Rescue method and system of rescue robot for exploration

Abstract

The invention discloses a rescue method and system of a rescue robot for exploration, and relates to the technical field of underground space rescue, and the method comprises the following steps: obtaining multi-path acoustic echo data of a karst cave and motion track data of the robot, and constructing a three-dimensional point cloud model according to the multi-path acoustic echo data and the motion track data of the robot; and identifying unmatched data in the multi-path acoustic echo data and the robot motion trail data, and taking an area where the unmatched data is located as an abnormal area. According to the method, the spatial resolution of signal acquisition is enhanced through the multi-microphone array, robust acoustic fingerprints are extracted in combination with a noise reduction algorithm and short-time Fourier transform, high-confidence human body sound source signals are screened out by using a feature template matching mechanism, accurate extraction and recognition of human body acoustic features in a karst cave complex noise environment are realized, and the accuracy of human body acoustic feature recognition is improved. The problem of misjudgment caused by confusion of sound source features and environmental noise in a traditional method is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of underground space rescue, and in particular to a rescue method and system for an exploration rescue robot. Background Art

[0002] With the continuous deepening of underground space development and geological exploration activities, the need to rescue missing persons in complex environments such as underground caves and mines has become increasingly prominent; traditional manual rescue methods often have many problems such as slow response speed, high personnel safety risks, and low positioning accuracy due to the complex spatial environment, low visibility, and unknown geological structure; in recent years, with the development of robotics, environmental perception, acoustic detection, and three-dimensional modeling technologies, research on search and rescue of missing persons based on rescue robots has become a hot topic in related fields.

[0003] However, in the process of implementing the technical solutions of the invention in the embodiments of this application, it was found that the above technology has at least the following technical problems:

[0004] Existing methods often fail to effectively combine the three-dimensional model of the environment with acoustic characteristics, resulting in inaccurate calculation results of the sound source location and a lack of correlation analysis between the sound source location and the abnormal environmental area. This leads to insufficient safety of the generated rescue path and an inability to assist rescue robots in making optimal decisions, which restricts the efficiency and success rate of rescue operations in cave environments. Summary of the Invention

[0005] The purpose of the present invention is to provide a rescue method and system for an exploration rescue robot to solve the problems raised in the above background technology.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention discloses a rescue method for an exploration rescue robot, which is applied to the rescue of lost persons in cave exploration, comprising the following steps:

[0008] Acquire multipath acoustic echo data of the cave and robot motion trajectory data, and use them to construct a three-dimensional point cloud model;

[0009] Identifying mismatched data in the multipath acoustic echo data and the robot motion trajectory data, and using an area where the mismatched data is located as an abnormal area;

[0010] Acquiring an audio signal from the cave, and extracting acoustic fingerprint features from the audio signal;

[0011] Calculating similarity between the acoustic fingerprint feature and a preset feature template, and determining a target audio signal based on the similarity calculation result;

[0012] Calculating the audio sound coordinates based on the target audio signal, determining whether the audio sound coordinates are located in an abnormal area, and if so, correcting the audio sound coordinates based on multipath acoustic echo data of the abnormal area, and calculating the overlap between the corrected audio sound coordinates and the abnormal area;

[0013] Determining whether the overlap is greater than a preset threshold, if so, determining the corrected audio coordinates as the coordinates of the potential missing person, and performing nonlinear calculation on the overlap and similarity calculation results to obtain a spatial confidence level;

[0014] Determining whether the spatial confidence is greater than a preset value, and if so, determining the corrected audio sound coordinates as the coordinates of the missing person;

[0015] A rescue path for the missing person is generated based on the missing person's coordinates and the robot's motion trajectory data.

[0016] In a second aspect, the present invention discloses a rescue system for an exploration rescue robot, comprising:

[0017] The 3D point cloud model construction module is used to obtain the multi-path acoustic echo data of the cave and the robot motion trajectory data, and use them to construct a 3D point cloud model;

[0018] an abnormal region identification module, configured to identify mismatched data between the multipath acoustic echo data and the robot motion trajectory data, and to use the region where the mismatched data is located as an abnormal region;

[0019] An audio signal acquisition module, used to acquire audio signals in the cave;

[0020] an audio signal processing module, configured to extract acoustic fingerprint features from the audio signal, perform similarity calculation between the acoustic fingerprint features and a preset feature template, and determine a target audio signal based on the similarity calculation result;

[0021] an audio sound coordinate determination module, configured to calculate the audio sound coordinates based on the target audio signal, determine whether the audio sound coordinates are located in an abnormal area, and if so, correct the audio sound coordinates based on multipath acoustic echo data of the abnormal area;

[0022] A missing person coordinate determination module is used to calculate the overlap between the corrected audio coordinates and the abnormal area;

[0023] Determining whether the overlap is greater than a preset threshold, if so, determining the corrected audio coordinates as the coordinates of the potential missing person, and performing nonlinear calculation on the overlap and similarity calculation results to obtain a spatial confidence level;

[0024] Determining whether the spatial confidence is greater than a preset value, and if so, determining the corrected audio sound coordinates as the coordinates of the missing person;

[0025] The rescue path generation module is used to generate a rescue path for the missing person based on the coordinates of the missing person and the robot motion trajectory data.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] 1. This application enhances the spatial resolution of signal acquisition through a multi-microphone array, combines noise reduction algorithm with short-time Fourier transform to extract robust acoustic fingerprints, and uses feature template matching mechanism to screen out high-confidence human sound source signals, thereby realizing the accurate extraction and recognition of human acoustic features in the complex noise environment of the cave, and effectively solving the misjudgment problem caused by the confusion between sound source features and environmental noise in traditional methods.

[0028] 2. This application innovatively combines system error compensation based on environmental structure modeling with dynamic coordinate correction based on signal feature learning. Through the dual correction mechanism, it effectively overcomes the limitations of a single correction method under complex geological conditions, effectively solves the problem of decreased positioning accuracy caused by multi-path propagation of sound waves in abnormal cave areas, and provides a reliable spatial coordinate data basis for rescue path planning.

