the motion amplitude standard, and obtain the motion gradient characteristic value;

[0024] Among them, G ij Represents the motion gradient value of the jth feature point in the i-th region between adjacent time frames, represents the displacement of the feature point of the kth frame in the i-th region, represents the displacement of the jth feature point in the adjacent frame, w ij Represents the weight coefficient of the jth feature point in the i-th region, Δt k represents the time interval between the kth frame and the k-1th frame, α ij Represents the deformation correction parameter of the jth feature point in the i-th region.

[0025] Optionally, the consciousness state calculation module includes:

[0026] The change rate calculation submodule analyzes the change rate of facial motion amplitude based on the facial dynamic feature trend, calculates the difference in facial motion rate in adjacent time windows, selects the time interval where the change rate exceeds the set standard, and generates a motion amplitude offset feature;

[0027] The mutation point extraction submodule calls the motion amplitude offset feature, compares the change rate gradients of adjacent time windows, identifies the change rate mutation points, screens the time points where the motion rate deviates from the standard, extracts the distribution of the mutation points on the time axis, analyzes the continuity and change trend of the mutation points, and obtains the mutation point characteristic map;

[0028] The error weight allocation submodule analyzes the correspondence between the temporal position of the mutation point and the injured area based on the mutation point characteristic map, determines the degree of influence of the mutation point on the facial movement pattern, adjusts the error allocation ratio of the injured area, determines the influence range of the mutation point, and obtains the movement trend deviation analysis results.

[0029] Optionally, the anomaly detection module includes:

[0030] The mutation point density calculation submodule extracts the mutation point data and the number of mutation points within the time window based on the motion trend deviation analysis results, calculates the average density of the mutation points within the time period, and obtains the mutation point density distribution;

[0031] The anomaly screening submodule calls the mutation point density distribution, compares the mutation point distribution characteristics in adjacent time periods, calculates the change amplitude of the number of mutation points, determines whether the number of mutation points in adjacent time periods exceeds the average density, screens abnormal mutation points and marks time nodes to obtain abnormal distribution nodes;

[0032] The anomaly classification submodule analyzes the time distribution pattern of the abnormal mutation points based on the abnormal distribution nodes, calculates the weight frequency ratio of the abnormal mutation points, classifies the abnormal types of the mutation points, determines the changes in the abnormal types in each time period, and generates mutation point marking data.

[0033] Optionally, the calculation of the change density ratio of the abnormal mutation point adopts the formula:

[0034]

[0035] Classify the abnormal type of mutation point, determine the change of abnormal type in each time period, and generate mutation point marking data;

[0036] Among them, D r Represents the change density ratio of abnormal mutation points, Q e represents the number of abnormal mutation points in the e-th time interval, Q e-1 represents the number of abnormal mutation points in the previous time interval, f represents the number of time intervals, and Ze Represents the end time of the e-th time interval, Z e-1 Represents the end time of the previous time interval.

[0037] Optionally, the consciousness state assessment module includes:

[0038] The mutation point feature calculation submodule extracts the start time of the mutation point based on the mutation point marking data, calculates the mutation point interval time and duration, analyzes the time stability of the mutation point within the difference time window, and obtains the mutation time stability index;

[0039] The abnormal pattern analysis submodule calls the mutation time stability index, analyzes the change pattern of the mutation point on the time axis, compares the distribution of mutation points in different time periods, identifies the fluctuation trend of the number of mutation points, determines the type and pattern of abnormal mutations, and obtains abnormal change identification data;

[0040] The state level division submodule calculates the proportion of mutation point types based on the abnormal change identification data, determines the correspondence between the mutation pattern and the state of consciousness according to the consciousness state assessment standard, divides the consciousness state into levels, calculates the time proportion corresponding to each level, and obtains the consciousness state assessment result.

[0041] Optionally, the calculation of the mutation point type ratio adopts the formula:

[0042]

[0043] According to the consciousness state assessment criteria, the corresponding relationship between the mutation pattern and the consciousness state is determined, the consciousness state is divided into levels, the time proportion corresponding to each level is calculated, and the consciousness state assessment result is obtained;

[0044] Among them, P p represents the proportion of the p-th type of mutation point, C p represents the number of mutation points of type p, U q,p Represents the cumulative duration of the p-th type of mutation point in the q-th time segment, N q represents the total duration of the qth time segment, Represents the average duration of the p-th type of mutation point in all time segments, N q,p represents the average duration of the p-th type of mutation point in the q-th time segment, κ represents the adjustment item weight coefficient, p represents the mutation point type index, q represents the time segment index, r represents the number of time segments, and s represents the number of mutation point types.

[0045] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0046] The innovative solution uses infrared cameras to capture subtle movements of the muscles around the eyes, corners of the mouth, and forehead, improving the accuracy and comprehensiveness of data collection. By evaluating blink frequency, displacement of the corners of the mouth, and stretching of the frontal muscles, the accuracy and dimensionality of the assessment are improved. By extracting deformation features and calculating motion amplitude gradients, combined with time series analysis, the assessment of the state of consciousness becomes more sensitive and accurate, enabling timely detection of potential changes in consciousness and rapid response. The density and distribution characteristic analysis of mutation points provides rich data interpretation for diagnosis and monitoring, allowing for rapid and accurate assessment of the state of consciousness of the injured person without contacting them. This is particularly suitable for first aid scenes and battlefield environments, providing real-time data to support medical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of 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 creative work.

