A method and system for detecting and prioritizing wounded personnel from an aerial view by using a UAV for medical service training, and a storage medium

Abstract

The application discloses a kind of unmanned plane overhead wounded detection and priority recommendation method and system for medical service training and storage medium, it is related to computer vision technical field.For realizing the quick identification of wounded in training scene and priority treatment decision.First, obtain the overhead video stream of training area, carry out ground texture suppression and illumination difference processing to continuous frame, construct ground structure gradient field and detect human body profile curvature abnormal section, generate profile confidence chart;Subsequently, according to the confidence chart positioning head and shoulder area, extract pitch attitude, bleeding distribution and limb movement characteristics, obtain motion frequency index;Respiratory rhythm is estimated based on head and shoulder area and pitch attitude, and is fused with bleeding proportion and motion frequency, constructs life-behavior joint parameter vector;Finally, input preset weight matrix to calculate the severity of illness emergency degree score of joint parameter vector, divide treatment priority and generate spatial distribution chart, realize the comprehensive quantitative evaluation and visual management of wounded state.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, specifically to a method, system, and storage medium for drone-based overhead wound detection and priority recommendation for medical support training. Background Technology

[0002] As training scenarios gradually transition from static displays to highly realistic, information-dense mission drills, rapidly identifying ground casualties and assessing their potential urgency has become a fundamental capability. Unmanned aerial vehicles (UAVs) can cover large areas in a short time, and their overhead view provides greater transparency, making "automatic aerial perception of casualty status" a real need. However, training environments often involve complex terrain textures, uneven lighting, and diverse background disturbances, making manual observation both time-consuming and prone to missing crucial details. Therefore, developing an automated method capable of stably extracting multi-source indicators such as human contours, bleeding features, posture, and respiration in dynamic environments is increasingly necessary.

[0003] Existing casualty detection and assessment solutions based on overhead video mostly focus on single visual features, such as color-based region segmentation or contour extraction based on ordinary edge detection. These methods are prone to false detections and boundary damage on surfaces with complex textures or significant lighting variations. Furthermore, many systems only focus on a single frame of a static image, failing to temporally model the dynamic behavioral information contained in consecutive frames (such as subtle fluctuations in respiratory brightness and changes in limb movement rhythms), resulting in a superficial representation of the casualty's actual condition. In addition, some methods simply use different indicators side-by-side, lacking a fusion strategy that reflects the combined mechanism of vital signs and behavioral characteristics, making the assessment results unstable or biased in complex scenarios. To obtain priority judgments that more closely resemble real-world treatment logic in the training environment, an analytical method is needed that can stably extract multimodal physiological and behavioral features of the human body in complex backgrounds and perform structured fusion of these features. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method, system, and storage medium for drone-based overhead casualty detection and priority recommendation for medical support training, thus resolving the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for drone-based overhead-view casualty detection and priority recommendation for medical support training, comprising the following steps: S1. Acquiring an overhead-view video stream of the training area, performing ground texture suppression and illumination difference processing on consecutive frames, constructing a surface structure gradient field based on the pixel density change rate, detecting curvature anomalies in the human contour within the gradient field, and performing temporal consistency checks on the curvature anomalies to generate a contour confidence map; S2. Based on the regions in the contour confidence map with confidence scores higher than a set threshold, performing local illumination difference and skin color probability estimation, determining the head and shoulder region range by combining the relative surface temperature gradient, calculating the casualty pitch attitude parameters based on the spatial deviation between the head and shoulder region and the overall contour, and simultaneously extracting the bleeding area within the confidence region. The color distribution and area ratio are calculated, and motion vectors are calculated for limb pixels in consecutive frames to obtain motion frequency indicators; S3. Based on the head and shoulder region and pitch posture parameters, the pixel set with the highest visual credibility is selected, and a slight brightness fluctuation curve is obtained by accumulating gray-level differences between frames. The pitch posture parameters are used to compensate the angle of the fluctuation curve to obtain the compensated respiratory rhythm estimate. The respiratory rhythm estimate is fused with the bleeding area ratio and motion frequency indicators to construct a life-behavior joint parameter vector; S4. The respiratory rhythm, bleeding ratio and motion frequency in the joint parameter vector are input into a preset weight matrix for weighted fusion calculation to generate an injury urgency score. According to the score range, the injured are divided into different treatment priority levels, and a spatial distribution map containing priority indicators is generated.

[0006] Furthermore, the process of acquiring the overhead video stream of the training area, performing ground texture suppression and illumination difference processing on consecutive frames, and constructing the surface structure gradient field based on the pixel density change rate is as follows: Multi-scale Gaussian smoothing is performed on consecutive frames in the overhead video stream. Complex texture regions are identified by calculating the standard deviation of pixel values ​​within a local window. Texture feature matching based on local binary patterns is performed in these complex texture regions, and successfully matched texture regions are marked as the ground background. Pixel-by-pixel difference operations are performed between the current frame and the background reference frame to obtain a difference image with ground texture removed. The pixel density change rate is calculated on the difference image using a sliding window approach. The first derivative of the pixel density is calculated in both the horizontal and vertical directions. The derivative results in the two directions are vectorized to form a gradient field reflecting the changes in surface undulations.

[0007] Furthermore, the specific process of detecting curvature anomalies in the human contour in the gradient field and performing temporal consistency checks on these anomalies to generate a contour confidence map is as follows: Continuous contour lines are extracted in the gradient field using an edge tracking algorithm. The curvature value is calculated for each sampling point on the contour line. A curvature change monitoring window is set. When the standard deviation of the curvature value within the window exceeds a set threshold, the contour segment is marked as an anomaly. In consecutive frames of the video sequence, motion trajectory records are established for each anomaly segment. The spatial position change of the anomaly segment between adjacent frames is calculated. Confidence is accumulated for anomalies with position changes less than a displacement threshold, and confidence is attenuated for anomalies with severe positional jitter. A confidence map is generated based on the confidence distribution to characterize contour stability.

[0008] Furthermore, based on the regions in the contour confidence map with confidence levels higher than a set threshold, the specific process of performing local illumination difference and skin color probability estimation, combined with the relative surface temperature gradient, to determine the head and shoulder region range is as follows: Local illumination compensation is performed within the confidence-compliant region, and the local contrast is adjusted by calculating the average brightness of neighboring pixels. Skin color features are extracted from the brightness-equalized image, and the distribution density of Cr and Cb components in the YCbCr color space is statistically analyzed. Simultaneously, the temperature distribution in the infrared image is analyzed, the temperature difference between each pixel and its eight neighbors is calculated, the correspondence between skin color density and temperature gradient is established, and regions that simultaneously satisfy high skin color density and stable temperature distribution are found. Based on the human head and shoulder size ratio constraint, the minimum bounding rectangle range of the head and shoulder region is determined.

