Intelligent pet state translation method and system based on multi-modal data

Abstract

The invention provides an intelligent pet state translation method and system based on multi-modal data. According to the method, multi-modal data of an outdoor environment where a target pet is located is collected, an obstacle distribution map is generated to quantify the environment obstacle density and the space shielding relation, and combined coding is carried out in combination with the multi-modal data to generate a feature vector reflecting environment pressure. The method comprises the following steps: synchronously obtaining body movement time sequence data of a pet, performing mode matching with a preset pet body language database, and extracting behavior abnormal fragments associated with outdoor environment temporal and spatial changes; by analyzing the mapping relation between the environment pressure feature vector and the abnormal behavior segment, the stress level of the pet is predicted, and finally a translation instruction containing a high-density obstacle area avoidance path prompt is generated, so that intelligent translation of linkage of the pet behavior state and the environment risk is realized. According to the technical scheme provided by the invention, the intelligent translation efficiency and accuracy of the pet state can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent translation of pet status, and in particular to a method and system for intelligent translation of pet status based on multimodal data. Background Art

[0002] In complex outdoor environments, a pet's stress response is crucial to their health and safety. With accelerating urbanization, pets and their owners have more opportunities to participate in outdoor activities together. Environmental changes, such as noise, unfamiliar creatures, or crowds, can trigger stress responses. Accurately identifying and promptly responding to these conditions can help prevent potential health issues and enhance communication and understanding. To achieve this, analyzing a pet's posture changes, vocal frequency, and environmental factors can provide deep insight into their immediate emotional state, enabling owners or trainers to take timely action and prevent potential risks.

[0003] Currently, a targeted solution to this technological need is to utilize intelligent monitoring systems installed in outdoor environments to assess pets' condition. Integrating high-definition cameras, audio capture equipment, and environmental sensors, these systems can comprehensively capture a pet's visual and auditory behavior and analyze it in conjunction with surrounding environmental parameters. By identifying characteristics such as the pet's behavioral trajectory, postural changes, and vocalization frequency, timely action can be taken. This approach provides a new approach for understanding and managing pets' stress responses in complex outdoor environments.

[0004] While the above solution meets the need for intelligent translation of pet status to a certain extent, it still has some significant drawbacks. First, due to its reliance on fixed-position monitoring equipment, its coverage is limited, and for pets with wide ranges of activity, blind spots may occur. Second, it exhibits certain limitations when dealing with complex background noise and differences between individual pets, which may lead to misjudgments or missed detections. Furthermore, this solution focuses primarily on analyzing external behavior and sounds, while ignoring changes in internal physiological indicators such as heart rate and body temperature, which are also crucial for fully understanding a pet's true condition. Therefore, overcoming these shortcomings and improving the accuracy and reliability of intelligent translation of pet status remains an important direction for future research. Summary of the Invention

[0005] The present application provides a pet status intelligent translation method and system based on multimodal data to solve the problems of low efficiency and poor accuracy of pet status intelligent translation in the prior art.

[0006] In a first aspect, the present application provides a pet status intelligent translation method based on multimodal data, comprising:

[0007] Acquire multimodal data of the target pet's complex outdoor environment, including noise decibels, crowd density, and three-dimensional spatial data captured by a LiDAR.

[0008] generating an obstacle distribution map based on the three-dimensional spatial data, and jointly encoding the multimodal data and the obstacle distribution map to generate an environmental pressure feature vector;

[0009] Synchronously collecting the target pet's body movement time series data, and performing pattern matching on the body movement time series data with a preset pet body language database to extract abnormal behavior segments that are temporally and spatially associated with the complex outdoor environment;

[0010] Determining a predicted result of the stress level of the target pet in the outdoor complex environment by mapping the environmental pressure feature vector with the abnormal behavior segment;

[0011] A translation instruction is generated according to the stress level prediction result, wherein the translation instruction includes an avoidance path prompt for a high-density area in the obstacle distribution map.

[0012] Optionally, after generating the translation instruction according to the stress level prediction result, the method further includes:

[0013] Building a large base model based on multimodal data, the large base model includes cross-modal analysis results of human speech, environmental sounds, and animal calls;

[0014] Jointly labeling the cross-modal parsing results of the animal calls according to the environmental stress feature vector and the abnormal behavior segment to obtain a labeling result, wherein the labeling result includes a mapping relationship between dog breed category, emotion type and environmental stress;

[0015] Extracting Mel-spectrum features of the annotation result using the large base model to obtain a spectrum vector, aligning the spectrum vector with the environmental pressure feature vector, and adjusting parameters of the large base model based on the alignment result to obtain an adjusted large base model;

[0016] Constructing dog language reasoning logic based on the adjusted base large model, and using the dog language reasoning logic to adjust the language generation strategy of the translation instruction to generate a natural language description that matches the personality characteristics of the target pet;

[0017] Based on the natural language description, combined with the body movement timing data and the noise decibel level, the emotional state of the target pet is identified, and emotional soothing suggestions corresponding to the emotional state are added to the translation instruction;

[0018] The translation instruction with the emotional comfort suggestion is pushed to the user terminal, and the avoidance path prompt in the translation instruction is dynamically updated according to the changes in the high-density area in the obstacle distribution map.

[0019] Optionally, the obstacle distribution map is constructed by quantifying the density of mobile obstacles and spatial occlusion coefficients within the target pet's activity range;

[0020] Generating an obstacle distribution map based on the three-dimensional spatial data includes:

[0021] Dividing the target pet's corresponding activity range into multiple spatial grid cells based on the three-dimensional spatial data, extracting the movement trajectory of obstacles in each spatial grid cell and their frequency of occurrence within a preset time period, and calculating the moving obstacle density and spatial occlusion coefficient of each spatial grid cell based on the change in the degree of line of sight occlusion between adjacent spatial grid cells;

[0022] The mobile obstacle density and spatial occlusion coefficient of each spatial grid unit are mapped into a three-dimensional thermal map, which is superimposed to generate the obstacle distribution map.

[0023] Optionally, the jointly encoding the multimodal data and the obstacle distribution map to generate an environmental pressure feature vector includes:

[0024] Extracting a parameter set from the multimodal data that overlaps with a high-density area in the obstacle distribution map based on the temporal trend of noise decibels and the spatial distribution of crowd density, and performing time-series accumulation on parameters in the parameter set whose fluctuation amplitude exceeds a preset threshold to generate a parameter sequence reflecting sudden environmental interference events;

[0025] Correlating and matching the mobile obstacle density and spatial occlusion coefficient with a parameter sequence, determining a diffusion path of the sudden environmental interference event in the obstacle distribution map, and adjusting the mobile obstacle density and spatial occlusion coefficient of each spatial grid unit according to the diffusion path;

[0026] Based on the adjusted mobile obstacle density, spatial occlusion coefficient and the temporal accumulation results of the parameter sequence, the temporal fluctuation characteristics of the multimodal data and the spatial topological characteristics of the obstacle distribution map are tensor-spliced, and the splicing results are encoded through a convolutional neural network to generate an environmental pressure feature vector.