[0048] Figure 1 A schematic diagram of the system of the present invention;

[0049] Figure 2 Schematic diagram of the system framework of the present invention;

[0050] Figure 3 This is a flow chart of the facial information extraction module of the present invention;

[0051] Figure 4 This is a flow chart of the facial feature analysis module of the present invention;

[0052] Figure 5 This is a flow chart of the consciousness state calculation module of the present invention;

[0053] Figure 6 This is a flow chart of the anomaly detection module of the present invention;

[0054] Figure 7 This is a flow chart of the consciousness state assessment module of the present invention. DETAILED DESCRIPTION

[0055] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0056] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0057] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0058] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0059] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0060] The embodiment of the present invention provides a system for assessing the consciousness status of a wounded person based on facial recognition, such as Figure 1 As shown, the system includes:

[0061] The facial information extraction module uses an infrared camera to collect movement change data around the eyes, mouth corners, and frontal muscles. It analyzes the degree of facial movement deformation in adjacent time frames, calculates blink frequency, mouth corner displacement, and frontal muscle stretch amplitude, and obtains the key sequence of facial movement.

[0062] The facial feature analysis module extracts the deformation features of key facial areas based on the facial motion key sequence, calculates the motion amplitude gradients of the eye area, mouth corners, and frontal muscle areas, analyzes the gradient differences in consecutive time frames, and tracks facial feature points in local areas to obtain facial dynamic features;

[0063] The consciousness state calculation module analyzes the movement trends of the eye area, mouth corners, and frontal muscles based on facial dynamic features. It calculates the rate of change of facial movement within the differential time window, extracts the mutation points of the change rate, and calculates the distribution of the mutation points on the time axis. It determines the degree of influence of the injured area on the facial mutation points, assigns error weights to the injured area, and obtains the movement trend deviation analysis results.

[0064] The anomaly detection module measures the density of mutation points on the time axis based on the results of motion trend deviation analysis, compares the distribution characteristics of mutation points in adjacent time periods, determines whether the number of mutation points in adjacent time periods exceeds the average number of mutation points, screens abnormal mutation points, marks time nodes, performs anomaly classification, and generates mutation point marking data;

[0065] The consciousness state assessment module calculates the duration, distribution density and time stability of mutation points based on mutation point marking data, analyzes the change pattern of mutation points on the time axis, determines the type and pattern of abnormal mutations, divides the consciousness state level, and obtains the consciousness state assessment results.

[0066] The key sequences of facial movement include eyelid movement sequence, mouth corner movement sequence, and forehead expression sequence. Facial dynamic features include deformation rate, deformation period, and deformation amplitude. The results of motion trend deviation analysis include mutation frequency, mutation persistence, and mutation intensity. The mutation point marking data includes abnormal timestamp, abnormal duration, and abnormal intensity level. The results of consciousness state assessment include consciousness state stability, consciousness state change frequency, and consciousness state classification level.

[0067] like Figure 2 and Figure 3 As shown, the facial information extraction module includes:

[0068] The facial motion data acquisition submodule uses an infrared camera to collect movement change data of the eye area, mouth corners, and frontal muscles. It extracts the position information of facial monitoring points in adjacent time frames, calculates the contraction amplitude of the eye area, the movement trajectory of the mouth corners, and the stretch length of the frontal muscles, calibrates the movement range, and generates the facial motion trajectory.

[0069] Set the acquisition frame rate and acquisition time period, use the device to collect a series of facial continuous time frame images, and extract the position information of the facial monitoring points in adjacent time frames. The facial monitoring points are the center points of the upper and lower eyelids around the eyes, the left and right edge points of the mouth corners, and the reference point in the middle area of ​​the forehead. Use pixel coordinates to convert to actual measurement coordinates, and use the Euclidean distance formula to calculate the displacement change between moving points. The formula is Where (x1, y1) and (x2, y2) represent the coordinates of the monitoring points in adjacent time frames. Assuming that the upper eyelid of the left eye is at (120, 200) in time frame 1 and (118, 198) in time frame 2, the displacement is calculated as The displacement is converted into actual length for subsequent motion amplitude analysis. The contraction amplitude of the eye socket is calculated by the maximum and minimum displacement difference of the continuous time frame. Assuming that the maximum displacement in 10 consecutive frames is 4.2mm and the minimum displacement is 1.1mm, the contraction amplitude is 3.1mm. The movement trajectory of the mouth corner is obtained by tracking the displacement curve of the mouth corner point to obtain the moving path length. The cumulative displacement is used for calculation. For example, if the displacement of the mouth corner point in the continuous time frame is 1.0mm, 1.2mm, and 1.5mm respectively, the trajectory length is 1.0+1.2+1.5=3.7mm. The stretching length of the frontalis muscle is calculated by monitoring the forehead reference point at The vertical displacement change is calculated. Assuming that the position of the forehead reference point in the initial frame is (150, 100), the position in the fifth frame is (150, 105), and the stretching length is 5 mm, the calibration motion interval range is set by the displacement threshold. The threshold is determined by referring to the statistical sample mean and standard deviation. Assuming that the sample mean is 2.5 mm and the standard deviation is 0.8 mm, the interval with a displacement greater than the mean plus 1.5 times the standard deviation is defined as the effective motion interval, that is, the threshold is 2.5 + 1.5 × 0.8 = 3.7 mm. Based on the data of all monitoring points, a set of facial monitoring point displacement curves is generated to obtain the facial motion trajectory.