[0009] Furthermore, the specific process of calculating the patient's pitch posture parameters based on the spatial deviation between the head and shoulder region and the overall contour, extracting the color distribution and area ratio of the bleeding area within the confidence region, and calculating motion vectors for limb pixels across consecutive frames to obtain motion frequency indices is as follows: Calculate the offset vector of the head and shoulder region center relative to the center of the circumscribed rectangle of the human contour; obtain the orientation angle of the head and shoulder region through principal component analysis; calculate the pitch angle by combining the magnitude and direction of the offset vector; convert the human contour region to the HSV color space; set the hue range threshold and saturation threshold for the blood color feature; extract connected bleeding regions using a region growing algorithm; and calculate the ratio of the number of pixels in the bleeding region to the total number of pixels in the human contour; calculate the motion displacement of corresponding pixel blocks across consecutive frames; obtain the motion vector by analyzing the intensity changes of the pixel blocks in the spatiotemporal domain; and statistically analyze the amplitude distribution and frequency of the motion vectors to extract the amplitude and frequency indices of the main motion features.

[0010] Furthermore, the process of obtaining a micro-brightness fluctuation curve by accumulating inter-frame gray-level differences and using pitch attitude parameters to perform angle compensation on the fluctuation curve to obtain the compensated respiratory rhythm estimate is as follows: A set of pixels with stable texture features is selected within a defined head and shoulder region; the gray-level difference between corresponding pixels in adjacent frames is calculated; the gray-level difference is accumulated over time to form the original fluctuation curve; the influence of pitch attitude parameters on gray-level changes is analyzed, and an angle-gray-level change compensation function is established to compensate the original fluctuation curve using a sine function based on the pitch angle; periodic signal components are extracted from the compensated fluctuation curve, and the respiratory cycle is identified using a peak detection algorithm to calculate the number of breaths per unit time, thus obtaining the respiratory rhythm estimate.

[0011] Furthermore, the specific process of fusing the estimated respiratory rhythm with the proportion of bleeding area and the exercise frequency index to construct a life-behavior joint parameter vector is as follows: The estimated respiratory rhythm is normalized and mapped to a standard physiological parameter range; the proportion of bleeding area is graded and quantified according to the severity of the injury to obtain a standardized bleeding score; the exercise frequency index is converted into an exercise capacity assessment value, and the exercise activity coefficient is calculated based on the exercise amplitude and frequency; the standardized respiratory rhythm, bleeding score, and exercise activity coefficient are combined in a predetermined order to form a life-behavior joint parameter vector containing three dimensions.

[0012] Further, step S4 includes the following steps: establishing a weight matrix according to medical support training guidelines, assigning different weight coefficients to the three parameters of respiratory rhythm, bleeding ratio, and movement frequency, performing matrix operation on the joint parameter vector and weight matrix to obtain the injury urgency score; setting a score threshold range, mapping the injury urgency score to the corresponding treatment priority level; associating the spatial location information of each injured person with the corresponding priority level, and visually displaying it on an electronic map using identifiers of different colors and shapes to form a spatial distribution map containing priority information.

[0013] A system for drone-based overhead-view casualty detection and priority recommendation for medical support training includes the following modules: a contour detection module, used to acquire overhead-view video streams of the training area, perform ground texture suppression and illumination difference processing on consecutive frames, construct a surface structure gradient field based on pixel density change rate, detect curvature aberrations in the human contour within the gradient field, perform temporal consistency checks on the curvature aberrations, and generate a contour confidence map; and a posture and bleeding recognition module, used to perform local illumination difference and skin color probability estimation on areas with confidence scores higher than a set threshold in the contour confidence map, determine the head and shoulder region range based on the relative surface temperature gradient, calculate the casualty pitch posture parameters based on the spatial deviation between the head and shoulder region and the overall contour, and simultaneously extract the color distribution and area ratio of bleeding areas within the confidence region. Motion vectors are calculated for limb pixels across consecutive frames to obtain motion frequency indices. A respiratory rhythm estimation module selects the set of pixels with the highest visual reliability based on head and shoulder region and pitch posture parameters. It obtains a micro-brightness fluctuation curve through inter-frame grayscale difference accumulation, compensates for the angle of the fluctuation curve using pitch posture parameters, and obtains a compensated respiratory rhythm estimate. This estimate is then fused with the bleeding area ratio and motion frequency indices to construct a life-behavior joint parameter vector. A priority assessment module inputs the respiratory rhythm, bleeding ratio, and motion frequency from the joint parameter vector into a preset weight matrix for weighted fusion calculation, generating an injury urgency score. Based on the score range, the injured are divided into different treatment priority levels, generating a spatial distribution map containing priority identifiers.

[0014] A computer-readable storage medium storing a processor-executable program, which, when executed by a processor, performs the method described above.

[0015] The present invention has the following beneficial effects:

[0016] (1) A method for UAV-based top-down casualty detection and priority recommendation for medical support training. Under the UAV's top-down view, the stability of early identification is significantly improved through a continuous frame enhancement mechanism. The structural gradient field constructed by ground texture suppression, illumination difference, and pixel density change rate can effectively suppress false edges caused by complex terrain, making it easier to distinguish abnormal curvature segments of the human body contour from the background. Combined with temporal consistency verification, the contour remains stable between frames, reducing detection drift caused by illumination changes and background disturbances, and providing a reliable foundation for subsequent identification. At the same time, local illumination compensation and skin color probability estimation are performed within the confidence region and fused with the temperature gradient, so that the head and shoulder region can still be clearly located under non-uniform illumination, low contrast, and complex background conditions, avoiding region misjudgment, offset, and boundary distortion that are prone to occur in the prior art, thereby providing accurate input for pose estimation, bleeding extraction, and motion analysis.

[0017] (2) A system for UAV-based top-down casualty detection and priority recommendation for medical support training achieves more realistic injury assessment by fusing and analyzing multi-source vital behavior parameters. The system utilizes a brightness fluctuation curve constructed from a reliable pixel set, combined with pitch angle compensation, to improve the reliability of respiratory rhythm estimation under top-down conditions, eliminating limitations imposed by viewpoint shifts and pixel micro-motions. Furthermore, respiratory rhythm, bleeding area ratio, and movement frequency are constructed as a joint parameter vector and weighted and fused using a preset weight matrix, enabling the urgency score to simultaneously reflect vital signs, injury extent, and voluntary movement ability, overcoming the priority distortion problem caused by single indicators in traditional methods. The generated priority spatial distribution map provides a clear basis for rescue route planning and resource allocation in medical support training, achieving more practical automated classification and labeling effects.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for drone-based overhead casualty detection and priority recommendation for medical support training, according to the present invention.