[0027] Optionally, the limb movement time series data includes a continuous change sequence of limb swing frequency, trunk contraction amplitude and head deflection angle;

[0028] The method of pattern matching the limb movement time series data with a preset pet body language database to extract abnormal behavior segments that are temporally and spatially associated with the outdoor complex environment includes:

[0029] Dividing the limb movement time series data into analysis windows of variable length, and dynamically adjusting the sliding interval of the analysis windows according to the spatial density change of the obstacle distribution map in the outdoor complex environment;

[0030] In each analysis window, based on the motion change characteristics of adjacent moments in the limb motion time series data, the degree of matching between the limb motion parameters in the current analysis window and the standard behavior patterns in the pet body language database is calculated, and the parameters related to visual interference in the environmental pressure feature vector are combined to generate dynamic matching conditions;

[0031] Filtering the analysis windows whose matching degree is lower than a preset benchmark according to the dynamic matching condition, extracting the periodic differences between the limb swing frequency and the trunk contraction amplitude within the analysis window, and combining the location information of the high-density area in the obstacle distribution map to mark the abnormal behavior segments that are positionally associated with the outdoor complex environment;

[0032] The abnormal behavior segments are correlated and compared with the spatial distribution changes of the crowd density in the multimodal data, and non-stress action data generated by the pet's autonomous behavior are excluded to generate the spatiotemporally correlated abnormal behavior segment set.

[0033] Optionally, the identifying the emotional state of the target pet based on the natural language description, combined with the body movement timing data and the noise decibels, and adding emotional comfort suggestions corresponding to the emotional state to the translation instruction includes:

[0034] Performing word segmentation processing on the natural language description, extracting context-related features, and generating a sentiment feature vector, wherein the sentiment feature vector includes word segmentation weight distribution and semantic dependency strength;

[0035] Establishing a two-dimensional feature space with body movement amplitude and movement frequency as axes, constructing a dynamic trajectory point set based on the emotional feature vector and the target pet's body movement time series data within a set time window, and dividing the emotional state into emotional intensity levels according to the distribution density of the dynamic trajectory point set in the two-dimensional feature space;

[0036] Extracting voiceprint pulse intervals whose duration exceeds a threshold in the noise decibel sequence, weightedly coupling the energy integral value of each voiceprint pulse interval with the motion frequency of the dynamic trajectory point set to generate an emotional state determination coefficient;

[0037] A bidirectional mapping table between emotion labels and body movement patterns is constructed based on the frequency domain harmonic components of the voiceprint pulse interval. When the emotion state determination coefficient exceeds the dynamic baseline, the context-related features are used to correct the confidence level of the emotion label corresponding to the current body movement pattern.

[0038] When the confidence of the corrected emotion label exceeds a preset threshold, an interactive instruction template that is positively correlated with the emotion intensity level is matched from the emotion soothing strategy library, and the interactive instruction template is semantically aligned with the context of the translation instruction to output a translation instruction with additional emotion soothing suggestions.

[0039] Optionally, the step of performing time-series accumulation on parameters in the parameter set whose fluctuation amplitude exceeds a preset threshold to generate a parameter sequence reflecting an environmental sudden interference event includes:

[0040] Establish a dynamic fluctuation threshold in time slices. Based on the historical trend of noise decibels over time and the historical spatial distribution of crowd density in the parameter set, set the range exceeding two standard deviations of the median of the historical trend or historical distribution change as the initial trigger boundary;

[0041] Calculate the parameter value fluctuation difference of adjacent time slices. When the difference direction of three consecutive time slices is consistent and the fluctuation amplitude exceeds the initial trigger boundary, lock the current time slice to the end mark of the interference event;

[0042] Based on the locked time slice, the frequency domain energy surge interval of noise decibels and the mutation direction of the spatial gradient of crowd density are extracted, and the overlapping areas in the time dimension are divided into independent interference segments, and interference segments with temporal continuity or overlapping spatial coverage are accumulated;

[0043] The accumulated interference segments are parameter-weighted, wherein the duration of the frequency domain energy surge interval and the spatial diffusion speed of the crowd density spatial gradient are used as weight factors to generate a parameter sequence reflecting the sudden interference event in the environment.

[0044] Optionally, the calculating the degree of matching between the limb movement parameters in the current analysis window and the standard behavior patterns in the pet body language database based on the movement change characteristics at adjacent moments in the limb movement time series data includes:

[0045] Decomposing the continuous change sequence of limb swing frequency, trunk contraction amplitude and head deflection angle in the limb movement time series data into motion trajectories of multiple joints, and extracting the motion change characteristics of each joint at adjacent moments;

[0046] Based on the gradient direction of the spatial density change in the obstacle distribution map, the weight distribution of the joint points of the motion trajectory decomposition is dynamically adjusted, and the offset of the joint points of the limb parts corresponding to the high-density areas is calculated;

[0047] Based on the offset calculation result, the cosine value of the angle between the motion change feature and the reference trajectory corresponding to the standard behavior pattern is counted in the analysis window, and the cosine value of the angle is dynamically thresholded based on the time distribution characteristics of the visual interference parameter in the environmental pressure feature vector;

[0048] The truncated cosine values ​​of the included angle are weighted and summed according to the joint weights to generate a local matching degree. The local matching degree is then phase-compensated based on the periodic difference between the limb swing frequency within the analysis window and the standard behavior pattern in the pet body language database to obtain the matching degree of the current analysis window.

[0049] Optionally, determining the predicted result of the stress level of the target pet in the outdoor complex environment by associating the environmental pressure feature vector with the abnormal behavior segment includes:

[0050] Constructing a spatiotemporal alignment relationship between the environmental pressure feature vector and the abnormal behavior segment, and extracting a joint feature set of the environmental pressure feature vector and the abnormal behavior segment within an overlapping time window;

[0051] Performing nonlinear correlation analysis on the joint feature set, capturing the coupling relationship between the environmental pressure feature vector and the behavioral abnormality segment through a multi-layer feature interaction network, and generating a coupling strength coefficient representing the strength of the correlation between the environment and the behavior;

[0052] Based on the numerical distribution range of the coupling strength coefficient, dividing the stress level threshold interval of the target pet in the outdoor complex environment;

[0053] According to the landing point of the coupling strength coefficient in the stress level threshold range, the confidence weight of the stress level is dynamically allocated, and the confidence weight is corrected in combination with the same environmental pressure in historical stress events to obtain a stress level prediction result.