[0070] The facial deformation analysis submodule calculates the deformation gradient of facial monitoring points in adjacent time frames based on facial motion trajectory, selects motion points that meet the deformation criteria, analyzes the blinking and closing cycle, mouth corner offset, and frontalis muscle stretching degree in continuous time frames, and obtains facial deformation features.

[0071] Call the monitoring point displacement curve data to calculate the deformation gradient of the facial monitoring points in adjacent time frames, using the gradient formula Where Δd represents the displacement difference, Δt is the time difference, assuming the time interval is 0.04 seconds and the displacement difference between adjacent frames is 1.6 mm, then the gradient The motion points that meet the deformation standard are screened. The deformation standard is set based on the sample mean and standard deviation. Assuming the mean is 25 mm / s and the standard deviation is 10 mm / s, the threshold is set to the mean plus 2 times the standard deviation, that is, 25+2×10=45 mm / s. Points with a deformation gradient greater than 45 mm / s are considered valid motion points. When analyzing the blink closure cycle in continuous time frames, by tracking the change in the distance between the upper and lower eyelids, it is assumed that the distance sequence in the continuous frames is (5.2, 4.1, 2.0, 1.0, 0.8, 1.2, 3.0, 5.0) mm. The closure cycle is determined as the time required for the distance to reach the maximum. If the closing to opening takes 0.16 seconds, the closure cycle is 0.16 seconds. The mouth corner offset is calculated by calculating the horizontal and vertical displacement changes of the mouth corner point. Assuming that the cumulative horizontal displacement is 3.5 mm and the vertical displacement is 2.8 mm, the total offset is The stretching degree of the frontalis muscle is determined by the difference between the maximum and minimum values ​​in the displacement curve of the continuous time frame. Assuming that the stretching displacement difference in the continuous time frame is 6.2 mm, the stretching degree is 6.2 mm. All valid motion points and related deformation data are sorted out to generate facial deformation features.

[0072] The key sequence screening submodule calls facial deformation features, screens facial motion sequences, determines the continuity of motion sequences, screens stable motion points according to the time axis, calculates the matching degree of facial motion patterns, determines the time period when the motion patterns are stable, and obtains the key sequence of facial motion.

[0073] Set the sequence length threshold and select the data segment that meets the continuous frame number requirement. Assume that the effective continuous frame segment length in the sequence must be more than 8 frames. When judging the continuity of the motion sequence, the inter-frame interval is calculated. If the inter-frame time difference is 0.04 seconds and there is no frame skipping, it is considered a continuous sequence. When screening stable motion points according to the time axis, the motion stability judgment index is used. where σ d is the displacement standard deviation, μ d is the displacement mean. Assuming that the sequence displacement mean is 3.0 mm and the standard deviation is 0.5 mm, the stability index is The threshold is set to 0.2, and S<0.2 is judged as a stable motion point. The dynamic time warping algorithm (DTW) is used to calculate the matching degree of the facial motion pattern. The reference motion template and the test sequence data are input. Assuming that the result after DTW distance calculation is 12.5, the matching threshold is set to 15. When the DTW distance is less than 15, it is judged as a highly matching sequence. When determining the stable time period of the motion pattern, the proportion of stable motion points in the time period is judged. Assuming that 9 out of 10 frames in the time period are stable points, the proportion is 90%, which exceeds the set threshold of 80%. The time period that meets the stability and matching requirements is selected to obtain the key sequence of facial motion.

[0074] like Figure 2 and Figure 4 As shown, the facial feature analysis module includes:

[0075] The deformation feature extraction submodule analyzes the movement trajectories of the eye area, mouth corners, and frontal muscle area based on the key sequence of facial movement, identifies and marks key feature points, obtains the position change of each feature point in continuous time frames, and generates key feature motion data;

[0076] The deformation feature extraction submodule analyzes the motion trajectory of the eye area, mouth corners and frontal muscle area based on the key sequence of facial movement, extracts the continuous time frame image of the target area, locates the coordinates of the feature points through the facial feature point detection method, and sets the eye area feature points as The feature points of mouth corners are The characteristic point of the frontalis muscle is Where x is the feature point index, u is the time frame number, and the feature point displacement is calculated using the following formula: in, and are the horizontal and vertical coordinates of the x-th feature point in the u-th frame, respectively. During the monitoring process, assuming that the coordinates of the mouth corner feature point in the 8th frame are (112.4mm, 198.6mm), and in the 7th frame are (111.1mm, 197.9mm), substitute them into the calculation:

[0077] Among all feature points, the displacement validity threshold is set as the weighted standard deviation of the sample displacement mean, that is: Among them, the sample displacement mean Standard deviation σ D =0.4mmm, weight coefficient β=1.5, calculated T d =1.2+1.5×0.4=1.8mm, if If the value is less than the threshold, it is not marked as significant movement. The displacement of each feature point in the eye area, mouth corner and frontalis area is calculated in turn to generate the feature matrix: Generate key feature motion data.