[0020] Figure 2 This is a system flowchart of an unmanned aerial vehicle (UAV) system for detecting and prioritizing casualties in medical support training, according to the present invention. Detailed Implementation

[0021] This application provides a method, system, and storage medium for drone-based overhead casualty detection and priority recommendation for medical support training. It addresses the problem that casualty status recognition in complex training scenarios is easily affected by changes in lighting, background interference, and viewing angle limitations, leading to unstable positioning, incomplete feature extraction, and inaccurate injury assessment.

[0022] The overall concept of the solution in this application embodiment is as follows:

[0023] Based on top-down video acquired by UAVs, stable extraction of human body contours, multimodal feature fusion analysis of key parts, and temporal acquisition of vital and behavioral features are achieved in complex environments. Finally, an emergency scoring model that integrates multiple features is constructed, thereby outputting patient priority recommendation results for medical support training.

[0024] Please see Figure 1This invention provides a technical solution: a method for drone-based overhead-view casualty detection and priority recommendation for medical support training, comprising the following steps: S1. Acquiring an overhead-view video stream of the training area, performing ground texture suppression and illumination difference processing on consecutive frames, constructing a surface structure gradient field based on the pixel density change rate, detecting curvature anomalies in the human contour within the gradient field, and performing temporal consistency checks on the curvature anomalies to generate a contour confidence map; S2. Based on the areas in the contour confidence map with confidence scores higher than a set threshold, performing local illumination difference and skin color probability estimation, determining the head and shoulder region range by combining the relative surface temperature gradient, calculating the casualty pitch attitude parameters based on the spatial deviation between the head and shoulder region and the overall contour, and simultaneously extracting the color of the bleeding area within the confidence region. S3. Based on the head and shoulder region and pitch posture parameters, select the pixel set with the highest visual credibility, obtain the micro-brightness fluctuation curve through the accumulation of gray-level differences between frames, use the pitch posture parameters to perform angle compensation on the fluctuation curve, obtain the compensated respiratory rhythm estimate, fuse the respiratory rhythm estimate with the bleeding area ratio and the motion frequency index to construct the life-behavior joint parameter vector; S4. Input the respiratory rhythm, bleeding ratio and motion frequency in the joint parameter vector into the preset weight matrix for weighted fusion calculation to generate the injury urgency score, divide the injured into different treatment priority levels according to the score range, and generate a spatial distribution map containing priority indicators.

[0025] In this implementation scheme, S1: Acquire the overhead video stream of the training area and perform preprocessing and contour candidate generation on consecutive frames to provide stable boundary information for subsequent target localization. Ground texture suppression: This refers to using image processing techniques such as filtering and texture matching to reduce the interference of high-frequency details generated by the ground surface (such as soil, grass, and rubble) in the image on target edge recognition, making the true human body boundary more prominent on the gradient or edge map. Illumination difference: This refers to using brightness comparison between frames or within a local window to eliminate artifacts caused by uneven illumination, making the pixel response of the same object more consistent under different lighting conditions. Pixel density change rate and ground structure gradient field: The pixel density change rate is a measure of the spatial variation of local pixel intensity or texture density. The gradient field is a field synthesized by differentiating this measure in the horizontal and vertical directions, used to characterize ground undulations or structural changes. Curvature anomaly segment: On the gradient / edge curve, a continuous segment where the curvature value deviates significantly from the neighborhood average, usually corresponding to the inflection point of complex edges or target boundaries. Contour confidence map: A two-dimensional image representation obtained by mapping the edge stability (such as temporal consistency and curvature significance) of each pixel or contour segment to a confidence value. Through the above processing, human boundary candidates can be stably extracted under complex backgrounds and lighting changes, significantly reducing false edges and boundary jitter-induced missed / false detections, providing a highly reliable input area for head and shoulder localization and pose estimation in S2, and ensuring the data coupling and robustness of the entire link. S2: Locates the head and shoulder region within the high-confidence area output by S1 and simultaneously extracts visual cues related to the injury (skin color probability, surface temperature gradient, bleeding color distribution, limb motion vectors, etc.). Local illumination difference and skin color probability estimation: Local illumination difference normalizes and differentiates the pixel brightness within the confidence area. Skin color probability estimation refers to inferring the probability distribution of a pixel belonging to skin based on pixel component statistics in a specific color space (such as brightness-color difference or HSV). Surface temperature relative gradient: Refers to the temperature difference between a pixel and its neighboring pixels calculated using infrared / thermal imaging data, used to distinguish between covered objects, exposed skin, or heat sources. Spatial deviation and pitch posture parameters: Spatial deviation is the positional offset of the head and shoulder region center relative to the centroid of the overall contour; pitch posture parameters describe the tilt angle of the human body in the vertical plane (e.g., the quantitative representation of supine / prone position). Motion vectors and motion frequency: Motion vectors are the displacement vectors of the same pixel or pixel block between consecutive frames, and motion frequency is a statistical index of the frequency of motion amplitude over time. This step overcomes the unreliability of single visual cues under occlusion, low contrast, or nighttime conditions by extracting multimodal (color + temperature + motion) features in parallel; pitch posture parameters provide the necessary geometric relationships for subsequent respiratory signal angle compensation and behavioral interpretation; bleeding ratio and motion frequency directly quantify the severity of injury and voluntary movement ability, constituting key indicators for subsequent fusion evaluation.S3: Extract minute brightness fluctuations from the visible, reliable pixel set of the head and shoulders region to estimate respiratory rhythm, and perform angle compensation based on pitch attitude on this signal. Simultaneously, combine this signal with hemorrhage and motion indicators to form a joint parameter vector. Highest Pixel Confidence Set: Within the head and shoulders region, a subset of pixels most likely reflecting true physiological changes is selected using metrics such as confidence level, skin color probability, and thermal stability to reduce noise impact. Inter-frame Gray-level Difference Accumulation and Minor Brightness Fluctuation Curve: Cumulative or short-time Fourier / temporal analysis is performed on the gray-level changes of selected pixels over time to extract periodic minute fluctuations. The period of these fluctuations corresponds to the respiratory rate. Angle Compensation: The pitch attitude parameters obtained in S2 are used to correct projection scaling or brightness distribution distortion caused by the imaging angle, making the extracted periodic signal closer to the true physiological rhythm. Vital-Behavioral Joint Parameter Vector: The estimated respiratory rhythm, hemorrhage area ratio, and motion frequency are organized into a vector form according to their defined order, serving as standardized input for subsequent scoring. By selecting high-reliability pixels and using posture compensation, the reliability of identifying weak vital signs (such as breathing) under overhead conditions is improved. Breathing, bleeding, and movement are converted into structured vectors, facilitating unified quantitative assessment and weight calculation in S4, thereby achieving coordinated judgment of vital signs and behavioral capabilities. S4: The joint parameter vector is input into a preset weight matrix for weighted fusion, calculating the urgency score and prioritizing treatment according to score intervals. The priority locations are then marked on a spatial distribution map for dispatching purposes. Preset weight matrix: A coefficient matrix set according to training objectives and empirical rules, used to weight the relative importance of each item in the joint parameter vector (e.g., breathing, bleeding, movement); weights can be determined offline using historical exercise data. Urgency score: A quantitative score obtained after weighted summation or more mathematical mappings, reflecting the relative urgency of treatment for a single target. Priority classification and spatial distribution map: The score is mapped to a classification level (e.g., level three or four), and the geographical location and corresponding level of each injured person are marked on the training area map with icons / colors, forming a visual dispatching basis. This step unifies multi-source parameters into a weighted scoring framework, taking into account vital signs, injury presentation, and behavioral capabilities, thus avoiding biases caused by single-indicator-driven decision-making. The output classification and spatial labeling directly support rescue route planning, resource allocation, and training evaluation, forming a closed loop from the perception layer to the decision-making layer, meeting the timeliness and assessability requirements of medical support training.