[0029] 3. Through dynamic signal monitoring and three-dimensional spatial mapping, this solution establishes a real-time correlation mechanism between path status and environmental changes, realizes dynamic safety control of rescue paths in complex cave environments, and effectively avoids robot jams or crashes caused by sudden changes in geological structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0031] Figure 1 is a flow chart of the steps of the present invention;

[0032] Figure 2 A schematic diagram of the process of determining the audio sound coordinates provided by the present invention;

[0033] Figure 3 A schematic diagram of the process of correcting audio utterance coordinates provided by the present invention;

[0034] Figure 4 A schematic diagram of the process of determining the rescue priority of a missing person provided by the present invention;

[0035] Figure 5 A schematic diagram of a scenario provided by the present invention;

[0036] Figure 6 This is a functional diagram of the system modules provided by the present invention. DETAILED DESCRIPTION

[0037] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0038] Application Overview:

[0039] In existing technologies, with the continuous deepening of underground space development and geological exploration activities, the rescue of missing persons in complex environments faces severe challenges; traditional rescue methods rely on manual operations, and there are problems of response delay and positioning deviation in low-visibility environments such as caves; existing technologies mostly use single sensor data, and sound source positioning is easily interfered by multipath effects, and the positioning results are not associated with environmental structural risks, resulting in safety hazards in rescue path planning.

[0040] In order to solve the above problems, in view of the phenomenon of distortion of sound wave propagation in complex geological structures, it was found that relying solely on acoustic signals for positioning will lead to coordinate offset; when further studying the structural characteristics of the cave, it was realized that three-dimensional environmental modeling can provide spatial constraints; after multiple experimental verifications, it was found that the fusion modeling of multi-path acoustic echo data and motion trajectory data can effectively restore the real spatial structure; when the coordinates of the sound source are detected in an abnormal area, it is inferred that the conventional positioning algorithm has systematic errors, and then an adaptive correction mechanism based on environmental characteristics is proposed.

[0041] After introducing the basic concept of the present invention, embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0042] Example 1:

[0043] See also Figure 1-Figure 5 A rescue method for an exploration rescue robot is applied to rescue persons lost in cave exploration, comprising the following steps:

[0044] Acquire multipath acoustic echo data of the cave and robot motion trajectory data, and use them to construct a three-dimensional point cloud model;

[0045] Identify mismatched data between multipath acoustic echo data and robot motion trajectory data, and use the area where the mismatched data is located as the abnormal area;

[0046] Acquire audio signals from the cave and extract acoustic fingerprint features from the audio signals;

[0047] Calculate the similarity between the acoustic fingerprint feature and the preset feature template, and determine the target audio signal based on the similarity calculation result;

[0048] Calculate the audio source coordinates based on the target audio signal, determine whether the audio source coordinates are located in an abnormal area, and if so, correct the audio source coordinates based on the multipath acoustic echo data of the abnormal area, and calculate the overlap between the corrected audio source coordinates and the abnormal area;

[0049] Determine whether the overlap is greater than a preset threshold. If so, determine the corrected audio coordinates as the coordinates of the potential missing person, and use the overlap and similarity calculation results to obtain the spatial confidence through nonlinear calculation. The specific calculation formula is as follows:

[0050]

[0051] Where, represents the spatial confidence, Indicates the degree of overlap, Indicates similarity, represents the adaptive weight coefficient, represents the nonlinear enhancement coefficient, represents the base of natural logarithm;

[0052] Determine whether the spatial confidence is greater than a preset value, and if so, determine the corrected audio coordinates as the coordinates of the missing person;

[0053] Generate a rescue path for the missing person based on the missing person's coordinates and the robot's motion trajectory data.

[0054] The 3D point cloud model refers to a set of spatial points generated by combining the propagation delay of sound waves with the spatial coordinate data of the robot. For example, it can be implemented by combining the ToF ranging algorithm with SLAM technology to accurately characterize the internal structure of the cave.

[0055] Abnormal areas refer to spatial areas with abnormal point cloud distribution, which can be identified, for example, by density mutation detection algorithms and used to warn of geological structural risks;

[0056] Acoustic fingerprint features refer to the spectral feature vectors of audio signals, which can be extracted through MFCC coefficients, for example, to distinguish human voice sources from environmental noise;

[0057] Overlap refers to the spatial coincidence ratio between the corrected coordinates and the abnormal area, which can be obtained, for example, by voxelized grid calculation and used to evaluate the credibility of the coordinates;

[0058] Spatial confidence refers to a composite indicator of comprehensive acoustic matching and spatial overlap, which can be calculated, for example, using a weighted product method for final positioning decision-making.

[0059] Specifically, the system first collects multipath echo data through the sonar device carried by the robot, combines it with the IMU sensor to record motion trajectory data, and uses the point cloud reconstruction algorithm to generate a three-dimensional environment model;

[0060] When the microphone array captures suspicious audio, it extracts its Mel-frequency cepstral coefficients as feature templates for matching, and filters out signals that match the human voiceprint characteristics;

[0061] After initially calculating the sound source coordinates based on the arrival time difference algorithm, it automatically detects whether the coordinates are located in the cavity, crack, or landslide sign area marked in the point cloud model. If there is overlap, the sound wave propagation data of the area is called up and the coordinate offset is corrected through the BP neural network.

[0062] The corrected coordinates must meet the overlap threshold with the abnormal area and participate in the spatial confidence calculation together with the acoustic matching degree. Ultimately, only coordinates with spatial confidence greater than the preset value are output for path planning.

[0063] Compared with existing technologies, existing solutions usually handle environmental modeling and sound source localization independently, while this solution establishes a data linkage mechanism; for example, traditional methods ignore the impact of environmental structure on sound wave propagation when calculating sound source coordinates, while this solution actively corrects positioning errors by identifying abnormal areas; existing technologies use fixed thresholds to judge the effectiveness of sound sources, while this solution achieves dynamic decision-making through nonlinear fusion of acoustic and spatial features; in addition, existing path planning is mostly based on straight-line distance, while this solution combines robot motion trajectory data to avoid risky areas.

[0064] This application solves the problem of large sound source positioning errors in complex geological environments, and improves coordinate accuracy by correcting environmental characteristics; at the same time, it establishes a correlation mechanism between rescue paths and geological risks to prevent robots from entering landslide-prone areas; the double verification mechanism effectively reduces the false alarm rate and ensures the safety and reliability of rescue operations.

[0065] This application further proposes that constructing a 3D point cloud model includes:

[0066] Collect multi-path acoustic echo data and robot motion trajectory data reflected from different interfaces and structures in the cave;

[0067] Multipath acoustic echo data includes sound wave emission time, sound wave reception time, sound wave propagation path, and sound speed;

[0068] The robot motion trajectory data includes the robot's spatial coordinates;

[0069] The sound wave propagation delay is calculated by subtracting the sound wave transmission time from the sound wave reception time. Combined with the speed of sound, the actual spatial distance traveled by the acoustic echo is calculated to obtain the echo path length data.