[0054] In a second aspect, the present application provides a pet status intelligent translation system based on multimodal data, comprising:

[0055] An acquisition module acquires multimodal data of the complex outdoor environment in which the target pet is located, wherein the multimodal data includes noise decibels, crowd density, and three-dimensional spatial data captured by a lidar;

[0056] an encoding module, which generates an obstacle distribution map based on the three-dimensional spatial data, and jointly encodes the multimodal data and the obstacle distribution map to generate an environmental pressure feature vector;

[0057] An extraction module synchronously collects the target pet's body movement time series data, performs pattern matching on the body movement time series data with a preset pet body language database, and extracts abnormal behavior segments that are temporally and spatially associated with the outdoor complex environment;

[0058] a mapping module, which determines a predicted result of the stress level of the target pet in the outdoor complex environment by associating the environmental pressure feature vector with the abnormal behavior segment;

[0059] A generation module generates a translation instruction according to the stress level prediction result, wherein the translation instruction includes an avoidance path prompt for a high-density area in the obstacle distribution map.

[0060] In a third aspect, an embodiment of the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a pet status intelligent translation method based on multimodal data as described in the first aspect above.

[0061] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a pet status intelligent translation method based on multimodal data as described in the first aspect.

[0062] In an embodiment of the present application, multimodal data of the complex outdoor environment in which the target pet is located is obtained, the multimodal data including noise decibels, crowd density, and three-dimensional spatial data captured by a lidar; an obstacle distribution map is generated based on the three-dimensional spatial data, and the multimodal data and the obstacle distribution map are jointly encoded to generate an environmental pressure feature vector; the limb movement timing data of the target pet is synchronously collected, and the limb movement timing data is pattern-matched with a preset pet body language database to extract abnormal behavioral segments that are temporally and spatially associated with the complex outdoor environment; a predicted result of the stress level of the target pet in the complex outdoor environment is determined by associating the environmental pressure feature vector with the abnormal behavioral segments; and a translation instruction is generated according to the stress level prediction result, the translation instruction including an avoidance path prompt for high-density areas in the obstacle distribution map.

[0063] The technical solution of this application has the following beneficial effects:

[0064] This application can fully understand the characteristics of the pet's environment by collecting noise decibels, crowd density, and three-dimensional spatial data captured by lidar, providing basic data support for subsequent analysis. By jointly encoding multimodal data with the obstacle distribution map to obtain the environmental pressure feature vector, it can effectively identify potential risk factors in the environment and quantify the stress level caused by these factors to pets, which helps to understand the stressors that pets may face. By extracting abnormal behavioral fragments of pets and monitoring and analyzing their behavior, it is possible to accurately detect abnormal behaviors of pets in specific environments, providing a direct basis for assessing the pet's stress state. Combining environmental information and abnormal pet behavior to assess the pet's stress level makes the judgment of the stress state more scientific and reasonable. Finally, avoidance path prompts for high-density areas in the obstacle distribution map are obtained, aiming to provide corresponding action guidance based on the pet's stress level, helping pets avoid high-risk areas and improve safety.

[0065] After generating the translated instructions, the system optimizes the language generation strategy based on the target pet's historical behavioral data and personalized behavior patterns, generating a natural language description tailored to its personality traits. It also combines the pet's emotional state with body movement timing data and noise decibel fluctuations to identify the pet's emotional state and add emotional soothing suggestions to the translated instructions. Ultimately, the system pushes instructions combining emotional soothing and avoidance routes to the user's terminal, dynamically adjusting the route prompts based on changes in the obstacle distribution map. This process not only accounts for individual pet differences but also enhances the pet's ability to cope with environmental changes through dynamic adjustment strategies, while also providing emotional support, improving the user experience, and enhancing the pet's safety and comfort.

[0066] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0068] Figure 1 A flowchart of a pet status intelligent translation method based on multimodal data provided by the present application is shown;

[0069] Figure 2 A schematic diagram of the structure of a pet status intelligent translation system based on multimodal data provided by the present application is shown;

[0070] Figure 3A schematic structural diagram of a computing device provided by the present application is shown. DETAILED DESCRIPTION

[0071] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0072] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0073] The technical solution of this application is applicable to stress response translation scenarios in complex outdoor environments. By integrating multimodal data, including noise decibels, crowd density, and three-dimensional spatial data captured by lidar, a comprehensive perception system for the target pet's environment is constructed, and the target pet's body movement time series data is simultaneously collected and analyzed. Finally, based on the assessment results of the pet's stress level, translation instructions containing avoidance path prompts are generated, aiming to help pets avoid high-risk areas in complex outdoor environments and improve the safety and comfort of pets' outdoor activities. This process embodies a systematic solution from environmental monitoring to pet behavior analysis to the generation of personalized guidance suggestions.

[0074] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0075] Figure 1 A flowchart of a pet status intelligent translation method based on multimodal data is provided for an embodiment of the present application. Figure 1 As shown, the method includes:

[0076] 101. Acquire multimodal data of a complex outdoor environment in which the target pet is located, wherein the multimodal data includes noise decibels, crowd density, and three-dimensional spatial data captured by a laser radar;

[0077] In this step, multimodal data refers to acoustic, visual, and spatial perception data collected synchronously by multiple sensors, specifically including noise decibels, crowd density, and lidar three-dimensional spatial data.

[0078] Noise decibel is a quantitative indicator of the ambient sound pressure level collected by a microphone array, which is used to reflect the instantaneous or continuous intensity of acoustic interference.

[0079] Crowd density is the number of human targets per unit area calculated by visual cameras or infrared thermal imaging technology, and is used to characterize the distribution density of social stressors in the environment.

[0080] Three-dimensional spatial data is the depth point cloud data generated by the lidar by emitting laser pulses and receiving reflected signals. It is used to reconstruct the three-dimensional topological structure of the environment, including the position, shape and movement trajectory of obstacles.