[0026] Specifically, the process of acquiring the overhead video stream of the training area, performing ground texture suppression and illumination difference processing on consecutive frames, and constructing the surface structure gradient field based on the pixel density change rate is as follows: Multi-scale Gaussian smoothing is performed on consecutive frames in the overhead video stream. Complex texture regions are identified by calculating the standard deviation of pixel values ​​within a local window. Texture feature matching based on local binary patterns is performed in these complex texture regions, and successfully matched texture regions are marked as the ground background. Pixel-by-pixel difference operations are performed between the current frame and the background reference frame to obtain a difference image with ground texture removed. The pixel density change rate is calculated on the difference image using a sliding window approach. The first derivative of the pixel density is calculated in both the horizontal and vertical directions. The derivative results in the two directions are vectorized to form a gradient field reflecting the changes in surface undulations.

[0027] In this implementation, the purpose of constructing the surface structure gradient field through the above operations is to separate the parts of surface texture, brightness, or material changes related to human geometric boundaries from background interference, enabling subsequent contour tracking algorithms to operate at the locations where structural changes are most pronounced. The pixel density change rate essentially reflects the density variation of pixel values ​​within a local region, and its first derivative can highlight the edge locations of surface structure changes. To improve quantization accuracy, this embodiment calculates the pixel density change rate for each sliding window in the difference image. ,in: ;in: The grayscale value of the h-th pixel within the window; The average pixel value of this window; The number of pixels in the window. The horizontal and vertical derivatives of the rate of change of pixel density are denoted as follows: and Through vector composition: ;in: : The gradient of the pixel density change rate in the horizontal direction; : The gradient of the pixel density change rate in the vertical direction; This represents the comprehensive gradient value reflecting surface undulations and structural changes. The threshold is determined by collecting several frames of background samples in the training area and statistically analyzing each frame. The mean and standard deviation of the gradient field are used, and the mean plus twice the standard deviation is used as the upper bound of the background variation. When the gradient value of a pixel exceeds this upper bound, it is considered that the location may contain human-related structures. Through the above algorithm, the difference between the ground surface and the human body boundary in the gradient field is effectively amplified, making the human contour extraction more stable and providing reliable input for subsequent anomaly segment identification.

[0028] Specifically, the process of detecting curvature anomalies in the human contour within a gradient field and performing temporal consistency checks on these anomalies to generate a contour confidence map is as follows: Continuous contour lines are extracted in the gradient field using an edge tracking algorithm. The curvature value is calculated for each sampling point on the contour line. A curvature change monitoring window is set. When the standard deviation of the curvature value within the window exceeds a set threshold, the contour segment is marked as an anomaly. Motion trajectory records are established for each anomaly segment in consecutive frames of the video sequence. The spatial position change of the anomaly segment between adjacent frames is calculated. Confidence is accumulated for anomalies with position changes less than a displacement threshold, and confidence is attenuated for anomalies with severe positional jitter. A confidence map is generated based on the confidence distribution to characterize contour stability.

[0029] In this implementation scheme, the purpose of identifying curvature anomaly segments in the gradient field is to find boundary changes corresponding to the geometric features of the human body shape in a complex background. The human contour typically exhibits significant curvature abrupt changes in areas such as the shoulders, head, and limbs; therefore, curvature anomaly segments can serve as preliminary candidates for key contour segments. The curvature calculation method used in this embodiment is as follows: [The text then abruptly shifts to a different topic:] ...contour sampling point sequence... Calculate curvature : ;in: : The coordinates of the m-th sampling point on the contour; The curvature value at the m-th sampling point is used to measure the degree of contour curvature. Three local sampling points (m-1, m, and m+1) are chosen to ensure locality of curvature calculation. Within the curvature change monitoring window, if the standard deviation of the curvature values... Exceeding the window threshold If so, the segment is marked as a curvature anomaly segment. Threshold The determination method is as follows: The standard deviation distribution of curvature of normal human contour segments is statistically analyzed in the training samples. The 95th percentile of this distribution is used as the threshold to make abnormal segments more distinguishable. In the temporal consistency check, a motion trajectory sequence is established for each abnormal segment, and the spatial position difference between adjacent frames is calculated. ;in: : The coordinates of the center position of the abnormal segment in the nth frame; : Displacement between adjacent frames. If Less than the displacement threshold This indicates that the location of the anomalous segment is stable, and the cumulative confidence level Q is: ;like Greater than The confidence level is determined by the attenuation coefficient. Adjustments will be made: ;in: Temporal confidence of outlier segments; Confidence gain coefficient; : Confidence decay coefficient. The setup involves selecting a human contour segment with a stable length of approximately N frames in the training video, making N×α close to 1, thereby ensuring that the real human contour can achieve a high confidence level within a reasonable time range. The setting method is to reduce the number of frames required to attenuate to half of the original value by no more than 3-4 frames, that is... Seek Finally, interpolation is performed on the pixel plane based on the confidence values ​​to obtain a contour confidence map that characterizes contour stability, resulting in a higher numerical distribution of human-related contour segments on the map. This confidence map provides a clear and reliable boundary region for subsequent feature fusion and target localization, and effectively reduces misjudgments caused by background perturbations.

[0030] Specifically, based on the regions in the contour confidence map with confidence levels higher than a set threshold, the following process is performed to determine the head and shoulder region range by performing local illumination difference and skin color probability estimation, combined with the relative surface temperature gradient: Local illumination compensation is performed within the confidence-compliant regions, and local contrast is adjusted by calculating the average brightness of neighboring pixels. Skin color features are extracted from the brightness-equalized image, and the distribution density of Cr and Cb components in the YCbCr color space is statistically analyzed. Simultaneously, the temperature distribution in the infrared image is analyzed, the temperature difference between each pixel and its eight neighbors is calculated, the correspondence between skin color density and temperature gradient is established, and regions that simultaneously satisfy high skin color density and stable temperature distribution are found. Based on the human head and shoulder size ratio constraint, the minimum bounding rectangle range of the head and shoulder region is determined.