[0078] The gradient calculation submodule calls key feature motion data to calculate the motion amplitude gradient values ​​of the eye area, mouth corners, and frontal muscle areas, compares the positions of motion points in consecutive time frames, analyzes the gradient change rate between adjacent time frames, filters out gradient data that exceeds the motion amplitude standard, and obtains the motion gradient feature value;

[0079] The motion amplitude gradients for the periocular, mouth corner, and frontalis regions were calculated using the formula:

[0080]

[0081] Compare the positions of motion points in consecutive time frames, analyze the gradient change rate between adjacent time frames, filter out gradient data that exceeds the motion amplitude standard, and obtain the motion gradient characteristic value;

[0082] Among them, G ij Represents the motion gradient value of the jth feature point in the i-th region between adjacent time frames, represents the displacement of the feature point of the kth frame in the i-th region, represents the displacement of the jth feature point in the adjacent frame, w ij Represents the weight coefficient of the jth feature point in the i-th region, Δt k represents the time interval between the kth frame and the k-1th frame, α ij Represents the deformation correction parameter of the jth feature point in the i-th region;

[0083] Assume that in the kth frame, the feature point of the i-th region is located at coordinates And in the k-1th frame, the corresponding jth feature point is located at By calculating the Euclidean distance between the two feature points, the displacement difference is obtained:

[0084]

[0085] Assume that the following data is measured by high-precision camera equipment:

[0086] In the kth frame, the coordinates of the feature points in the i-th region are:

[0087] In the k-1 frame, the coordinates of the j-th feature point are:

[0088] Substitute into the calculation:

[0089]

[0090] Determine the weight coefficient w of the feature point ij For example, the weight of the eye area is set to 1.2, the mouth corner area is 1.0, and the frontalis area is 0.8. Assume that the weight coefficient w of the i-th area is ij =1.0.

[0091] Time interval Δt k is the time difference between two frames. For example, if the frame rate of the camera device is 30 frames per second, then:

[0092]

[0093] Deformation correction parameter α ij It is used to adjust the deformation effect of different feature points in motion. For example, by measuring the displacement changes of feature points under different expressions, the standard deviation is calculated as α ij The value of α is assumed to be ij =0.5mm.

[0094] Substituting the above values ​​into the original formula:

[0095]

[0096] Calculate the denominator:

[0097]

[0098] Therefore, the motion amplitude gradient G ij for:

[0099]

[0100] The results show that in adjacent time frames, the feature points of the i-th region move at a speed of about 5.59 mm per second. By calculating the G of all feature points ij The motion gradient feature value of the entire face can be obtained.

[0101] The trend analysis submodule analyzes the changing trend of facial movement based on the motion gradient eigenvalue, compares the facial dynamics in different time periods, tracks the motion state of feature points in local areas, determines the stability and change pattern of facial dynamics, and obtains the facial dynamic feature trend.

[0102] The trend analysis submodule analyzes the changing trend of facial motion based on the motion gradient eigenvalue and extracts the effective gradient sequence from the input data. Where p is the feature point number, q is the time window number, and the motion gradient mean is calculated in each time window: If the gradient values ​​of M=4 feature points in the third time window are {5.2, 6.1, 5.8, 5.6} mm / s, calculate the mean:

[0103] Dynamic stability is calculated by the fluctuation coefficient: in is the standard deviation of the gradient within the window,

[0104] calculate:

[0105] The coefficient of fluctuation is: The difference rate is used to judge the trend change: If the mean of the previous time window Substitute into the calculation: If the preset threshold T Δ =0.07, then Determine whether the trend is stable. Arrange all window dynamic parameters to generate trend matrix: Get the trend of facial dynamic features.

[0106] like Figure 2 and Figure 5 As shown, the consciousness state calculation module includes:

[0107] The change rate calculation submodule analyzes the change rate of facial motion amplitude based on the trend of facial dynamic features, calculates the difference in facial motion rate within adjacent time windows, filters the time intervals where the change rate exceeds the set standard, and generates motion amplitude offset features;

[0108] The motion rate data in the continuous time window is extracted from the facial dynamic feature trend. The total monitoring time is set to 10 seconds, the time window size is 2 seconds, and there are 5 window segments in total. A high frame rate camera is used for data acquisition. The frame rate is set to 60 frames per second and the interval between each frame is 0.0167 seconds. During the detection period, the motion data of the eye area, mouth corner and frontal muscle area are collected. The average rate of the facial monitoring point in each time window is calculated. The rate mean of the 1st to 5th time windows is set as To calculate the rate of change of facial motion within adjacent time windows, the following formula is used: in, represents the rate of change of motion velocity in the vth time window, and is the mean motion rate of adjacent time windows, is the weight coefficient of the vth time window, T δ is the time window length, is the time correction term. Let the weight coefficient of the 4th time window be Time correction item Time window length T δ = 2.0 seconds, substitute the average of the 4th and 3rd time window rates into the calculation:

[0109]

[0110] After completing the rate change calculation, it is necessary to set the rate change threshold for abnormal screening. Based on historical monitoring data and experimental samples, the rate change threshold is set to 0.4 mm / s2. When , mark the time window as an abnormal time period. If the threshold is exceeded, the fourth time window is determined to be an abnormal window, and abnormal time windows are screened. All window sequences that exceed the threshold are counted to form an abnormal time interval set. The screened abnormal time intervals are compared with the rate trends in continuous time periods to confirm the distribution characteristics and change patterns of continuous abnormal windows and generate motion amplitude offset features.