[0081] In the embodiment of the present application, a laser radar is used to perform an all-round scan of the target area, and the distance to the object is calculated by the reflection time of the laser beam. Point cloud data containing the coordinates of the obstacle is generated, and the outline of the obstacle is extracted using a density-based spatial clustering algorithm. A distributed microphone array is deployed to collect environmental soundprint signals, and the frequency domain energy integral is calculated after Fourier transforming the sound wave signal to obtain the noise decibel value of different time slices. The image is captured by a visual camera, and a pre-trained detection convolutional neural network model is used to identify and count the number of pet targets in the image. Finally, the laser radar, microphone, and camera data are associated under a unified timing reference through timestamp alignment technology to ensure the spatiotemporal consistency of multimodal data.

[0082] In a park trail scenario, a lidar system scans trees, benches, and pedestrians along the trail, generating point cloud data containing the three-dimensional coordinates of obstacles. A distributed microphone array detects a sudden increase in the sound pressure level of children playing in the distance, reaching 85 decibels and continuing to fluctuate. A visual camera captures a crowd gathering at the trail entrance, and a pedestrian detection model calculates a crowd density of three people per square meter. These three datasets are timestamp-aligned to form a synchronized multimodal dataset.

[0083] 102. Generate an obstacle distribution map based on the three-dimensional spatial data, and jointly encode the multimodal data and the obstacle distribution map to generate an environmental pressure feature vector;

[0084] In this step, the obstacle distribution map is a spatial topological model generated through three-dimensional spatial data analysis, which quantifies the density gradient, line of sight obstruction coefficient and moving obstacle trajectory of obstacles in the activity area.

[0085] The environmental pressure feature vector is a composite feature vector generated by fusing noise decibels, crowd density, and obstacle distribution maps, and is used to characterize the comprehensive interference intensity of the environment on pets.

[0086] In an embodiment of the present application, the three-dimensional spatial data generated by the laser radar is divided into uniform grid units, the frequency of occurrence of obstacles in each unit is counted, and the line of sight occlusion coefficient between adjacent units (that is, the probability that the line of sight is blocked by an obstacle) is calculated. The trajectory of moving obstacles (such as pedestrians and vehicles) is tracked, and their movement direction and speed are predicted by the Kalman filter algorithm. The noise decibel sequence is mapped to the corresponding grid unit according to the time slice, and the mean sound pressure level and fluctuation variance of each unit are calculated. The crowd density distribution is superimposed on the grid unit to calculate the coverage of the high-density area. The noise, crowd and obstacle data are weightedly fused through the attention mechanism, in which the weights of the sudden high-decibel area and the high-density crowd area are increased to generate an environmental pressure feature vector containing the joint features of space, acoustics and vision.

[0087] Continuing with the above example, the lidar point cloud data was divided into 0.5m x 0.5m grid cells. Calculations showed that the obstacle density at the trail entrance was 0.8 (maximum 1), the view obstruction coefficient in the bench area was 0.9, and the moving obstacle tracking module predicted a runner approaching the pet at 2 meters per second. The noise decibel sequence was mapped to the corresponding grid cells. The average sound pressure level at the entrance was 80 decibels, and the crowd density distribution showed 70% entrance coverage. After weighted fusion using the attention mechanism, the environmental pressure feature vector reinforced the composite characteristics of "high decibels + high density + approaching moving obstacles" at the entrance.

[0088] 103. Synchronously collect the target pet's body movement time series data, and perform pattern matching on the body movement time series data with a preset pet body language database to extract abnormal behavior segments that are temporally and spatially associated with the outdoor complex environment;

[0089] In this step, the limb movement timing data is the joint motion parameters continuously recorded by the inertial sensor in the pet wearable device, including limb swing frequency, trunk contraction amplitude and head deflection angle.

[0090] The pet body language database is a predefined standardized behavior pattern library that contains joint movement feature templates for common stress actions (such as freezing and fleeing) and non-stress actions (such as walking and sniffing).

[0091] Abnormal behavioral segments refer to sequences of movements exhibited by pets within a specific time period that significantly deviate from the pre-set standard behavioral pattern. These segments contain abnormal fluctuations in joint motion parameters (such as limb swing frequency, trunk contraction amplitude, and head deflection angle) within the limb movement time series data, and must be strongly correlated with the spatiotemporal labels of the environmental pressure feature vector (such as periods of sudden noise decibel increases and areas with dense obstacles).

[0092] In an embodiment of the present application, a nine-axis inertial measurement unit (accelerometer, gyroscope, magnetometer) is embedded in the pet collar or vest to collect three-axis acceleration, angular velocity and direction data of the limb joints. The original data is segmented into sliding windows to extract parameters such as the swing frequency and contraction amplitude within the window. The action sequence of the current window is aligned with the standard behavior template in the database through the dynamic time warping algorithm, and the similarity score of the joint trajectory is calculated. Combined with the spatiotemporal label of the environmental pressure feature vector (such as high noise period, obstacle-dense area), the time window with a similarity lower than the preset threshold is screened and marked as an abnormal behavior segment associated with the outdoor environment in time and space.

[0093] The nine-axis inertial sensor in the pet wearable device detected a sudden increase in trunk contraction from 5 cm to 15 cm in 2 seconds (the average for normal walking is 8 cm), and a persistent rightward head deflection of more than 30 degrees. The Dynamic Time Warping algorithm calculated that this pattern was only 40% similar to a "normal exploration" pattern. Combined with the high-risk entry point in the current environmental pressure feature vector, the segment (10:05:23-10:05:25) was identified as a behavioral anomaly.

[0094] 104. Determine a predicted result of the stress level of the target pet in the outdoor complex environment by mapping the environmental pressure feature vector with the abnormal behavior segment;

[0095] In this step, association mapping is to establish a nonlinear relationship between environmental stress feature vectors and behavioral abnormality segments through machine learning models to quantify the impact of environmental interference on pet behavior.

[0096] The stress level prediction result is a dynamic classification result based on the clustering of historical data, including three levels: low stress, medium stress, and high stress.

[0097] In the embodiment of the present application, the environmental pressure feature vector (such as noise decibel gradient, obstacle density) and the motion parameters of the abnormal behavior segment (such as swing frequency offset, contraction amplitude difference) are input into the multi-layer perceptron model. The model learns the nonlinear relationship between the environment and behavior through the fully connected layer and outputs the stress intensity score. Based on the Gaussian mixture model, the historical stress scores are clustered and analyzed to dynamically divide the threshold intervals of low, medium and high stress levels. For example, when the product of the noise decibel and the obstacle density in the environmental pressure feature vector exceeds the threshold, and the motion parameter offset of the abnormal behavior segment reaches the critical value, it is determined to be a high stress level.