[0070] The echo path is positioned in three-dimensional space according to the echo path length data and the spatial coordinates of the robot;

[0071] According to the sound wave propagation path and echo path, a spatial point set is generated, which is gradually accumulated to form dense three-dimensional point cloud data, and then a three-dimensional point cloud model of the cave is generated.

[0072] Among them, multipath acoustic echo data refers to the set of echo signals formed by multiple reflections of sound waves on the interfaces of different media in the cave. Specifically, it can be achieved by using sonar equipment to emit pulse sound waves and receive reflected signals to capture the reflection characteristics of structures such as cave walls and cracks.

[0073] Among them, the robot motion trajectory data refers to the spatial coordinate sequence recorded by the rescue robot during its movement in the cave. It can be implemented by an inertial navigation system or a lidar positioning device to provide a dynamic spatial reference for the acoustic echo data.

[0074] Among them, the sound wave propagation delay refers to the time difference between the emission and reception of the sound wave. It can be measured through high-precision clock synchronization technology and used to calculate the actual propagation distance of the sound wave in the medium.

[0075] The echo path length data refers to the spatial distance set of the path traversed by the sound wave from emission to reception. It can be calculated using the product formula of the speed of sound and the propagation delay, and is used to determine the relative position of the reflection point.

[0076] Specifically, the sonar device transmits sound wave signals in different directions of the cave, receives multipath echo data reflected from rock walls, cavities and other structures, and records the real-time coordinate information of the robot during movement;

[0077] The time difference between sound wave emission and reception is accurately measured using time synchronization technology, and the actual length of each echo path is calculated based on the sound velocity distribution parameters in the cave.

[0078] Based on the robot's dynamic coordinates, the starting point and reflection point of each echo path are mapped to a three-dimensional coordinate system to form discrete spatial positioning points. By solving the spatial geometric relationship between the sound wave propagation path and the echo path, a spatial point set containing the cave structure characteristics is generated.

[0079] As the robot moves and the acoustic waves continue to scan, the point set is incrementally registered and density optimized, ultimately constructing a dense three-dimensional point cloud model reflecting the true shape of the cave.

[0080] Compared with existing technologies, traditional methods usually only use single sensor data or static coordinates for modeling, which makes it difficult to accurately reflect the multipath reflection characteristics of complex caves. This solution fuses multipath acoustic echoes with robot dynamic motion data to dual-locate the reflection points in three-dimensional space, effectively solving the coupling error problem between the sound wave propagation path and the spatial coordinates. For example, the existing technology does not consider the coordinate reference drift caused by robot movement, while this solution ensures the spatial consistency of point cloud data by updating the relative relationship between the robot coordinates and the sound wave emission position in real time.

[0081] Through the above technical solution, this application realizes the dynamic matching and fusion of multi-source perception data, improving the accuracy and reliability of cave three-dimensional modeling; through the collaborative solution of sound wave propagation delay and robot motion trajectory, it can effectively eliminate the path positioning deviation caused by a single acoustic data source, and provide a high-precision environmental model foundation for subsequent positioning and path planning of missing persons; combined with the geometric relationship between the sound wave propagation path and the robot's spatial coordinates, it can accurately restore the spatial distribution characteristics of complex structures such as cavities and cracks inside the cave, and solve the problem of incomplete environmental perception caused by the single data source of traditional modeling methods.

[0082] This application further proposes to identify mismatched data in multipath acoustic echo data and robot motion trajectory data, and to use the area where the mismatched data is located as the abnormal area, specifically including:

[0083] Analyze the spatial distribution of multipath acoustic echo data and motion trajectory data in the 3D point cloud model, identify mismatched data in the multipath acoustic echo data and motion trajectory data, and use the area where the mismatched data is located as the abnormal area;

[0084] Perform spatial density analysis, cluster segmentation, and structural change detection on the 3D point cloud data near the abnormal area to identify the type of abnormal area, including cavity anomalies, crack anomalies, and landslide precursors;

[0085] Mark the spatial coordinate range of the abnormal area;

[0086] For each abnormal area that is identified and marked, count the coordinates of all point clouds in the corresponding area;

[0087] Multipath acoustic echo data associated with these point cloud coordinates is retrieved.

[0088] Spatial density analysis refers to the statistical distribution density of point clouds within a unit volume, for example, using three-dimensional grid division and density histogram calculation to identify areas with sudden changes in point cloud density.

[0089] Cluster segmentation refers to grouping based on the spatial location of the point cloud. For example, the Euclidean distance clustering algorithm is used to divide the continuously distributed point cloud into independent areas.

[0090] Structural change detection refers to analyzing the geometric differences between adjacent point clouds, such as through surface curvature calculation or normal vector change detection, to identify areas that may indicate landslides.

[0091] Specifically, the multi-path acoustic echo data and motion trajectory data in the 3D point cloud model are spatially registered to form a related dataset. When the deviation between the acoustic echo path estimation result and the spatial coordinates of the robot's motion trajectory exceeds the allowable error range, the area is marked as an abnormal candidate area.

[0092] The density of the three-dimensional point cloud is calculated for the candidate area, for example, by counting the number of point clouds in a cubic grid. If the density value of a certain area is lower than the threshold of the adjacent area, it is judged as a cavity anomaly. For sparse point clouds with a strip distribution, a density-based clustering algorithm is used for segmentation. If the segmented point cloud clusters show linear extension characteristics, they are classified as crack anomalies. Structural change detection is carried out by comparing the height distribution differences between the current point cloud and the historical point cloud, for example, using the elevation gradient analysis method. If a sudden change in the height value of a local area exceeds the safety threshold, it is judged as a sign of landslide.

[0093] The spatial coordinate boundaries of all abnormal areas are marked with the minimum circumscribed cube, and a point cloud coordinate index table is established. The associated sound wave emission time, reception time and path length data in the area are retrieved through database query.

[0094] Compared with existing technologies, traditional methods only rely on a single data source for anomaly detection, such as analyzing point cloud density or acoustic echo path separately, without establishing spatial correlation between multimodal data, resulting in isolated noise points being misjudged as structural anomalies; this scheme effectively eliminates false abnormal signals caused by equipment errors or environmental interference through spatial correspondence verification of acoustic echo path and motion trajectory; existing technologies classify abnormal areas only based on morphological features, for example, uniformly treating a decrease in point cloud density as a cavity, without considering geological structure differences; this scheme introduces a structural change detection algorithm to distinguish between cavities formed by dissolution and collapse signs caused by changes in rock stress, providing refined data support for rescue path risk assessment.