[0031] In this implementation, local illumination difference and skin color probability estimation are performed on the high-confidence regions selected from the contour confidence map to address the issues of uneven illumination, ground reflection interference, and skin color shifts under different lighting conditions in UAV overhead scenes. By performing local illumination compensation on the confidence regions, skin color features are made more concentrated and more suitable for subsequent statistical models. To quantify skin color distribution, this embodiment establishes a skin color probability density function for the brightness-equalized regions, which can be expressed as: ;in: Pixel position The probability density of skin color; The chromaticity component U of this pixel in the chromaticity space (corresponding to the original Cb component); The chromaticity component V of this pixel in the chromaticity space (corresponding to the original Cr component). The mean values ​​of the U and V components of skin color obtained from the training samples; Standard deviations of skin color U and V in the training samples; : Normalization factor, used to ensure that the sum of probabilities is 1. The method for determining the model is as follows: human skin color samples are collected under various indoor and outdoor lighting conditions. The distribution of each sample is statistically analyzed, and the mean and variance of the overall distribution are used as the final parameters to ensure the model's universality. Simultaneously, temperature gradient calculations are performed on the infrared images. In this embodiment, the cumulative temperature difference between each pixel and its eight neighbors is calculated. ;in: :pixel Cumulative temperature difference; The infrared temperature value of this pixel; :The eighth neighboring domain The temperature values ​​of each neighboring pixel are considered. A smaller cumulative temperature difference indicates a more stable temperature distribution, which better matches the single heat source structure of the human head and shoulders. By combining skin color probability density and cumulative temperature difference as constraints, regions with both high skin color probability and low temperature gradient are identified. Finally, based on the geometric proportions of the human head and shoulders (width-to-height ratio approximately 2:3), the spatial location of the head and shoulders region is determined using minimum bounding rectangle fitting, and the rectangular boundary is output.

[0032] Specifically, the process of calculating the patient's pitch posture parameters based on the spatial deviation between the head and shoulder region and the overall contour, extracting the color distribution and area ratio of the bleeding area within the confidence region, and calculating motion vectors for limb pixels across consecutive frames to obtain motion frequency indices is as follows: The offset vector of the head and shoulder region center relative to the center of the circumscribed rectangle of the human contour is calculated; the orientation angle of the head and shoulder region is obtained through principal component analysis; the pitch angle is calculated by combining the magnitude and direction of the offset vector; the color space within the human contour region is converted to HSV color space; the hue range threshold and saturation threshold of the blood color feature are set; connected bleeding regions are extracted using a region growing algorithm; the ratio of the number of pixels in the bleeding region to the total number of pixels in the human contour is calculated; motion displacement is calculated for corresponding pixel blocks across consecutive frames; motion vectors are obtained by analyzing the intensity changes of pixel blocks in the spatiotemporal domain; the amplitude distribution and frequency of change of the motion vectors are statistically analyzed; and the amplitude and frequency indices of the main motion features are extracted.

[0033] In this implementation scheme, the pitch attitude is calculated by the spatial deviation between the head and shoulder region and the overall contour to address the geometric distortion caused by changes in the direction and posture of the injured person's head from a top-down perspective. To obtain the attitude angle, this embodiment first calculates the center point of the head and shoulder region (…). , ) and the center point of the circumscribed rectangle of the overall human body outline ( , Offset vector of ) ;in: : A vector describing the direction of head and shoulder center offset; : Coordinates of the center of the head and shoulder region; The coordinates of the center of the circumscribed rectangle of the overall outline are given. Principal component analysis is then performed on the pixel cloud of the head and shoulder region to obtain the orientation angles of the first principal component. The pitch angle is calculated by combining the direction angle with the direction of the offset vector. : ;in: : The pitch angle of the wounded soldier; The head and shoulder principal component orientation angle is used to correct for body rotation factors. For bleeding area detection, this embodiment converts the color space to HSV within the contour area and sets the hue threshold range corresponding to the blood color. and saturation threshold The method for determining this is as follows: bloodstain samples are collected under different lighting conditions, and the results are obtained by expanding the maximum and minimum hue distribution ranges. The result is obtained by subtracting three standard deviations from the mean of the saturation distribution. To enhance robustness, region growing is performed on pixels that meet the hue and saturation conditions to obtain connected suspected bleed regions. The bleed area ratio is expressed as: ;in: : Percentage of bleeding area; Number of pixels in the bleeding area; Total number of pixels in the body contour region. The motion frequency index is obtained through the motion vectors of corresponding pixel blocks in consecutive frames. In this embodiment, the displacement is calculated for each pixel block: ;in: The displacement magnitude of the pixel block in the k-th frame; The pixel block is located at the center of the k-th frame; Previous frame position. The motion frequency index is obtained by calculating the dominant frequency component of the displacement sequence. Its physical meaning corresponds to the periodic movements of the injured person's limbs, such as struggling or slight twitching, providing a quantitative basis for subsequent assessment of the urgency of the injury.

[0034] Specifically, the process of obtaining a micro-brightness fluctuation curve by accumulating inter-frame gray-level differences and then using pitch attitude parameters to perform angle compensation on the fluctuation curve to obtain the compensated respiratory rhythm estimate is as follows: A set of pixels with stable texture features is selected within a defined head and shoulder region; the gray-level difference between corresponding pixels in adjacent frames is calculated; the gray-level difference is accumulated over time to form the original fluctuation curve; the influence of pitch attitude parameters on gray-level changes is analyzed, and an angle-gray-level change compensation function is established to compensate the original fluctuation curve using a sine function based on the pitch angle; periodic signal components are extracted from the compensated fluctuation curve, and the respiratory cycle is identified using a peak detection algorithm to calculate the number of breaths per unit time, thus obtaining the respiratory rhythm estimate.