[0098] The environmental pressure characteristics at the entrance (noise decibel gradient +15dB / s, obstacle density 0.8) and the behavioral abnormality segment parameters (contraction amplitude offset +87.5%) were input into the multi-layer perceptron model, outputting a stress intensity score of 92 out of 100. Gaussian mixture model cluster analysis showed that the historical lower limit of the high stress threshold was 85 points, and the current score triggered the "high stress" level.

[0099] 105. Generate a translation instruction based on the stress level prediction result, where the translation instruction includes an avoidance path prompt for a high-density area in the obstacle distribution map.

[0100] In this step, the translation instruction is a composite output including a natural language description and a path planning instruction, which is used to guide the user to adjust the pet's actions.

[0101] High-density areas are defined as spatial units in the obstacle distribution map where the frequency of obstacles is significantly higher than the average level, or where the line of sight obstruction coefficient exceeds a preset threshold. These areas include densely populated static obstacles (such as fixed buildings and trees) and overlapping areas where the movement trajectories of dynamic obstacles (such as crowds and moving vehicles) overlap.

[0102] The avoidance path prompt is a dynamic obstacle avoidance route generated based on the topological structure of high-density areas in the obstacle distribution map, avoiding areas with dense moving obstacles.

[0103] In an embodiment of the present application, a predefined instruction template library is matched according to the stress level. For example, a high stress level corresponds to "a high-risk area is detected ahead, and it is recommended to detour immediately." Combined with the predicted trajectory of the moving obstacles in the obstacle distribution map, a path search algorithm is used to calculate the minimum risk path from the current position to the safe area. The path weight is determined by the obstacle density, the line of sight obstruction coefficient and the noise decibel, and low-density, high-visibility routes are given priority. The path coordinates are converted into natural language descriptions (such as "turn right and go 50 meters along the green belt") and integrated with emotional comfort suggestions (such as "keep the leash loose to reduce pet anxiety") to generate translation instructions.

[0104] Based on the high-stress level command template, the path planning module calculates an obstacle avoidance path, turning right from the current location and circumventing the green belt, avoiding the entrance crowd area and the bench obstruction area (path risk value drops from 0.78 to 0.21). A translated command is generated, such as "A crowd and high-decibel noise are detected ahead. It is recommended to immediately turn right and circumvent the green belt for 50 meters. Keep the leash loose to relieve the pet's anxiety." The path is simultaneously highlighted on the user terminal map and marked as "High-risk area with dense crowds."

[0105] In summary, steps 101 to 105 use multimodal data to perceive the combined risks of crowds, noise, and moving obstacles at the entrance, and accurately determine it as a high-stress state based on abnormal movements such as pet trunk contraction and head deflection. The generated instructions not only provide a dynamic obstacle avoidance path (bypassing crowds and blocked areas), but also provide additional soothing suggestions to achieve a closed-loop response between environmental risks and pet behavior. Ultimately, it helps users guide their pets safely out of high-pressure areas, verifying the effectiveness of the entire link from data collection to decision output.

[0106] In order to further improve the personalized adaptation capability and environmental dynamic response accuracy of pet stress translation instructions, the solution customizes the language style of translation instructions by calling the base large model to analyze historical behavior and real-time data, ensuring that the instructions are consistent with the pet's personality, and can identify the pet's emotional state and provide corresponding emotional comfort suggestions. In addition, these suggestions are pushed to the user, and the path guidance is dynamically updated according to environmental changes to enhance the pet's sense of security and comfort. In some embodiments, after generating the translation instructions according to the stress level prediction results in step 105, it also includes: 201, constructing a base large model based on multimodal data, the base large model including cross-modal analysis results of human voice, environmental sounds and animal calls;

[0107] In step 201, the pedestal model is a deep learning model pre-trained on massive amounts of multimodal audio data from the internet. Its inputs include human speech waveforms, ambient sound spectra, and animal call time-domain signals, and its output is a cross-modal semantic parsing result. The cross-modal parsing result is an intermediate feature representation of the pedestal model's unified understanding of different types of audio data. It can map the emotional tendencies in human speech, the danger signals in ambient sounds, and the emotional fluctuations in animal calls into the same semantic space.

[0108] In an embodiment of the present application, by loading a pre-trained general audio large model as a base, the model adopts a fusion architecture of deep convolutional neural network and self-attention mechanism. First, multimodal data collected in the outdoor environment is input and processed through parallel branches. The human voice branch performs voice activity detection and frame processing, the environmental sound branch performs Mel-frequency cepstral coefficient feature extraction, and the animal call branch performs time domain envelope analysis. The output features of the three branches are interactively fused through the cross-modal attention layer to generate a cross-modal parsing result containing semantic relevance, which serves as the parsing basis for subsequent steps. This process enables the model to have the ability to understand the sound association between "people, environment and animals".

[0109] 202. Jointly labeling the cross-modal parsing results of the animal calls according to the environmental stress feature vector and the abnormal behavior segment to obtain a labeling result, wherein the labeling result includes a mapping relationship between dog breed category, emotion type, and environmental stress;

[0110] In step 202, joint labeling involves adding multidimensional labels to animal vocalization data using environmental stress feature vectors and behavioral anomaly segments. Within the labeling dimension, the dog breed category corresponds to the target pet's breed identifier, and emotion types include discrete categories such as fear and excitement. The environmental stress mapping involves establishing a quantitative correlation matrix between vocalization features and specific environmental stress values.

[0111] In an embodiment of the present application, cross-modal analysis results of the target pet in a complex outdoor environment are collected, and the environmental stress feature vectors and abnormal behavior segments are synchronously associated. The labeling system uses semi-automatic rules. First, the environmental stress feature vectors are divided into high / medium / low risk levels through a clustering algorithm. Secondly, the preset emotion label library is matched according to the amplitude of the body movements in the abnormal behavior segments. Finally, a labeling result is generated that includes the mapping relationship between dog breed category, emotion type, and environmental stress. This labeling result serves as a supervisory signal to guide the model to learn the causal chain between environment and emotion.

[0112] 203. Extracting Mel-spectrum features of the annotation result using the large base model to obtain a spectrum vector, aligning the spectrum vector with the environmental pressure feature vector, and adjusting parameters of the large base model based on the alignment result to obtain an adjusted large base model.

[0113] In step 203, the spectral vector is the frequency domain feature sequence converted from the annotated call signal using a Mel filter bank, representing the call's acoustic fingerprint. Feature alignment involves projecting the spectral vector and the ambient pressure feature vector into the same metric space to minimize the distribution distance between the two types of features.