[0111] The mutation point extraction submodule uses the motion amplitude offset feature to compare the change rate gradients of adjacent time windows, identify the change rate mutation points, screen the time points where the motion rate deviates from the standard, extract the distribution of the mutation points on the time axis, analyze the continuity and change trend of the mutation points, and obtain the mutation point characteristic map;

[0112] The mutation point extraction submodule calls the motion amplitude offset feature, compares the gradient of the rate of change in adjacent time windows, identifies the mutation point, and uses the gradient difference rate formula: in, is the gradient value of the rate of change in the mth time window, λ m is the weight correction coefficient, τm is the window time span, θ m is the mutation correction term, extracting the change rate of the 6th and 7th time windows Time span τ7 = 0.06s, correction term θ7 = 0.03s, weight coefficient λ7 = 1.05, substitute into the calculation:

[0113]

[0114] Set the mutation point threshold T G =25.0mm / s2, because If the threshold is exceeded, it is marked as a mutation point, and the mutation point time index set {t3, t7, t 10}, the continuity calculation uses the time interval difference rate: If t7=3.4s, t 10 =4.2s, set the reference time ΔT = 1.0s, then: Obtain the mutation point characteristic map.

[0115] The error weight allocation submodule analyzes the correspondence between the temporal position of the mutation point and the injured area based on the mutation point characteristic map, determines the degree of influence of the mutation point on the facial movement pattern, adjusts the error allocation ratio of the injured area, determines the influence range of the mutation point, and obtains the movement trend deviation analysis results.

[0116] The error weight allocation submodule analyzes the relationship between the time position of the mutation point and the injured area based on the mutation point characteristic map, and sets the mutation point position. and injured area number Calculate the influence coefficient of the mutation point on the injured area: in, is the impact value of the k-th mutation point on the n-th region, is the mutation intensity, ω n is the regional weight, β k is the time offset correction term, ψ n is the regional correction coefficient, measured by data ω2=1.2,β3=0.02ms,ψ2=0.05ms,substitute into the calculation:

[0117]

[0118] If the impact threshold T I =50.0, Mark the significantly affected area and calculate the error allocation adjustment ratio of the injured area: If the cumulative impact value of the second area The number of mutation points K = 3, we get: Obtain motion trend deviation analysis results.

[0119] like Figure 2 and Figure 6 As shown, the anomaly detection module includes:

[0120] The mutation point density calculation submodule extracts the mutation point data and number of mutation points within the time window based on the results of motion trend deviation analysis, calculates the average density of mutation points within the time period, and obtains the mutation point density distribution;

[0121] Set the time window size W α and the total time T α , calculate the number of time windows Count the number of mutation points in each time window The calculation formula of mutation point density is: in, represents the density of mutation points in the a-th time window, is the window correction coefficient, For the adjustment item, set the time window size W α = 2.0 seconds, total time T α = 10.0 seconds, calculate the number of windows N α =5, the number of mutation points in the third time window in actual monitoring Window correction factor Adjustment Substituting the values ​​into the formula: By calculating the mutation point density of all time windows, the distribution of mutation point density in each time period can be obtained, and the mutation point density distribution can be obtained.

[0122] The anomaly screening submodule calls the mutation point density distribution, compares the distribution characteristics of mutation points in adjacent time periods, calculates the change in the number of mutation points, determines whether the number of mutation points in adjacent time periods exceeds the average density, screens abnormal mutation points and marks time nodes to obtain abnormal distribution nodes;

[0123] Extract the number of mutation points in the time period, set the time window size and the total time period length, compare and analyze the number of mutation points in each time period in the input data with the average value of the mutation point density, set the total monitoring time to 12 seconds, the time window size to 2 seconds, divide the monitoring data into 6 time periods, and measure the number of mutation points in each time period. The following formula is used to calculate the change in the number of mutation points: in, is the change range of the number of mutation points in the qth time period, and are the number of mutation points in the current and previous time periods, is the time period weight coefficient, is the mean density of mutation points, is the density correction term, and the number of mutation points in the fourth time period and the third time period are Set the weight coefficient mean mutation point density Density correction Substitute into the formula to calculate:

[0124]

[0125] According to the calculation results of the change range of the number of mutation points, the threshold value T of the change range of the number of mutation points is set. τ =1.2mpoints, when When the number of mutation points fluctuates significantly, Therefore, the change in the number of mutation points in the fourth time period was determined to be an abnormal mutation interval;

[0126] Determine whether the number of mutation points in adjacent time periods exceeds the average density, and set the mutation point density benchmark value to the average number of mutation points in all time periods:

[0127]

[0128] If the number of mutation points in a certain period of time exceeds It is marked as a high-density segment, and the number of mutation points in the fourth time period is The time period marked as the abnormal mutation point is combined with all the detection data to screen out the abnormal mutation points and mark the corresponding time nodes t4=6.0 seconds and t5=8.0 seconds to obtain the abnormal distribution nodes.

[0129] The anomaly classification submodule is based on the anomaly distribution nodes, analyzes the time distribution pattern of the anomaly mutation points, calculates the weight frequency ratio of the anomaly mutation points, classifies the anomaly types of the mutation points, determines the changes of the anomaly types in each time period, and generates mutation point marking data.