[0035] In this implementation scheme, a slight brightness fluctuation curve is obtained by accumulating inter-frame gray-level differences, and angle compensation is performed in conjunction with pitch attitude parameters to obtain a breathing rhythm estimate. Specifically, a set of pixels Ω with stable texture features is selected within a defined head and shoulder region, and inter-frame gray-level difference calculation is performed on each pixel within set Ω, where the gray level of the k-th frame pixel is denoted as . , where p is the pixel index. The grayscale difference between adjacent frames is defined as... Then, spatial summation is performed on all pixels to obtain the temporal gray-level difference sequence. ;in, : Uncompensated temporal grayscale difference; p: Pixel index; k: Frame number. To eliminate brightness variation interference caused by UAV pitch angle, the pitch angle sequence obtained from the airborne attitude sensor is used. Construct an angle compensation function ;in, Angle sensitivity coefficient, used to reflect the degree of influence of pitch angle change on grayscale response; The pitch angle of the k-th frame is calculated by the attitude sensor. Parameters Based on static calibration experiments, the brightness variation data of a stationary target was collected at different pitch angles and obtained using least-squares fitting. The compensation amount was then superimposed onto the uncompensated gray-level difference sequence to obtain the angle-compensated brightness fluctuation sequence. ;in, The compensated fluctuation sequence. Subsequently, the sequence... Periodic component extraction is performed, and the positional intervals of adjacent peaks are identified using a peak detection method. Let the frame number of the adjacent peaks be denoted as . and Then a single respiratory cycle can be represented as A stable breathing cycle is obtained by averaging multiple cycles, and then the number of breaths per unit time is calculated by combining this with the video frame rate f. Where R: estimated respiratory rhythm; f: frame rate; : Cycle average. This process yields robust estimates of respiratory rhythms for subsequent vital sign fusion analysis.

[0036] Specifically, the process of fusing the estimated respiratory rhythm with the proportion of bleeding area and the exercise frequency index to construct a life-behavior joint parameter vector is as follows: The estimated respiratory rhythm is normalized and mapped to a standard physiological parameter range; the proportion of bleeding area is graded and quantified according to the severity of the injury to obtain a standardized bleeding score; the exercise frequency index is converted into an exercise capacity assessment value, and the exercise activity coefficient is calculated based on the exercise amplitude and frequency; the standardized respiratory rhythm, bleeding score, and exercise activity coefficient are combined in a predetermined order to form a life-behavior joint parameter vector containing three dimensions.

[0037] In this implementation plan, the estimated respiratory rhythm is fused with the hemorrhage area ratio and exercise frequency to construct a joint vital-behavioral parameter vector. Specifically, the estimated respiratory rhythm R is first subjected to interval mapping to normalize it into standard physiological parameters. Where L is the normalized respiratory parameter; , The minimum and maximum values ​​are obtained from the interval statistics of healthy and abnormal states in the training samples. These values ​​are determined through distribution analysis of historical labeled data to ensure that the normalized R maintains its physiologically meaningful linear discriminative power. Subsequently, the proportion of bleeding area, denoted as S, is calculated based on the injury region segmentation results. According to the clinical trauma severity model, S is converted into a discrete bleeding score. Where H: bleeding score; Indicator functions; , Two-level area thresholds were set based on medical experience; , : The scoring coefficient corresponding to the trauma level. The scoring coefficient is empirically calibrated by referring to actual combat trauma data, making different degrees of bleeding clearly distinguishable in the scoring. For the motion frequency index, let the detected motion change intensity sequence be . The motion frequency F is obtained by counting local maxima. F is then mapped to an activity coefficient reflecting athletic ability. Where A: the coefficient of motor activity; : Motion sensitivity adjustment coefficient; The coefficient is derived from the average value of normal movements in the training sample, referencing the motion frequency. The influence of activity assessment on the final vector was determined through cross-validation. Finally, the normalized respiratory parameter L, hemorrhage score H, and exercise activity coefficient A were combined in a fixed order to form a three-dimensional life-behavioral joint parameter vector. ;vector This is used to support subsequent casualty priority inference, enabling comparability and calculability of vital signs, bleeding status, and motor ability within the same parameter space.

[0038] Specifically, step S4 includes the following steps: establishing a weight matrix according to medical support training guidelines, assigning different weight coefficients to the three parameters of respiratory rhythm, bleeding rate, and movement frequency, performing matrix operation on the joint parameter vector and weight matrix to obtain the injury urgency score; setting a score threshold range, mapping the injury urgency score to the corresponding treatment priority level; associating the spatial location information of each injured person with the corresponding priority level, and visually displaying it on an electronic map using identifiers of different colors and shapes to form a spatial distribution map containing priority information.

[0039] In this embodiment, step S4 is used to determine the life-behavior joint parameter vector as described above. The system generates injury urgency scores and prioritizes treatment, enabling visualization of the spatial distribution of the wounded. The specific process is as follows: First, a weight matrix is ​​established based on medical support training principles. ;in: The respiratory rhythm weighting coefficient represents the relative importance of respiratory status in injury assessment and is determined through statistical analysis of historical training data and emergency experience. Bleeding proportion weighting coefficient, representing the impact of bleeding severity on emergency score, is obtained by fitting the statistical distribution of trauma type and bleeding volume; The exercise frequency weighting coefficient represents the impact of exercise capacity on the urgency of the injury, and is determined in combination with limb activity indicators and rehabilitation difficulty. This describes the method for constructing the diagonal matrix. It involves the joint parameter vector. With weight matrix Perform matrix synthesis to obtain the severity score of the injury. : ;in: : The current urgency level of the injured; Normalized respiratory rhythm value; Standardized bleeding score; The exercise activity coefficient is used to weight and fuse three key parameters with different weights, achieving a comprehensive quantification of multi-dimensional injury characteristics. The weighting coefficients are fitted using regression analysis of historical training scenario data to ensure the score reflects actual training needs, and can be dynamically adjusted according to different training objectives. Subsequently, a scoring threshold range is set. Score the urgency of the injury Mapped to the corresponding treatment priority level : ;in: Priority levels for treating the wounded; The scoring threshold is determined through statistical analysis of training cases and emergency simulation assessments to ensure a clear distinction in the urgency of casualties corresponding to different priorities. This step maps injury scores to actual treatment decisions, allowing scores generated under UAV-based overhead conditions to be directly used for training and drill scheduling. Finally, the spatial location information of each casualty is correlated with its corresponding priority level. The data is linked and visualized on an electronic map using identifiers of different colors and shapes, forming a spatial distribution map that includes priority information. The technical purpose of this step is to: intuitively display the distribution of wounded and the order of treatment within the training area, providing medical training commanders with a basis for rapid decision-making and achieving unified management of the wounded's status and location.