[0114] In an embodiment of the present application, the annotation results are input into the call parsing branch of the base large model, and the Mel spectrum features are extracted through multiple layers of one-dimensional convolutional layers to generate a spectrum vector. At the same time, the environmental pressure feature vector is input into the environmental parsing branch of the model. The feature alignment module is used to achieve fusion. Based on the aligned fusion features, the gradient clipping strategy is used to fine-tune the parameters of the base large model. When a high risk of environmental pressure is detected, the weights of the fear-related neurons in the call branch are strengthened. When the abnormal behavior fragment shows low stress, the response of the irrelevant frequency channel is weakened. Iterate until the accuracy of the base large model in recognizing the environment-related emotions of the test set is stabilized above the threshold, and the adjusted base large model is obtained.

[0115] 204. Constructing dog language reasoning logic based on the adjusted base large model, and using the dog language reasoning logic to adjust the language generation strategy of the translation instruction to generate a natural language description that matches the personality characteristics of the target pet;

[0116] In step 204, the dog language inference logic, a lightweight inference module that encapsulates fine-tuned large model parameters, converts real-time pet calls into natural language descriptions. The language generation strategy, based on a library of instruction templates constructed from a natural language processing model, adjusts the wording and information density of natural language descriptions based on the pet's personality tags, which include categories such as sensitive, exploratory, and stable. Natural language descriptions are text instructions generated based on the pet's personality traits, guiding users to take interactive actions tailored to the pet's behavioral characteristics.

[0117] In an embodiment of the present application, a parameter subset related to animal call analysis in the adjusted base model is derived, and the parameter subset is compressed into dog language reasoning logic through knowledge distillation technology. When the target pet makes a new call, the logic calls the personalized behavior pattern library, and inputs the personalized behavior matching the target pet in the personalized behavior pattern library into the pre-trained natural language generation model. The model uses the attention mechanism to screen the instruction templates that match the personality characteristics. For example, for sensitive pets, the description "A dense crowd is detected ahead, and the target pet is prone to anxiety. Please slowly guide it to the left for a detour" is generated, while for exploratory pets, the description "It is recommended to explore and detour to the left" is generated. The final output is a natural language instruction containing a personalized description.

[0118] 205. Based on the natural language description, combined with the body movement time series data and the noise decibels, identify the emotional state of the target pet, and add emotional soothing suggestions corresponding to the emotional state to the translation instruction;

[0119] In step 205, the emotional state is the classification result of the pet's instantaneous psychological state determined by multimodal data fusion, including labels such as fear, anxiety, and calmness. The limb movement time series data is the pet's joint movement parameters collected by inertial sensors during the current time period, such as the frequency of paw tremors and the angle of the ear back. The noise decibel is a quantified sequence of the ambient sound pressure level. Emotional comfort suggestions are a set of intervention strategies that match the emotional state, including voice comfort instructions (such as playing white noise of a specific frequency), physical guidance suggestions (such as gently pulling the leash), and other actionable instructions.

[0120] In an embodiment of the present application, keywords (such as "anxiety" and "rigidity") are extracted from natural language descriptions, and the micro-movement features in the current limb movement time series data are analyzed through a convolutional neural network, such as calculating the temporal correlation between the paw tremor frequency and the ear back angle. The frequency domain energy distribution of the noise decibel sequence is synchronously analyzed to identify the time coverage interval of sudden acoustic interference (such as car horns). The micro-movement features, acoustic features, and keyword vectors are input into the emotion classification model, and an emotion label (such as "high fear") is output. According to the emotion label, the instructions with the highest positive rate of historical interactive feedback are screened from the soothing strategy library. For example, the suggestion of "whispering the pet's name and patting the back" is matched to the "high fear" label, and it is appended to the end of the translation instruction.

[0121] 206. Push the translation instruction with the emotional comfort suggestion to the user terminal, and dynamically update the avoidance path prompt in the translation instruction according to the change of the high-density area in the obstacle distribution map.

[0122] In step 206, the dynamic update mechanism uses path correction logic based on changes in high-density areas in the obstacle distribution map to ensure that the avoidance path prompts are synchronized with the current environmental state. High-density areas are defined as the collection of spatial cells in the obstacle distribution map where the obstacle density or line of sight obstruction coefficient exceeds a preset threshold. The avoidance path prompts are topologically generated minimal risk routes that avoid high-density areas.

[0123] In an embodiment of the present application, after the translation instruction is pushed to the user terminal, the movement trajectory changes in the high-density areas in the obstacle distribution map are continuously monitored. For example, it is detected that the crowd is spreading to the northeast, causing the original path to be blocked. The conflict probability between the original path and the new obstacle distribution is calculated by the path replanning algorithm. If the conflict probability exceeds the dynamic threshold, an incremental update strategy is adopted to retain the main segment of the path and partially adjust the detour direction (such as changing "turn right and detour" to "cross the green belt diagonally in front of the right"). The updated path coordinates are reintegrated with the emotional comfort suggestions to generate incremental instructions and push them, such as "crowd movement is detected, please adjust to cross the green belt diagonally in front of the right, and continue to pat the pet's back."

[0124] Here's a specific example:

[0125] In the park trail scenario, when the target pet passes through a tourist gathering area, the environmental perception system collects crowd density, noise decibels, and lidar point clouds in real time. The large base model analyzes the environmental pressure feature vector and simultaneously identifies the dog's rapid barking as "high-frequency whimpering." Combined with historical behavior annotations, the current barking is jointly annotated. The dog language reasoning logic generates a natural language description based on fine-tuning parameters, and combines real-time limb data (tail clamped) and noise data to determine that it is highly fearful. The server generates soothing suggestions, and the initial avoidance path prompts "turn right and detour." When the lidar detects that the scooter suddenly turns, it obtains the new obstacle map through the application programming interface, immediately updates the instruction to "accelerate and go straight through", and pushes it to the owner's mobile terminal.

[0126] In summary, steps 201 to 206 establish a joint annotation system for environmental pressure and vocal emotions, ensuring that the fine-tuning process of the large base model has clear environmental relevance. Feature alignment technology is used to unify the semantic space of multimodal data, significantly improving the accuracy of emotion recognition in complex environments. Language generation strategies are adjusted based on personalized behavioral patterns to ensure that translation results match the pet's personality traits. Combining dynamic obstacle perception with real-time path updates forms a closed-loop decision-making process from emotion recognition to action guidance. Ultimately, full-link intelligent translation is achieved in outdoor scenarios, encompassing "accurate perception of environmental threats, precise translation of individual emotions, and dynamic optimization of avoidance paths."