[0095] Through the above technical solution, the present application solves the problems of low accuracy in abnormal area identification and rough classification in the existing technology; through spatial correlation analysis of multi-path acoustic data and motion trajectories, it avoids misjudgment caused by errors in single sensor data and improves the accuracy of abnormal area positioning; combined with multi-level feature extraction of density analysis, cluster segmentation and structural change detection, it realizes the refined classification of different types of abnormal areas such as cavities, cracks, and landslide signs, providing reliable structural risk data for subsequent rescue path safety assessment; the abnormal area coordinate marking and associated data retrieval mechanism ensures the data traceability during the sound source coordinate correction calculation, forming a complete data processing chain from environmental modeling to rescue decision-making.

[0096] This application further proposes to calculate the similarity between the acoustic fingerprint feature and the preset feature template, and determine the target audio signal according to the similarity calculation result, specifically including:

[0097] The microphone array carried by the robot collects audio signals in the cave in real time;

[0098] Filter, reduce noise and normalize audio signals;

[0099] The short-time Fourier transform method is used to extract the acoustic fingerprint features in the audio signal;

[0100] Calculate the similarity between the acoustic fingerprint feature and the preset feature template using the feature distance measurement method;

[0101] It is determined whether the similarity calculation result is greater than a set value, and if so, the audio signal is determined to be the target audio signal.

[0102] Among them, the microphone array refers to a collection system composed of multiple spatially distributed microphones, which can be implemented using a four-element ring array or a spherical array to collect sound source information covering different directions in the cave.

[0103] Among them, filtering processing refers to filtering out high-frequency mechanical noise and low-frequency geological vibration noise through a bandpass filter. Specifically, it can be implemented using a Butterworth filter to eliminate environmental interference signals.

[0104] Among them, noise reduction processing refers to eliminating multipath reflection noise through an adaptive noise suppression algorithm, which can be implemented by spectral subtraction or deep noise reduction network to improve the signal-to-noise ratio.

[0105] Among them, normalization processing refers to the dynamic range compression of the signal amplitude, which can be achieved by using the maximum amplitude normalization method to eliminate device acquisition differences.

[0106] Among them, the short-time Fourier transform method refers to dividing the time domain signal into a windowed frame sequence and then performing spectrum analysis. Specifically, it can be implemented using a Hamming window function to extract the time-frequency domain characteristics of the signal.

[0107] Among them, the feature distance measurement method refers to calculating the similarity between feature vectors, which can be implemented using the dynamic time warping algorithm or the cosine similarity algorithm to quantify the matching degree between the target signal and the preset template.

[0108] The set value refers to a similarity screening threshold determined in advance through experiments, which can be specifically set to a normalized value in the range of 0.7 to 0.9 to exclude low-confidence interference signals.

[0109] Specifically, the method first collects acoustic signals from multiple angles inside the cave through a spatially distributed microphone array, and uses a bandpass filter and adaptive noise reduction algorithm to eliminate environmental noise to ensure the signal quality of subsequent processing;

[0110] The normalized audio signal is segmented into short-time frame sequences, and the acoustic fingerprint including the spectrum envelope and harmonic structure is extracted through short-time Fourier transform;

[0111] The extracted fingerprint features are compared with the pre-established acoustic template library of missing persons for similarity, and the dynamic time warping algorithm is used to calculate the time alignment similarity of the time-frequency features;

[0112] When the calculated result exceeds the preset threshold, it is determined that the audio signal comes from the human body rather than environmental noise; this multi-stage processing mechanism effectively solves the problem of difficult feature extraction caused by severe sound wave reverberation and complex noise spectrum in the cave environment.

[0113] Compared with the existing technology, the traditional method only collects signals through a single microphone and does not establish an acoustic feature template library, resulting in the inability to distinguish between human voice sources and environmental noise; this method enhances the spatial resolution of signal acquisition through a multi-microphone array, combines noise reduction algorithm with short-time Fourier transform to extract robust acoustic fingerprints, and uses the feature template matching mechanism to screen out high-confidence human sound source signals; the existing technology does not consider the spectral distortion problem caused by the multipath effect of the cave, while this method uses a dynamic time warping algorithm to achieve elastic matching of time-frequency features, significantly improving the recognition accuracy in complex acoustic environments.

[0114] Through the above technical solution, this application realizes the accurate extraction and identification of human acoustic characteristics in the complex noise environment of the cave, effectively solving the misjudgment problem caused by the confusion between sound source characteristics and environmental noise in traditional methods; by constructing an acoustic fingerprint template library and a dynamic similarity calculation mechanism, it can reliably distinguish the sound source of the missing person from geological noise, providing high-quality input data for subsequent sound source positioning, thereby ensuring the safety of rescue path generation.

[0115] The present application further proposes that the audio utterance coordinates can be calculated based on the target audio signal, including:

[0116] Calculate the arrival time difference of the target audio signal between different microphones respectively;

[0117] Obtain the precise coordinates of each microphone in the 3D point cloud model;

[0118] According to the arrival time difference of the target audio signal between different microphones, the microphone coordinates and the speed of sound, a multivariate nonlinear equation system is established to solve the three-dimensional spatial coordinates of the sound source;

[0119] The calculated three-dimensional spatial coordinates of the sound source are mapped to the three-dimensional point cloud model and the sound source coordinate results are output.

[0120] Calculating the arrival time difference refers to measuring the time difference between the target audio signal arriving at multiple microphones. This can be achieved using a cross-correlation algorithm to eliminate multipath interference caused by sound wave reflections in a cave environment.

[0121] Among them, the precise coordinates of the microphone in the three-dimensional point cloud model refer to the position data obtained through laser ranging positioning or visual synchronous positioning and mapping technology to ensure the positioning accuracy of the spatial reference network.

[0122] Among them, the multivariate nonlinear equations refer to a mathematical model established based on the physical relationship between sound speed, time difference and spatial distance. By introducing multi-path sound wave propagation constraints, the errors of the traditional hyperbolic positioning method are corrected.

[0123] Among them, mapping to a three-dimensional point cloud model means spatially aligning the calculated sound source coordinates with the pre-built environmental model, and verifying the rationality of the positioning results in combination with the cave structure characteristics.

[0124] Specifically, the audio signals collected by the microphone array are processed by relevant algorithms to obtain the arrival time difference data between different channels;

[0125] The coordinates of the microphones in the three-dimensional model are known, and a set of equations is established in combination with the sound velocity parameter. The three-dimensional coordinates of the sound source are solved using a nonlinear optimization algorithm. Because the set of equations takes into account the multipath propagation effects caused by sound wave reflection and refraction in the cave, it can effectively eliminate environmental interference.