[0130] Calculate the change density ratio of abnormal mutation points using the formula:

[0131]

[0132] Classify the abnormal type of mutation point, determine the change of abnormal type in each time period, and generate mutation point marking data;

[0133] Among them, D r Represents the change density ratio of abnormal mutation points, Q e represents the number of abnormal mutation points in the e-th time interval, Q e-1represents the number of abnormal mutation points in the previous time interval, f represents the number of time intervals, and Z e Represents the end time of the e-th time interval, Z e-1 Represents the end time of the previous time interval;

[0134] The user's gait data is monitored by laser sensors and inertial measurement units. The total number of mutation points is counted every 30 seconds. The abnormal mutation point is defined as the time node when the step length changes by more than 15%. The step length change data is recorded and analyzed in real time by the inertial measurement unit. The sampling frequency of the sensor is 100Hz. The number of abnormal mutation points Q e In actual testing, they are recorded as:

[0135] Q0=4 points, Q1=6 points, Q2=7 points, Q3=5 points, Q4=8 points, Q5=6 points;

[0136] The parameter f represents the number of time intervals. The data acquisition source is the preset detection period division. The total monitoring time is 180 seconds, each interval is 30 seconds, and there are 6 time intervals in total, so f = 6;

[0137] Parameter Z e Represents the end time of the e-th time interval. Based on the data collection time setting, the end time of the time interval is:

[0138] Z0 = 30 seconds, Z1 = 60 seconds, Z2 = 90 seconds, Z3 = 120 seconds, Z4 = 150 seconds, Z5 = 180 seconds;

[0139] The calculation steps are as follows:

[0140] Calculation of the sum of step size differences:

[0141]

[0142] The square root of the sum of squares of time intervals is calculated:

[0143]

[0144] Formula calculation:

[0145]

[0146] The threshold for abnormal mutation points is a step length change of more than 15%. According to gait monitoring literature, normal walking step length fluctuations range from 5% to 10%, and a change exceeding 15% can be considered abnormal. The sensor sampling frequency is selected to be 100Hz to meet the real-time monitoring accuracy requirements, and the time interval length is 30 seconds to cover the natural gait cycle without excessively prolonging the detection time.

[0147] The results show that the change density ratio of abnormal mutation points within the detection period is 0.0249. This value is used to quantify the degree of fluctuation in the number of mutation points. A higher value indicates a more drastic change in the number of mutation points. This indicator value is the input data for subsequent anomaly classification and analysis of changes in anomaly types over time periods. It helps to distinguish the types of mutation point anomalies and generate mutation point marker data.

[0148] like Figure 2 and Figure 7 As shown, the consciousness state assessment module includes:

[0149] The mutation point feature calculation submodule extracts the start time of the mutation point based on the mutation point marker data, calculates the mutation point interval time and duration, analyzes the time stability of the mutation point within the difference time window, and obtains the mutation time stability index;

[0150] The mutation point feature calculation submodule extracts the mutation point start time based on the mutation point mark data Set the total number of mutation points N δ , calculate the mutation point interval time and duration. The calculation formula for the mutation point interval time is: in, is the interval between the d-th mutation point and the previous mutation point, is the time correction factor, is the data sampling interval, is the time offset correction term. Extract the data to get the time of the third mutation point The second mutation point time Time correction factor Sampling interval Offset correction term Substituting into the formula:

[0151]

[0152] Mutation point duration The calculation formula is: in, Correction Substitute into the calculation: Mutation point time stability Using the coefficient of variation method: set up Seconds, substitute into the calculation: Get mutation temporal stability indicators.

[0153] The abnormal pattern analysis submodule uses the mutation time stability index to analyze the change pattern of the mutation point on the time axis, compare the distribution of mutation points in different time periods, identify the fluctuation trend of the number of mutation points, determine the type and pattern of abnormal mutations, and obtain abnormal change identification data;

[0154] The calculation formula for the fluctuation trend is: in, is the fluctuation trend value of the mutation point in the e-th time period, is the trend correction coefficient, is the time span, is the correction parameter. The stability index of the 4th and 3rd time periods is set up Time span Correction parameters Substitute into the calculation:

[0155]

[0156] Criteria for judging abnormal mutation types: For low volatility, For medium fluctuation, For high volatility, this calculation It belongs to the medium fluctuation type. Obtain abnormal change identification data.

[0157] The state level classification submodule calculates the proportion of mutation point types based on abnormal change identification data, determines the correspondence between mutation patterns and consciousness states according to the consciousness state assessment criteria, divides the consciousness state into levels, calculates the time proportion corresponding to each level, and obtains the consciousness state assessment results.

[0158] To calculate the proportion of mutation point types, use the formula:

[0159]

[0160] According to the consciousness state assessment criteria, the corresponding relationship between the mutation pattern and the consciousness state is determined, the consciousness state is divided into levels, the time proportion corresponding to each level is calculated, and the consciousness state assessment result is obtained;

[0161] Among them, P p represents the proportion of the p-th type of mutation point, C p represents the number of mutation points of type p, U q,p Represents the cumulative duration of the p-th type of mutation point in the q-th time segment, N q represents the total duration of the qth time segment, Represents the average duration of the p-th type of mutation point in all time segments, N q,prepresents the average duration of the p-th type of mutation point in the q-th time segment, κ represents the adjustment term weight coefficient, p represents the mutation point type index, q represents the time segment index, r represents the number of time segments, and s represents the number of mutation point types;

[0162] C p Represents the number of mutation points of type p. This value is obtained by classifying and counting the mutation points detected in the time segment. For example, within a 10-minute monitoring period, the facial expression data is counted and the number of mutation points of type 1 identified is 5;

[0163] U q,p Represents the cumulative duration of the p-th type of mutation point in the q-th time segment. This value is obtained by collecting data through video or sensors and recording the duration of the mutation points. For example, in the first time segment, the duration of the first type of mutation point is 2.5 seconds, and in the second time segment it is 1.8 seconds;

[0164] N q Represents the total duration of the qth time segment. This value is determined by the time segment division standard. For example, if each time segment is set to 10 seconds, then N q = 10 seconds;