[0127] To improve the accuracy of identifying environmental obstacles within the target pet's activity range, this solution uses three-dimensional spatial data to comprehensively assess the obstacles within the pet's activity area, including obstacle density and spatial occlusion, and generates an obstacle distribution map. In some embodiments, the obstacle distribution map in step 102 is constructed by quantifying the density of mobile obstacles and spatial occlusion coefficients within the target pet's activity range.

[0128] Generating an obstacle distribution map based on the three-dimensional spatial data includes:

[0129] 301. Divide the target pet's corresponding activity range into multiple spatial grid cells based on the three-dimensional spatial data, extract the movement trajectory of obstacles in each spatial grid cell and their frequency of occurrence within a preset time period, and calculate the moving obstacle density and spatial occlusion coefficient of each spatial grid cell based on the change in the degree of line of sight occlusion between adjacent spatial grid cells;

[0130] In step 301, the spatial grid unit is a uniform spatial block that divides the target pet's activity range into fixed-size blocks, which is used to quantify the spatial resolution of the obstacle distribution. The moving trajectory is a collection of continuous position coordinates of the obstacle within a preset time period, reflecting its movement direction and speed changes. The frequency of occurrence is the proportion of the number of times an obstacle is detected in a single spatial grid unit, which is used to characterize the distribution density of static obstacles. The degree of line of sight occlusion is the proportion of visibility reduction caused by the presence of obstacles between adjacent spatial grid units, which is calculated by the probability of the line of sight path being blocked. The density of mobile obstacles is the weighted value of the number of trajectories and the frequency of occurrence of dynamic obstacles (such as pedestrians and vehicles) in the spatial grid unit. The spatial occlusion coefficient is a unit-level oppression index generated by combining the degree of line of sight occlusion and the density of static obstacles.

[0131] In the embodiment of the present application, the park trail is divided into square grid units with a side length of one meter using lidar point cloud data. The frequency of occurrence of static obstacles such as benches and trees is counted for each unit, and the movement trajectory of pedestrians is tracked to calculate the number of trajectories. The line of sight tracking algorithm is used to calculate the line of sight occlusion probability between adjacent units. For example, the line of sight occlusion coefficient between the unit where the bench is located and the unit on the east side is 0.9. The density of mobile obstacles is calculated by combining the frequency of occurrence and the number of trajectories. For example, the density of the trail entrance unit is 0.7 (the maximum is 1).

[0132] 302. Map the mobile obstacle density and spatial occlusion coefficient of each spatial grid unit into a three-dimensional thermal map, and overlay them to generate the obstacle distribution map.

[0133] In step 302, the 3D heatmap is a visualization tool used to display the density of mobile obstacles and the intensity of spatial obstruction coefficients within different grid cells. The obstacle distribution map is generated by integrating this information to reflect the distribution of obstacles within the entire activity range.

[0134] In this embodiment, the calculated mobile obstacle density and spatial obstruction coefficient for each spatial grid cell are first converted into corresponding color values ​​to generate a three-dimensional thermal map. The thermal maps for all grid cells are then stacked to form a complete obstacle distribution map. This map not only intuitively displays the location and density of obstacles but also reveals the visual obstruction caused by the presence of obstacles, providing basic data support for subsequent environmental pressure assessments.

[0135] Here's a specific example:

[0136] In the park trail scene, the system divides the trail and the surrounding area into a one-meter square three-dimensional grid. In a certain grid unit at the entrance, a pedestrian trajectory density of 0.7, a bench appearance frequency of 0.9, a visual occlusion degree of 0.8, a calculated moving obstacle density of 0.78, and a spatial occlusion coefficient of 0.62 were detected. The adjacent green belt unit has no dynamic obstacles and clear vision, so the density and occlusion coefficients are 0.1 and 0.08 respectively. Through three-dimensional thermal layer mapping, the entrance unit is displayed as dark red superimposed on dark blue (high density and high occlusion), and the green belt unit is displayed as light red superimposed on light blue.

[0137] In summary, steps 301 to 302 accurately depict the distribution of environmental obstacles by quantitatively analyzing the density of mobile obstacles and the spatial occlusion coefficient within the target pet's activity range. By monitoring and analyzing the distribution of obstacles, the safety and comfort of pets and other visitors in public places are improved.

[0138] In order to improve the accuracy of identifying the environmental pressure characteristics of the target pet in a complex outdoor environment, in some embodiments, the step 102 of jointly encoding the multimodal data with the obstacle distribution map to generate an environmental pressure feature vector includes:

[0139] 401. Extract a parameter set from the multimodal data that overlaps with a high-density area in the obstacle distribution map based on a temporal trend of noise decibels and a spatial distribution change of crowd density. Perform time-series accumulation on parameters in the parameter set whose fluctuation amplitude exceeds a preset threshold to generate a parameter sequence reflecting a sudden environmental interference event.

[0140] In step 401, the parameter set is a noise decibel sequence and crowd density distribution data set that overlaps with high-density areas in the multimodal data. The fluctuation amplitude is the degree of deviation of the parameter value from the historical mean, measured by the multiple of the standard deviation. Time series accumulation is the energy integration and spatial superposition calculation within the time window of the parameters that exceed the fluctuation threshold. Sudden environmental interference events are complex interference scenarios triggered by abnormal fluctuations in multimodal parameters, such as crowds accompanied by high-decibel noise. The parameter sequence is a time-series feature set that records the start time, peak intensity, and spatial range of the sudden interference event.

[0141] In the embodiment of the present application, the noise decibel sequence and crowd density distribution data corresponding to the high-density area in the obstacle distribution map are extracted to calculate the fluctuation amplitude of each parameter. The interval where the fluctuation amplitude exceeds the preset threshold is detected by sliding the time window. The parameter value in each window is energy integrated (such as the frequency domain energy accumulation of noise decibels) and the spatial coverage range is calculated (such as the number of grids for crowd density diffusion). The windows that meet the conditions are marked as candidate interference events, and adjacent or overlapping windows are merged in chronological order to generate a parameter sequence and record its start time, peak intensity and impact range.

[0142] 402. Correlate and match the mobile obstacle density and spatial occlusion coefficient with a parameter sequence to determine a diffusion path of the sudden environmental interference event in the obstacle distribution map, and adjust the mobile obstacle density and spatial occlusion coefficient of each spatial grid unit according to the diffusion path.

[0143] In step 402, the diffusion path is the propagation direction and coverage of the sudden environmental interference event in the obstacle distribution map, determined by spatial correlation analysis of the parameter sequence. Correlation matching is to align the density of mobile obstacles and spatial occlusion coefficients with the parameter sequence in time and space to calculate the probability model of the event diffusion.