[0040] Please see Figure 2 A system for drone-based overhead-view casualty detection and priority recommendation for medical support training includes the following modules: a contour detection module, used to acquire overhead-view video streams of the training area, perform ground texture suppression and illumination difference processing on consecutive frames, construct a surface structure gradient field based on pixel density change rate, detect curvature aberrations in the human contour within the gradient field, perform temporal consistency checks on the curvature aberrations, and generate a contour confidence map; and a posture and bleeding recognition module, used to perform local illumination difference and skin color probability estimation on areas with confidence scores higher than a set threshold in the contour confidence map, determine the head and shoulder region range based on the relative surface temperature gradient, calculate the casualty pitch posture parameters based on the spatial deviation between the head and shoulder region and the overall contour, and simultaneously extract the color distribution and area ratio of bleeding areas within the confidence region. The system calculates motion vectors for limb pixels across consecutive frames to obtain motion frequency indices. A respiratory rhythm estimation module selects the set of pixels with the highest visual credibility based on head and shoulder region and pitch posture parameters. It obtains a micro-brightness fluctuation curve through inter-frame grayscale difference accumulation, compensates for the angle of the fluctuation curve using pitch posture parameters, and obtains a compensated respiratory rhythm estimate. This estimate is then fused with the bleeding area ratio and motion frequency indices to construct a life-behavior joint parameter vector. A priority assessment module inputs the respiratory rhythm, bleeding ratio, and motion frequency from the joint parameter vector into a preset weight matrix for weighted fusion calculation, generating an injury urgency score. Based on the score range, the injured are divided into different treatment priority levels, generating a spatial distribution map containing priority identifiers.

[0041] In this implementation scheme, the contour detection module performs preliminary processing on the overhead video stream of the training area. It constructs a surface structure gradient field using ground texture suppression and illumination difference methods, detects curvature anomalies in the human contour within this gradient field, performs temporal consistency checks, and generates a contour confidence map. The technical role of this module is to provide stable and reliable human contour information, providing foundational data for subsequent pose analysis and bleeding recognition, significantly reducing detection drift caused by complex backgrounds or illumination changes. The pose and bleeding recognition module filters high-confidence regions from the contour confidence map, performs local illumination compensation and skin color probability estimation, and determines the head and shoulder region range by combining surface temperature gradients. The module further calculates the patient's pitch and posture parameters, extracts the color distribution and area ratio of bleeding regions within the confidence area, and statistically analyzes the motion vectors and motion frequency indices of limb pixels. The technical role of this module is to accurately extract the patient's pose and injury features, providing precise input for respiratory rhythm estimation and comprehensive scoring. The respiratory rhythm estimation module, based on head and shoulder region and pitch posture parameters, selects the pixel set with the highest visual credibility. It generates a micro-brightness fluctuation curve through inter-frame grayscale difference accumulation and performs angle compensation to obtain the estimated respiratory rhythm value. This module fuses the respiratory rhythm with the bleeding ratio and motion frequency to construct a vital-behavioral joint parameter vector. The technical function of this module is to achieve quantitative assessment of the vital signs and motion status of the injured through multi-feature fusion, improving the accuracy of injury judgment. The priority assessment module inputs the various indicators in the vital-behavioral joint parameter vector into a preset weight matrix for weighted fusion calculation, generating an injury urgency score and classifying the injured person's treatment priority level according to the score range. This module associates the spatial location information of the injured person with the priority level, forming a visualized spatial distribution map. The technical function of this module is to provide an intuitive and quantitative distribution of injured person priority, assisting in rescue decision-making and resource scheduling in training scenarios.

[0042] This application also discloses a computer program product or computer program stored in a computer-readable storage medium. A processor of a computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, causing the computer device to perform the methods or systems described above. Similarly, the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0043] In summary, this application has at least the following effects:

[0044] A method, system, and storage medium for drone-based overhead casualty detection and priority recommendation for medical support training are presented. Through the synergistic effect of modules such as continuous frame processing of drone-based overhead video, contour detection, attitude and bleeding recognition, respiratory rhythm estimation, and priority assessment, the system achieves comprehensive analysis and quantitative assessment of casualty vital signs, injury severity, and movement status. It can generate a spatial distribution map containing priority information, providing an intuitive and reliable decision-making basis for rescue process planning and resource scheduling in training scenarios.

[0045] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0046] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0047] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0048] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0049] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0050] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for detecting and prioritizing wounded personnel from an aerial view by using a UAV for medical service training, characterized in that, The method comprises the following steps: S1. Obtain an overhead video stream of a training area, perform ground texture suppression and illumination difference processing on consecutive frames, construct a ground surface structure gradient field according to a pixel density change rate, detect a curvature abnormal section of a human body contour in the gradient field, and perform time sequence consistency checking on the curvature abnormal section to generate a contour confidence map; S2. According to the area with a confidence higher than a set threshold in the contour confidence map, perform local illumination difference and skin color probability estimation, determine the head and shoulder area range in combination with the relative gradient of the surface temperature, calculate the casualty pitch parameter according to the spatial deviation of the head and shoulder area and the overall contour, extract the color distribution and area proportion of the bleeding area in the confidence area, and calculate the motion vector of the limb pixels between consecutive frames to obtain a motion frequency indicator; S3. Based on the head and shoulder area and the pitch parameter, select the highest visible and reliable pixel set, obtain a slight brightness fluctuation curve through the accumulation of frame-to-frame gray difference, perform angle compensation on the fluctuation curve using the pitch parameter to obtain a compensated respiration rhythm estimation value, and fuse the respiration rhythm estimation value with the bleeding area proportion and the motion frequency indicator to construct a life-behavior joint parameter vector; S4. Input the respiration rhythm, bleeding proportion and motion frequency in the joint parameter vector into a preset weight matrix for weighted fusion calculation to generate a casualty emergency degree score, divide the casualty into different treatment priority levels according to the score interval, and generate a spatial distribution map containing a priority identifier.

2. The method for detecting and prioritizing wounded personnel from an aerial view by using a UAV for medical service training according to claim 1, wherein: The specific process of obtaining an overhead video stream of a training area and performing ground texture suppression and illumination difference processing on consecutive frames is as follows: Multi-scale Gaussian smoothing is performed on consecutive frames in the overhead video stream, the standard deviation of pixel values in a local window is calculated to identify texture complex regions, texture feature matching based on local binary pattern is performed on the texture complex regions, and the matching successful texture regions are marked as ground background; Pixel-by-pixel difference operation is performed on the current frame and the background reference frame to obtain a difference image without ground texture, the pixel density change rate is calculated in a sliding window manner on the difference image, the first derivative of the pixel density is calculated in the horizontal and vertical directions respectively, and the derivative results in the two directions are vector synthesized to form a gradient field reflecting the ground surface fluctuation change.

3. The method for detecting and prioritizing wounded personnel from an aerial view by using a UAV for medical service training according to claim 2, characterized in that: The specific process of detecting a curvature abnormal section of a human body contour in the gradient field and performing time sequence consistency checking on the curvature abnormal section to generate a contour confidence map is as follows: The continuous contour lines are extracted in the gradient field by an edge tracking algorithm, the curvature value of each sampling point on the contour line is calculated, a curvature change monitoring window is set, when the standard deviation of the curvature value in the window exceeds a set threshold, the section of the contour is marked as an abnormal section; In the consecutive frames of the video sequence, a motion trajectory record is established for each abnormal section, the spatial position change amount of the abnormal sections between adjacent frames is calculated, the confidence of the abnormal sections with a position change amount less than a displacement threshold is accumulated, and the confidence of the abnormal sections with severe position jitter is attenuated; A confidence map representing contour stability is generated according to the confidence distribution.