[0165] Represents the average duration of the p-th type of mutation point in all time segments, calculated as:

[0166]

[0167] Among them, N q,p represents the average duration of the p-th type of mutation point in the q-th time segment, and r is the number of time segments. For example, in three time segments, the average duration of the first type of mutation point is 1.5 seconds, 2.0 seconds, and 1.8 seconds respectively, then the calculation result is:

[0168]

[0169] κ represents the adjustment weight coefficient, which is used to balance the influence of different types of mutation points in the calculation. It is determined through experiments and changes with data fluctuations. For example, set κ = 0.8;

[0170] Calculation process:

[0171] The first step is to calculate the square of the duration of the p-th type of mutation point in each time segment and sum them up:

[0172]

[0173] Assuming that the cumulative duration of the first type of mutation points in three time segments is 2.5 seconds, 1.8 seconds, and 3.2 seconds respectively, and the time segment length is 10 seconds, the calculation is as follows:

[0174]

[0175] Take the square root:

[0176]

[0177] The second step is to calculate the molecular part:

[0178] |C p |×0.444;

[0179] Let C p =5:

[0180] 5×0.444=2.22;

[0181] The third step is to calculate the first part of the denominator:

[0182]

[0183] Assume there are three types of mutation points, with the number of each type being 5, 3, and 4 respectively. Then:

[0184] 5+3+4=12;

[0185] Step 4. Calculate the second part of the denominator:

[0186] Calculate the square root of the sum of the absolute values ​​of the differences between the duration of the p-th type of mutation point in the time segment and the overall average value:

[0187]

[0188] Calculated based on the above data:

[0189] |1.5-1.77|=0.27,|2.0-1.77|=0.23,|1.8-1.77|=0.03;

[0190] ∑=0.27+0.23+0.03=0.53;

[0191]

[0192] Multiply by the adjustment item weight coefficient κ = 0.8:

[0193] 0.8×0.728=0.582;

[0194] Step 5. Calculate the denominator:

[0195] 12+0.582=12.582;

[0196] The final calculation of the mutation point type ratio is:

[0197]

[0198] The results show that the p-type mutation points account for 17.65% of all mutation point types. This value can be used to analyze the distribution of mutation points and further analyze the corresponding correlation with the state of consciousness to identify the impact of the mutation pattern.

[0199] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0200] In the present invention, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, mb, mc, m ab, m ac, mb-c, m, or abc, where a, b, and c can be single or plural.

[0201] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0202] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0203] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0204] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of the system or unit, which can be electrical, mechanical or other forms.

[0205] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0206] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0207] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0208] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A system for assessing the consciousness of a casualty based on facial recognition, characterized in that: The system comprises: The facial information extraction module uses an infrared camera to collect movement data of the eye area, mouth corners, and forehead muscles, calculates blink frequency, mouth corner displacement, and forehead muscle stretch amplitude, and obtains the key sequence of facial movement; The facial feature analysis module extracts deformation features based on the facial motion key sequence, calculates the motion amplitude gradient of the facial area, analyzes the gradient difference, identifies facial feature points, and obtains facial dynamic features; The consciousness state calculation module analyzes facial movement trends based on the facial dynamic features, calculates the movement change rate, identifies and counts the distribution of change rate mutation points, determines the impact of the injured area on the mutation points, and obtains movement trend deviation analysis results; The anomaly detection module measures the density of mutation points based on the movement trend deviation analysis results, compares the distribution characteristics of mutation points, determines whether the number of mutation points exceeds the average value of mutation points, marks abnormal mutation points and classifies them, and generates mutation point marking data; The consciousness state assessment module calculates the duration, distribution density and time stability of the mutation point based on the mutation point marking data, determines the type and pattern of the abnormal mutation, divides the consciousness level, and obtains the consciousness state assessment result.

2. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 1, characterized in that: The key facial movement sequences include eyelid movement sequences, mouth corner movement sequences, and forehead expression sequences; the facial dynamic features include deformation rate, deformation period, and deformation amplitude; the movement trend deviation analysis results include mutation frequency, mutation persistence, and mutation intensity; the mutation point marking data include abnormal timestamp, abnormal duration, and abnormal intensity level; the consciousness state assessment results include consciousness state stability, consciousness state change frequency, and consciousness state classification level.

3. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 1, characterized in that: The facial information extraction module includes: The facial motion data acquisition submodule uses an infrared camera to collect movement change data of the eye area, mouth corners, and frontal muscles. It extracts the position information of facial monitoring points in adjacent time frames, calculates the contraction amplitude of the eye area, the movement trajectory of the mouth corners, and the stretch length of the frontal muscles, calibrates the movement range, and generates the facial motion trajectory. The facial deformation analysis submodule calculates the deformation gradient of facial monitoring points in adjacent time frames based on the facial motion trajectory, selects motion points that meet the deformation criteria, analyzes the blinking and closing cycle, mouth corner offset, and frontalis muscle stretching degree in consecutive time frames, and obtains facial deformation features; The key sequence screening submodule calls the facial deformation features, screens the facial motion sequence, determines the continuity of the motion sequence, screens stable motion points according to the time axis, calculates the facial motion pattern matching degree, determines the time period when the motion pattern is stable, and obtains the facial motion key sequence.

4. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 1, characterized in that: The facial feature analysis module includes: The deformation feature extraction submodule analyzes the motion trajectories of the eye area, mouth corners, and frontal muscle area based on the facial motion key sequence, identifies and marks key feature points, obtains the position change of each feature point in continuous time frames, and generates key feature motion data; The gradient calculation submodule calls the key feature motion data, calculates the motion amplitude gradient values ​​of the eye area, mouth corners and frontalis muscle area, compares the motion point positions of consecutive time frames, analyzes the gradient change rate between adjacent time frames, filters the gradient data that exceeds the motion amplitude standard, and obtains the motion gradient feature value; The trend analysis submodule analyzes the changing trend of facial movement based on the motion gradient eigenvalue, compares the facial dynamics in different time periods, tracks the motion state of feature points in local areas, determines the stability and change pattern of facial dynamics, and obtains the facial dynamic feature trend.

5. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 4, characterized in that: The motion amplitude gradient of the eye area, mouth corner and frontalis area is calculated using the formula: Compare the positions of motion points in consecutive time frames, analyze the gradient change rate between adjacent time frames, filter out gradient data that exceeds the motion amplitude standard, and obtain the motion gradient characteristic value; Among them, G ij Represents the motion gradient value of the jth feature point in the i-th region between adjacent time frames, represents the displacement of the feature point of the kth frame in the i-th region, represents the displacement of the jth feature point in the adjacent frame, w ij Represents the weight coefficient of the jth feature point in the i-th region, Δt k represents the time interval between the kth frame and the k-1th frame, α ij Represents the deformation correction parameter of the jth feature point in the i-th region.

6. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 1, characterized in that: The consciousness state calculation module includes: The change rate calculation submodule analyzes the change rate of facial motion amplitude based on the facial dynamic feature trend, calculates the difference in facial motion rate in adjacent time windows, selects the time interval where the change rate exceeds the set standard, and generates a motion amplitude offset feature; The mutation point extraction submodule calls the motion amplitude offset feature, compares the change rate gradients of adjacent time windows, identifies the change rate mutation points, screens the time points where the motion rate deviates from the standard, extracts the distribution of the mutation points on the time axis, analyzes the continuity and change trend of the mutation points, and obtains the mutation point characteristic map; The error weight allocation submodule analyzes the correspondence between the temporal position of the mutation point and the injured area based on the mutation point characteristic map, determines the degree of influence of the mutation point on the facial movement pattern, adjusts the error allocation ratio of the injured area, determines the influence range of the mutation point, and obtains the movement trend deviation analysis results.

7. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 1, characterized in that: The anomaly detection module includes: The mutation point density calculation submodule extracts the mutation point data and the number of mutation points within the time window based on the motion trend deviation analysis results, calculates the average density of the mutation points within the time period, and obtains the mutation point density distribution; The anomaly screening submodule calls the mutation point density distribution, compares the mutation point distribution characteristics in adjacent time periods, calculates the change amplitude of the number of mutation points, determines whether the number of mutation points in adjacent time periods exceeds the average density, screens abnormal mutation points and marks time nodes to obtain abnormal distribution nodes; The anomaly classification submodule analyzes the time distribution pattern of the abnormal mutation points based on the abnormal distribution nodes, calculates the weight frequency ratio of the abnormal mutation points, classifies the abnormal types of the mutation points, determines the changes in the abnormal types in each time period, and generates mutation point marking data.

8. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 7, characterized in that: The calculation of the change density ratio of abnormal mutation points adopts the formula: Classify the abnormal type of mutation point, determine the change of abnormal type in each time period, and generate mutation point marking data; Among them, D r Represents the change density ratio of abnormal mutation points, Q e represents the number of abnormal mutation points in the e-th time interval, Q e-1 represents the number of abnormal mutation points in the previous time interval, f represents the number of time intervals, and Z e Represents the end time of the e-th time interval, Z e-1 Represents the end time of the previous time interval.

9. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 1, characterized in that: The consciousness state assessment module includes: The mutation point feature calculation submodule extracts the start time of the mutation point based on the mutation point marking data, calculates the mutation point interval time and duration, analyzes the time stability of the mutation point within the difference time window, and obtains the mutation time stability index; The abnormal pattern analysis submodule calls the mutation time stability index, analyzes the change pattern of the mutation point on the time axis, compares the distribution of mutation points in different time periods, identifies the fluctuation trend of the number of mutation points, determines the type and pattern of abnormal mutations, and obtains abnormal change identification data; The state level division submodule calculates the proportion of mutation point types based on the abnormal change identification data, determines the correspondence between the mutation pattern and the state of consciousness according to the consciousness state assessment standard, divides the consciousness state into levels, calculates the time proportion corresponding to each level, and obtains the consciousness state assessment result.

10. The system for assessing the state of consciousness of a casualty based on facial recognition according to claim 9, characterized in that: The calculation of the mutation point type ratio adopts the formula: According to the consciousness state assessment criteria, the corresponding relationship between the mutation pattern and the consciousness state is determined, the consciousness state is divided into levels, the time proportion corresponding to each level is calculated, and the consciousness state assessment result is obtained; Among them, P p represents the proportion of the p-th type of mutation point, C p represents the number of mutation points of type p, U q,p Represents the cumulative duration of the p-th type of mutation point in the q-th time segment, N q represents the total duration of the qth time segment, Represents the average duration of the p-th type of mutation point in all time segments, N q,p represents the average duration of the p-th type of mutation point in the q-th time segment, κ represents the adjustment item weight coefficient, p represents the mutation point type index, q represents the time segment index, r represents the number of time segments, and s represents the number of mutation point types.

Images