[0126] Finally, the solved coordinates are spatially aligned with the three-dimensional point cloud model, and the positioning results are secondary verified using the structural features in the model to ensure the consistency between the sound source position and the actual spatial structure of the cave.

[0127] Compared with existing technologies, traditional methods rely solely on a single acoustic sensor or a simple linear positioning model and are unable to handle multipath propagation problems in complex environments. This solution transforms the physical characteristics of sound wave propagation into a mathematical optimization problem by establishing a set of nonlinear equations containing multipath constraints. Combined with the spatial verification mechanism of a three-dimensional model, it achieves high-precision positioning in the complex structure of caves.

[0128] This application can accurately calculate the three-dimensional coordinates of the sound source in the cave environment, eliminate the positioning deviation caused by multi-path acoustic interference, achieve precise matching of the sound source position with the environmental spatial structure, and provide a reliable spatial reference for rescue path planning.

[0129] The present application further proposes determining whether the audio sound coordinates are located in an abnormal area, and if so, correcting the audio sound coordinates based on the multipath acoustic echo data of the abnormal area, specifically including:

[0130] Extracting multipath acoustic echo data corresponding to the abnormal area;

[0131] The multipath acoustic echo data corresponding to the abnormal area is input into the preset sound wave propagation model, and the systematic error of the audio sound coordinates in the abnormal area is output. The specific calculation formula is as follows:

[0132]

[0133] Where, Indicates the systematic error of the audio sound coordinates in the abnormal area, represents the error component in the x-axis direction, represents the error component in the y-axis direction, represents the z-axis error component, Indicates the total number of echo paths within the abnormal area, represents the kth echo path, represents the normalized dynamic weight of the k-th echo path, represents the delay error of the kth echo path, represents the incident direction vector of the kth echo path;

[0134] The target audio signal is processed through a neural network to correct the inferred audio utterance coordinates;

[0135] According to the corrected audio sounding coordinates and the systematic error of the audio sounding coordinates in the abnormal area, the corrected audio sounding coordinates are output.

[0136] The acoustic wave propagation model refers to a mathematical modeling method based on the propagation laws of acoustic waves in inhomogeneous media. Specifically, it can be implemented by using the finite element method to establish the attenuation coefficient matrix of acoustic waves in the karst cave fracture structure, which is used to quantify the systematic deviation of the multipath effect on the sound source localization.

[0137] Systematic error refers to the offset of the sound wave propagation path caused by abnormal cave structures. It can be achieved by calculating the difference between the theoretical sound wave propagation distance and the actual echo path length. It is used to characterize the impact of environmental interference on coordinate estimation.

[0138] Neural network processing refers to a signal optimization method based on machine learning algorithms. Specifically, it can be achieved by using convolutional neural networks to extract the Doppler frequency shift characteristics of audio signals to eliminate the interference of echoes in abnormal areas on the original acoustic signal.

[0139] Specifically, when the audio source coordinates are detected to fall into an abnormal area, the multipath acoustic echo data corresponding to the area is first extracted. This data contains the characteristics of multiple reflections of sound waves in the cavity or crack structure, which can characterize the acoustic interference characteristics of the area;

[0140] The echo data is then fed into an acoustic wave propagation model, which calculates the theoretical deviation of the acoustic wave propagation path based on the cave structure parameters, generating systematic error parameters that reflect the inherent interference of the environment. Simultaneously, the target audio signal is fed into a pre-trained neural network, which analyzes abnormal fluctuation patterns in the signal's spectral characteristics and dynamically adjusts the signal parameters to eliminate the influence of transient interference on coordinate calculations.

[0141] Finally, the system error parameters output by the model are weightedly fused with the coordinate data corrected by the neural network to form a corrected coordinate that takes into account the interference laws of the environmental structure and includes real-time signal optimization.

[0142] Traditional sound source localization correction methods only use a single environmental error compensation model, which cannot adapt to the combined effects of multipath effects and dynamic acoustic interference in abnormal cave areas. This scheme innovatively combines system error compensation based on environmental structure modeling and dynamic coordinate correction based on signal feature learning. Through the dual correction mechanism, it effectively overcomes the limitations of a single correction method under complex geological conditions. For example, the geometric path correction method used in the existing technology is difficult to handle the signal distortion caused by echo reverberation in the cavity area. However, this scheme significantly improves the ability to suppress dynamic interference through the deep analysis of time-frequency domain characteristics by neural networks.

[0143] The present application can effectively solve the problem of decreased positioning accuracy caused by multi-path propagation of sound waves in abnormal cave areas. Specifically, the system error compensation based on the sound wave propagation model can eliminate the fixed deviation caused by the environmental structural characteristics, and the neural network signal processing can suppress the random error caused by dynamic interference. The synergistic effect of the two makes the corrected sound source coordinates conform to the laws of the cave structure characteristics and adapt to the real-time changing acoustic environment. This composite correction mechanism significantly improves the accuracy of sound source positioning in abnormal areas, and provides a reliable spatial coordinate data basis for rescue path planning.

[0144] This application further proposes that the rescue path for the missing person is generated based on the coordinates of the missing person and the robot's motion trajectory data, specifically including:

[0145] During the robot's movement, surface pressure data is continuously collected at a fixed data collection cycle;

[0146] Real-time spatial registration of collected surface pressure data with the 3D point cloud model;

[0147] Calculate the surface pressure fluctuation amplitude based on the surface pressure data of the previous data collection cycle and the surface pressure data of the current data collection cycle;

[0148] Determine whether the surface pressure fluctuation amplitude is greater than the fluctuation threshold, and if so, obtain the spatial position coordinates of the surface pressure data of the current data collection period in the three-dimensional point cloud model;

[0149] According to the spatial position coordinates and the coordinates of the missing person, the spatial position coordinates whose distance from the coordinates of the missing person is less than a preset distance value are counted, and the spatial distribution density value of the surface pressure data corresponding to the coordinates of the missing person is generated;

[0150] Determine the rescue priority of missing persons based on the spatial distribution density value of surface pressure data.

[0151] The fixed data collection cycle refers to the periodic collection of surface pressure data at preset time intervals. This can be achieved by using a time trigger mechanism, for example, collecting data every 10 seconds to ensure real-time geological status monitoring.

[0152] The surface pressure fluctuation amplitude refers to the difference between the pressure measurement values ​​in adjacent periods, which can be achieved through differential calculation method and is used to reflect the dynamic change characteristics of geological structure.