4. The method for detecting and prioritizing wounded personnel from an aerial view using a UAV for medical service training according to claim 1, wherein: The specific process of determining the head-shoulder region range according to the region with a confidence higher than a set threshold in the contour confidence map, combining local illumination difference and skin color probability estimation, and determining the relative gradient of the surface temperature is as follows: Local illumination compensation is performed in the region with a confidence up to standard, local contrast is adjusted by calculating the brightness mean value of the neighborhood pixels, skin color features are extracted in the brightness balanced image, and the distribution density of Cr and Cb components in the YCbCr color space is counted; The temperature distribution in the infrared image is synchronously analyzed, the temperature difference between each pixel and its eight neighborhood is calculated, the corresponding relationship between the skin color density and the temperature gradient is established, the region satisfying the high skin color density and stable temperature distribution is found, and the minimum circumscribed rectangle range of the head-shoulder region is determined according to the size ratio constraint of the human head-shoulder.

5. The method for detecting and prioritizing wounded personnel from an aerial view by using a UAV for medical service training according to claim 4, wherein: The specific process of calculating the patient's pitch attitude parameter according to the spatial deviation of the head-shoulder region and the overall contour, extracting the color distribution and area ratio of the bleeding region in the confidence region, and calculating the motion vector of the limb pixels between continuous frames to obtain the motion frequency index is as follows: The offset vector of the head-shoulder region center relative to the center of the circumscribed rectangle of the human contour is calculated, the direction angle of the head-shoulder region is obtained by principal component analysis, and the pitch angle is calculated in combination with the modulus and direction of the offset vector; The color space is converted to HSV in the human contour region, the hue range threshold and the saturation threshold of the blood color feature are set, the connected bleeding region is extracted by region growing algorithm, and the ratio of the number of bleeding region pixels to the total number of human contour pixels is counted; The motion displacement of the corresponding pixel block between continuous frames is calculated, the motion vector is obtained by analyzing the intensity change of the pixel block in the time and space domain, the amplitude distribution and change frequency of the motion vector are counted, and the amplitude and frequency index of the main motion feature are extracted.

6. The method for detecting and prioritizing wounded personnel from an aerial view using a UAV for medical service training according to claim 1, wherein: The specific process of obtaining the micro-amplitude brightness fluctuation curve by accumulating the gray difference between frames, and compensating the fluctuation curve by the pitch attitude parameter to obtain the compensated respiratory rhythm estimation value is as follows: Select the pixel set with stable texture features in the determined head-shoulder region, calculate the gray difference of the corresponding pixels between adjacent frames, and form the original fluctuation curve by time series accumulation of the gray difference; The influence law of the pitch attitude parameter on the gray change is analyzed, the compensation function of angle-gray change is established, and the original fluctuation curve is compensated based on the sine function of the pitch angle; Periodic signal components are extracted in the compensated fluctuation curve, the respiratory cycle is identified by peak detection algorithm, the number of breaths per unit time is calculated, and the respiratory rhythm estimation value is obtained.

7. The method for detecting and prioritizing wounded personnel from an aerial view using a UAV for medical service training according to claim 1, wherein: The specific process of fusing the respiratory rhythm estimation value with the bleeding area ratio and the motion frequency index to construct the life-behavior joint parameter vector is as follows: The respiratory rhythm estimation value is normalized to map to the standard physiological parameter range, and the bleeding area ratio is quantitatively graded according to the severity of trauma to obtain a standardized bleeding score; The motion frequency index is converted into a motion ability evaluation value, and the motion activity coefficient is calculated according to the motion amplitude and frequency; The standardized respiratory rhythm, bleeding score and motion activity coefficient are combined in a predetermined order to form a life-behavior joint parameter vector with three dimensions.

8. The unmanned aerial vehicle overhead wounded detection and priority recommendation method for medical service training according to claim 1, characterized in that: Step S4 includes the following steps: According to the medical service training criterion, a weight matrix is established, different weight coefficients are respectively given to the three parameters of breathing rhythm, bleeding proportion and movement frequency, matrix comprehensive operation is performed on the joint parameter vector and the weight matrix, and a severity of injury emergency score is obtained; A score threshold interval is set, and the severity of injury emergency score is mapped to a corresponding treatment priority level; The spatial position information of each wounded person is associated with the corresponding priority level, and different colors and shapes of identifiers are used for visual display on the electronic map to form a spatial distribution map containing priority information.

9. A system for detecting and prioritizing wounded personnel from an aerial view for medical service training, applied to the method for detecting and prioritizing wounded personnel from an aerial view for medical service training according to any one of claims 1-8, characterized in that, The method comprises the following modules: A contour detection module is configured to acquire an overhead video stream of a training area, perform ground texture suppression and illumination difference processing on consecutive frames, construct a ground structure gradient field according to a pixel density change rate, detect a curvature abnormal section of a human body contour in the gradient field, and perform time sequence consistency checking on the curvature abnormal section to generate a contour confidence map; An attitude and bleeding identification module is configured to perform local illumination difference and skin color probability estimation according to a region with a confidence higher than a set threshold in the contour confidence map, determine a head-shoulder region range in combination with a relative gradient of surface temperature, calculate a wounded person's pitch attitude parameter according to a spatial deviation of the head-shoulder region and the overall contour, extract a color distribution and an area proportion of a bleeding region in the confidence region, and calculate a motion vector of limb pixels between consecutive frames to obtain a movement frequency indicator; A breathing rhythm estimation module is configured to select a pixel set with the highest visibility and credibility based on the head-shoulder region and the pitch attitude parameter, obtain a micro-brightness fluctuation curve through frame-by-frame gray difference accumulation, perform angle compensation on the fluctuation curve by using the pitch attitude parameter to obtain a compensated breathing rhythm estimation value, and fuse the breathing rhythm estimation value with the bleeding area proportion and the movement frequency indicator to construct a life-behavior joint parameter vector; A priority evaluation module is configured to input the breathing rhythm, the bleeding proportion and the movement frequency in the joint parameter vector into a preset weight matrix for weighted fusion calculation to generate a severity of injury emergency score, divide the wounded person into different treatment priority levels according to a score interval, and generate a spatial distribution map containing priority identifiers.

10. A computer readable storage medium having stored therein a program that is executable by a processor, characterized in that, The program executable by the processor, when executed by the processor, is used to perform the method of any one of claims 1-8.

Images