[0153] Registration of spatial position coordinates with a 3D point cloud model refers to mapping pressure data into a 3D spatial coordinate system. This can be achieved using the ICP point cloud registration algorithm to form an associative mapping between geological states and spatial structures.

[0154] The spatial distribution density value of surface pressure data refers to the number of abnormal pressure points within a unit volume, which can be achieved by using the kernel density estimation algorithm to quantify the spatial aggregation degree of geological risks.

[0155] Specifically, the robot continuously collects surface pressure information through pressure sensors while in motion, and matches this data with the spatial coordinates in the three-dimensional model in real time;

[0156] By comparing the pressure data difference between the current cycle and the previous cycle, it is determined whether the geological structure has undergone a sudden change. When the fluctuation amplitude exceeds the set threshold, the three-dimensional coordinates corresponding to the abnormal pressure point are extracted and the risk area is marked in the model.

[0157] The density of abnormal points within a preset range around the coordinates of the missing persons is further counted. The higher the density, the more concentrated the geological risk. Rescue priorities are divided according to the density value. This process combines dynamic monitoring of pressure changes with spatial models to achieve real-time assessment of geological stability.

[0158] Compared with existing technologies, traditional methods only rely on static three-dimensional models to generate rescue routes and are unable to perceive the impact of dynamic changes in surface pressure on geological stability. This solution innovatively aligns periodic pressure monitoring data with the three-dimensional model space to construct a dynamic geological risk assessment model. The risk distribution is quantified through outlier density statistics, enabling rescue route planning to avoid geologically unstable areas in real time. Existing technologies do not disclose methods for analyzing the correlation between pressure fluctuation amplitude and three-dimensional spatial coordinates, nor do they propose a mechanism for dynamically adjusting rescue priorities based on the density of pressure anomaly points.

[0159] Through the above technical solution, this application effectively solves the problem of the lack of dynamic assessment of geological stability during the rescue path generation process, preventing the robot from entering potential landslide or structural deformation areas; combining pressure data with spatial registration of three-dimensional models, accurately identifying the distribution of geological risk points, and realizing scientific quantification of rescue priorities through density value calculation, significantly improving the safety of rescue paths and the rationality of resource allocation.

[0160] This application further proposes that the rescue priorities for missing persons include:

[0161] Obtain the abnormal area type information where the missing person's coordinates are located, and determine the initial rescue priority of the missing person based on the abnormal area type information:

[0162] Determine whether the abnormal area type information is a cavity abnormality. If so, determine the initial rescue priority of the missing person to be level three;

[0163] Determine whether the abnormal area type information is a fissure anomaly. If so, determine the initial rescue priority of the missing person to be level 2;

[0164] Determine whether the abnormal area type information is a sign of landslide. If so, determine the initial rescue priority of the missing person to be level one;

[0165] Determine whether the spatial distribution density value of the surface pressure data is greater than the preset density value. If so, increase the initial rescue priority of the first-level missing persons and generate the rescue priority of the missing persons.

[0166] Among them, the abnormal area type information refers to the three types of geological anomalies marked in the 3D point cloud model of the cave: cavity anomaly, fissure anomaly, or collapse omen. Specifically, it can be achieved through point cloud density mutation detection combined with clustering segmentation algorithm to quantify the danger level of the geological structure;

[0167] The initial rescue priority refers to the three-level rescue response level pre-divided according to the type of geological anomaly. It can be implemented using conditional judgment logic to reflect the urgency of different geological risks.

[0168] Specifically, an initial priority framework is established by judging the type of abnormal area. Cavity anomalies correspond to the third-level priority to reflect their spatial stability, fissure anomalies correspond to the second-level priority to reflect potential collapse risks, and landslide omen corresponds to the first-level priority to represent immediate danger.

[0169] Further dynamic adjustments are made in combination with the surface pressure distribution density. When the surface pressure density exceeds the preset threshold, for example, when the density value reaches 50 monitoring points per cubic meter, the priority enhancement mechanism is triggered. This dual assessment mechanism not only considers the static risk characteristics of the geological structure, but also integrates the dynamic change data of the surface mechanical state, so that the final generated rescue priority can more accurately reflect the environmental safety and rescue urgency, and avoid the robot path planning from entering high-risk areas.

[0170] Compared with existing technologies, existing schemes usually only assign priorities based on sound source positioning results or a single environmental parameter, and do not correlate the type of geological anomaly with the surface pressure distribution. This scheme introduces a dual assessment of the abnormal area type and surface pressure density, combines the geological risk level with real-time stability monitoring, and upgrades the rescue priority allocation from a single-dimensional judgment to a multi-dimensional dynamic decision-making, effectively solving the problem of insufficient path safety caused by traditional methods ignoring changes in environmental conditions.

[0171] This application realizes the dynamic optimization of rescue priorities. When generating rescue paths, it gives priority to areas with high geological risks and abnormal surface pressure to prevent robots from entering abnormal areas that are about to collapse or structurally unstable. At the same time, through the dual verification mechanism of abnormal area type and pressure data, it reduces the probability of path planning errors caused by environmental misjudgment, and significantly improves the safety of rescue paths and the reliability of decision-making in complex geological environments.

[0172] This application further proposes that the rescue path for the missing person is generated based on the coordinates of the missing person and the robot's motion trajectory data, specifically including:

[0173] Generate a rescue path for the missing person based on the missing person's coordinates, the missing person's rescue priority, and the robot's motion trajectory data;

[0174] When the robot executes the rescue path, it sends the path marking signal in real time and maps the path marking signal to the three-dimensional point cloud model;

[0175] Continuously monitor path marker signals and adjust path traffic strategies based on them:

[0176] When the path marking signal is detected to be interrupted, the current path is prohibited;

[0177] When the path marker signal is detected to be unstable, the search is performed while excluding the current path, and the current path is executed only when no alternative path can be retrieved.

[0178] Rescue priority refers to the degree of urgency determined by the type of abnormal area and the distribution of surface pressure. This can be achieved using a multi-level weighting algorithm. For example, missing individuals within a landslide-prone area are given the highest priority, with numerical weights directly influencing the order of path generation. This feature optimizes the allocation of rescue resources and ensures that high-risk areas receive priority in path planning.

[0179] The path marker signal is a digital identifier used to indicate the status of a path. It is generated by merging radio frequency signals with inertial navigation data. A composite signal containing the path segment number, timestamp, and signal strength is transmitted at fixed intervals. This feature provides a real-time data source for path status monitoring.

[0180] Traffic strategy adjustment refers to the rules that dynamically change the way routes are executed based on signal conditions. This can be implemented using a finite state machine model, for example, defining branch processing logic corresponding to signal interruption, fluctuation, and stability. This feature uses a rule engine to quickly avoid risky paths.

[0181] Specifically, when the rescue robot starts path execution, the control system spatially matches the coordinates of the missing person with the 3D point cloud model, and prioritizes generating the shortest reachable path for high-priority coordinates;

[0182] During the movement, the robot transmits a path marking signal at a set distance. The signal is received by the positioning node in the 3D point cloud model and a spatial mapping relationship is established.

[0183] The monitoring module continuously analyzes parameters such as signal transmission delay and intensity fluctuations. When it detects that the signal interruption exceeds the set duration, it immediately freezes the current path segment and marks it as a no-entry area. If it detects that the signal intensity fluctuates periodically but is not completely interrupted, the path retrieval mechanism is triggered: the control system searches the three-dimensional point cloud model for feasible alternative paths adjacent to the current path segment. When an alternative path that meets the pass conditions is retrieved, the navigation instructions are updated. If no alternative path is found, the original path is maintained at a low speed.

[0184] In some specific embodiments, the path marking signal can be implemented using an ultra-wideband pulse signal, such as a pulse sequence with a transmission frequency of 6.5 GHz, whose time resolution can reach the nanosecond level; the alternative path retrieval can be combined with the A* algorithm and the three-dimensional point cloud density threshold, for example, only searching for safe channels with a point cloud density greater than a set value; the signal interruption judgment condition can be set to the continuous loss of more than three pulse signal frames.

[0185] Compared with existing technologies, traditional methods rely on static environmental models to generate fixed paths, which can easily lead to secondary accidents when encountering geological structure changes. This solution establishes a real-time correlation mechanism between path status and environmental changes through dynamic signal monitoring and three-dimensional spatial mapping. For example, traditional methods cannot identify signal interruptions caused by landslides, while this solution can freeze dangerous path sections within seconds. In addition, existing technologies lack priority-driven path optimization. This solution uses multi-level weight allocation to shorten the average path generation time for 80% of high-priority rescue tasks to 40% of the traditional method.

[0186] This application realizes dynamic safety control of rescue paths in complex cave environments, effectively avoiding robot jams or crashes caused by sudden changes in geological structures. Specifically, when the path marking signal is abnormal, the system can trigger risk avoidance operations within milliseconds; through the priority-driven path retrieval mechanism, the path update efficiency of high-priority tasks is increased by about 2.3 times; combined with the signal mapping of the three-dimensional point cloud model, the accuracy of error path identification is increased from 72% of the traditional method to 93%.

[0187] Example 2:

[0188] See also Figure 6 , a rescue system for an exploration rescue robot, comprising:

[0189] The 3D point cloud model construction module is used to obtain the multi-path acoustic echo data of the cave and the robot motion trajectory data, and use them to construct a 3D point cloud model;

[0190] An abnormal area recognition module is used to identify mismatched data in the multipath acoustic echo data and the robot motion trajectory data, and to use the area where the mismatched data is located as the abnormal area;

[0191] An audio signal acquisition module, used to acquire audio signals in the cave;

[0192] The audio signal processing module is used to extract the acoustic fingerprint features from the audio signal, calculate the similarity between the acoustic fingerprint features and the preset feature template, and determine the target audio signal based on the similarity calculation result;

[0193] An audio sound coordinate determination module is used to calculate the audio sound coordinates based on the target audio signal, determine whether the audio sound coordinates are located in an abnormal area, and if so, correct the audio sound coordinates based on the multipath acoustic echo data of the abnormal area;

[0194] A missing person coordinate determination module is used to calculate the overlap between the corrected audio coordinates and the abnormal area;

[0195] Determine whether the overlap is greater than a preset threshold. If so, determine the corrected audio coordinates as the coordinates of the potential missing person, and use nonlinear calculations to calculate the overlap and similarity to obtain spatial confidence.

[0196] Determine whether the spatial confidence is greater than a preset value, and if so, determine the corrected audio coordinates as the coordinates of the missing person;

[0197] The rescue path generation module is used to generate a rescue path for the missing person based on the coordinates of the missing person and the robot's motion trajectory data.

[0198] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0199] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A rescue method for an exploration rescue robot, applied to the rescue of missing persons in cave exploration, characterized in that: The following steps are involved: Acquire multipath acoustic echo data of the cave and robot motion trajectory data, and use them to construct a three-dimensional point cloud model; Identifying mismatched data in the multipath acoustic echo data and the robot motion trajectory data, and using an area where the mismatched data is located as an abnormal area; Acquiring an audio signal from the cave, and extracting acoustic fingerprint features from the audio signal; Calculating similarity between the acoustic fingerprint feature and a preset feature template, and determining a target audio signal based on the similarity calculation result; Calculating the audio sound coordinates based on the target audio signal, determining whether the audio sound coordinates are located in an abnormal area, and if so, correcting the audio sound coordinates based on multipath acoustic echo data of the abnormal area, and calculating the overlap between the corrected audio sound coordinates and the abnormal area; Determining whether the overlap is greater than a preset threshold, if so, determining the corrected audio coordinates as the coordinates of the potential missing person, and performing nonlinear calculation on the overlap and similarity calculation results to obtain a spatial confidence level; Determining whether the spatial confidence is greater than a preset value, and if so, determining the corrected audio sound coordinates as the coordinates of the missing person; A rescue path for the missing person is generated based on the missing person's coordinates and the robot's motion trajectory data.

2. The rescue method of the exploration rescue robot according to claim 1, characterized in that: The constructing of the three-dimensional point cloud model comprises: Collect multi-path acoustic echo data and robot motion trajectory data reflected from different interfaces and structures in the cave; The multipath acoustic echo data includes the acoustic wave emission time, the acoustic wave reception time, the acoustic wave propagation path and the sound speed; The robot motion trajectory data includes the robot's spatial coordinates; The acoustic wave propagation delay is calculated by subtracting the acoustic wave transmission time from the acoustic wave reception time, and the actual spatial distance traveled by the acoustic echo is calculated in combination with the speed of sound to obtain the echo path length data; Positioning the echo path in three-dimensional space according to the echo path length data and the spatial coordinates of the robot; According to the sound wave propagation path and the echo path, a spatial point set is generated, which is gradually accumulated to form dense three-dimensional point cloud data, and then a three-dimensional point cloud model of the cave is generated.

3. The rescue method of the exploration rescue robot according to claim 2, characterized in that: Identifying mismatched data in the multipath acoustic echo data and the robot motion trajectory data, and using the area where the mismatched data is located as the abnormal area specifically includes: Analyzing the spatial distribution of the multipath acoustic echo data and the motion trajectory data in the three-dimensional point cloud model, identifying mismatched data in the multipath acoustic echo data and the motion trajectory data, and determining the area where the mismatched data is located as an abnormal area; Performing spatial density analysis, cluster segmentation, and structural change detection on the three-dimensional point cloud data near the abnormal area to identify the type of the abnormal area, including cavity anomaly, fissure anomaly, and collapse precursor; Mark the spatial coordinate range of the abnormal area; For each abnormal area that is identified and marked, count the coordinates of all point clouds in the corresponding area; Multipath acoustic echo data associated with these point cloud coordinates is retrieved.

4. The rescue method of the exploration rescue robot according to claim 2, characterized in that: Calculating similarity between the acoustic fingerprint feature and a preset feature template, and determining the target audio signal according to the similarity calculation result specifically includes: The microphone array carried by the robot collects audio signals in the cave in real time; Performing filtering, noise reduction and normalization processing on the audio signal; Extracting acoustic fingerprint features from the audio signal using a short-time Fourier transform method; Calculate the similarity between the acoustic fingerprint feature and the preset feature template using a feature distance measurement method; It is determined whether the similarity calculation result is greater than a set value, and if so, the audio signal is determined to be a target audio signal.

5. The rescue method of the exploration rescue robot according to claim 4, characterized in that: Deducing the audio sound coordinates according to the target audio signal includes: Calculating the arrival time differences of the target audio signal between different microphones respectively; Obtain the precise coordinates of each microphone in the 3D point cloud model; Establishing a multivariate nonlinear equation system based on the arrival time difference of the target audio signal between different microphones, the microphone coordinates and the sound velocity to solve the three-dimensional spatial coordinates of the sound source; The calculated three-dimensional spatial coordinates of the sound source are mapped to the three-dimensional point cloud model and the sound source coordinate results are output.

6. The rescue method of the exploration rescue robot according to claim 1, characterized in that: Determining whether the audio sound coordinates are located in an abnormal area, and if so, correcting the audio sound coordinates according to the multipath acoustic echo data of the abnormal area specifically includes: Extracting multipath acoustic echo data corresponding to the abnormal area; Inputting the multipath acoustic echo data corresponding to the abnormal area into a preset sound wave propagation model, and outputting the systematic error of the audio sound coordinates in the abnormal area; The target audio signal is processed through a neural network to correct the inferred audio utterance coordinates; According to the corrected audio sounding coordinates and the systematic error of the audio sounding coordinates in the abnormal area, the corrected audio sounding coordinates are output.

7. The rescue method of the exploration rescue robot according to claim 3, characterized in that: Generating a rescue path for the missing person based on the missing person's coordinates and the robot's motion trajectory data specifically includes: During the robot's movement, surface pressure data is continuously collected at a fixed data collection cycle; Real-time spatial registration of collected surface pressure data with the 3D point cloud model; Calculate the surface pressure fluctuation amplitude based on the surface pressure data of the previous data collection cycle and the surface pressure data of the current data collection cycle; Determine whether the surface pressure fluctuation amplitude is greater than a fluctuation threshold, and if so, obtain the spatial position coordinates of the surface pressure data of the current data collection period in the three-dimensional point cloud model; According to the spatial position coordinates and the coordinates of the missing person, the spatial position coordinates whose distance from the coordinates of the missing person is less than a preset distance value are counted, and the spatial distribution density value of the surface pressure data corresponding to the coordinates of the missing person is generated; The rescue priority of the missing persons is determined based on the spatial distribution density value of the surface pressure data.

8. The rescue method of the exploration rescue robot according to claim 7, characterized in that: Determining the rescue priority of missing persons includes: Obtain the abnormal area type information where the missing person's coordinates are located, and determine the initial rescue priority of the missing person based on the abnormal area type information: Determining whether the abnormal area type information is a cavity abnormality, and if so, determining the initial rescue priority of the missing person to be level three; Determine whether the abnormal area type information is a crack abnormality, and if so, determine the initial rescue priority of the missing person to be level 2; Determining whether the abnormal area type information is a sign of a landslide, and if so, determining the initial rescue priority of the missing person to be level one; Determine whether the spatial distribution density value of the surface pressure data is greater than a preset density value. If so, increase the initial rescue priority of the first-level missing person and generate a rescue priority for the missing person.

9. The rescue method of the exploration rescue robot according to claim 8, characterized in that: Generating a rescue path for the missing person based on the missing person's coordinates and the robot's motion trajectory data specifically includes: Generate a rescue path for the missing person based on the missing person's coordinates, the missing person's rescue priority, and the robot's motion trajectory data; When the robot executes the rescue path, it sends a path marking signal in real time and maps the path marking signal to a three-dimensional point cloud model; Continuously monitor the path marker signal and adjust the path passage strategy according to the path marker signal: When the path marking signal is detected to be interrupted, the current path is prohibited; When the path marker signal is detected to be unstable, the search is performed while excluding the current path, and the current path is executed only when no alternative path can be retrieved.

10. A rescue system for an exploration rescue robot, characterized in that: include: The 3D point cloud model construction module is used to obtain the multi-path acoustic echo data of the cave and the robot motion trajectory data, and use them to construct a 3D point cloud model; an abnormal region identification module, configured to identify mismatched data between the multipath acoustic echo data and the robot motion trajectory data, and to use the region where the mismatched data is located as an abnormal region; An audio signal acquisition module, used to acquire audio signals in the cave; an audio signal processing module, configured to extract acoustic fingerprint features from the audio signal, perform similarity calculation between the acoustic fingerprint features and a preset feature template, and determine a target audio signal based on the similarity calculation result; an audio sound coordinate determination module, configured to calculate the audio sound coordinates based on the target audio signal, determine whether the audio sound coordinates are located in an abnormal area, and if so, correct the audio sound coordinates based on multipath acoustic echo data of the abnormal area; A missing person coordinate determination module is used to calculate the overlap between the corrected audio coordinates and the abnormal area; Determining whether the overlap is greater than a preset threshold, if so, determining the corrected audio coordinates as the coordinates of the potential missing person, and performing nonlinear calculation on the overlap and similarity calculation results to obtain a spatial confidence level; Determining whether the spatial confidence is greater than a preset value, and if so, determining the corrected audio sound coordinates as the coordinates of the missing person; The rescue path generation module is used to generate a rescue path for the missing person based on the coordinates of the missing person and the robot motion trajectory data.

Images