[0144] In the embodiment of the present application, the spatial range of the interference event recorded in the parameter sequence is aligned with the density distribution of mobile obstacles in the obstacle distribution map in time and space, and the event diffusion direction is calculated through a probabilistic graphical model. For example, based on the consistency between the crowd density diffusion trend and the direction of the mobile obstacle trajectory, the interference event is predicted to propagate along a specific path. The mobile obstacle density of the grid cells along the diffusion path is updated (such as increasing the weight of the number of trajectories) and the spatial occlusion coefficient.

[0145] 403. Based on the adjusted mobile obstacle density, spatial occlusion coefficient, and the time series accumulation result of the parameter sequence, the time series fluctuation characteristics of the multimodal data and the spatial topological characteristics of the obstacle distribution map are tensor-spliced, and the splicing result is encoded through a convolutional neural network to generate an environmental pressure feature vector.

[0146] In step 403, the time series accumulation result is a weighted aggregate value of the duration, peak intensity, and spatial coverage of the sudden interference event in the parameter sequence. Tensor concatenation is a technique for combining data of different dimensions. Convolutional neural networks are a type of deep learning model that excels at processing data with spatial topological structures. The environmental pressure feature vector is a composite feature generated by integrating the adjusted mobile obstacle density, spatial occlusion coefficient, and the time series accumulation result. It is used to characterize the comprehensive pressure intensity of the environment on the pet.

[0147] In this embodiment, we first use a tensor splicing method to combine multimodal data with temporal fluctuation characteristics and the spatial topological characteristics of the obstacle distribution map based on the adjusted mobile obstacle density, spatial occlusion coefficient, and time-series accumulation results of the parameter sequence. We then use a convolutional neural network to encode the spliced ​​data, ultimately generating a feature vector that comprehensively reflects the environmental pressure characteristics.

[0148] Here's a specific example:

[0149] In the park trail scenario, the system divides the activity area into one-meter-square grid cells. Upon detecting a surge in crowd density at the entrance cell, accompanied by a sudden horn honking, it generates a parameter sequence labeled "Entrance Sudden Disturbance Event." This event is predicted to spread northeastward along the trail, increasing the obstacle density and occlusion coefficient of the grid cells along the path. Finally, these adjusted parameters are integrated to generate an environmental pressure feature vector, which is used for subsequent stress level prediction.

[0150] In summary, steps 401 to 403 achieve refined quantification of environmental stress sources through dynamic association of spatial grid modeling and multimodal parameters. Based on the diffusion path prediction and parameter adjustment mechanism of sudden interference events, the accuracy of the obstacle distribution map is improved. The environmental pressure feature vector effectively integrates spatial occlusion, mobile obstacle density and sudden interference intensity, providing high-resolution input for stress level prediction. In the park trail scenario, the system accurately identifies the combined interference of crowds and noise at the entrance, dynamically adjusts the grid parameters along the path, and makes the environmental pressure assessment results highly consistent with the actual risk scenario, laying a data foundation for generating accurate obstacle avoidance paths.

[0151] In order to further improve the spatiotemporal correlation and environmental adaptability of the pet behavior abnormality segment extraction, in some embodiments, the limb movement time series data in step 103 includes a continuous change sequence of limb swing frequency, trunk contraction amplitude and head deflection angle;

[0152] The method of pattern matching the limb movement time series data with a preset pet body language database to extract abnormal behavior segments that are temporally and spatially associated with the outdoor complex environment includes:

[0153] 501. Divide the limb movement time series data into analysis windows of variable lengths, and dynamically adjust the sliding interval of the analysis windows according to the spatial density change of the obstacle distribution map in the outdoor complex environment;

[0154] In step 501, the analysis window is a variable-length interval that segments the limb movement time series data along the time dimension. Its sliding interval is dynamically adjusted based on the gradient direction of the spatial density change in the obstacle distribution map. Spatial density change is the rate of increase or decrease in the number of obstacles or movement trajectories per unit area in the obstacle distribution map, calculated by the difference in grid cell statistics between adjacent time slices. The dynamic sliding interval automatically adjusts the time span between the start and end points of the analysis window to reflect fluctuations in ambient pressure. For example, as high-density areas expand, the interval is shortened to improve detection sensitivity.

[0155] In this embodiment of the present application, the spatial density change rate of each grid cell in the obstacle distribution map is monitored, and the density gradient direction (e.g., diffusion from southeast to northwest) is calculated. The sliding interval of the analysis window is adjusted according to the gradient direction. If the density change rate exceeds the threshold, the interval is shortened to increase the sampling frequency; if the change is gentle, the interval is extended to reduce the computational load. The limb movement timing data (e.g., swing frequency, contraction amplitude) within each window is segmented and cached by timestamp.

[0156] 502. In each analysis window, based on the motion change characteristics at adjacent moments in the limb motion time series data, calculate the degree of matching between the limb motion parameters in the current analysis window and the standard behavior patterns in the pet body language database, and combine the parameters related to visual interference in the environmental pressure feature vector to generate dynamic matching conditions;

[0157] In step 502, the motion change feature is a set of differential values ​​of limb motion parameters at adjacent moments within the analysis window, such as the increment in swing frequency or the rate of change in contraction amplitude. The degree of match is a quantitative indicator of the similarity between the motion sequence in the current window and the standard behavioral patterns in the pet body language database, calculated using a trajectory alignment algorithm. The dynamic matching condition is a similarity threshold generated by combining visual interference parameters (such as obstacle density and line of sight occlusion coefficient) in the environmental pressure feature vector, which is used to dynamically adjust the matching benchmark.

[0158] In the embodiment of the present application, for each action sequence in the analysis window, a dynamic time warping algorithm is used to calculate its trajectory similarity with the template in the standard behavior pattern library. The matching threshold is dynamically adjusted according to the visual interference parameters (such as high occlusion coefficient) in the environmental pressure feature vector. When the visual interference increases, the similarity threshold is lowered to capture more subtle action anomalies. For example, if the occlusion coefficient exceeds 0.7, the matching threshold is lowered from 0.65 to 0.55.

[0159] 503. Filter the analysis windows whose matching degree is lower than a preset benchmark according to the dynamic matching condition, extract the periodic differences between the limb swing frequency and the trunk contraction amplitude within the analysis window, and mark the abnormal behavior segments that are associated with the location of the complex outdoor environment in combination with the location information of the high-density areas in the obstacle distribution map;

[0160] In step 503, periodic differences are characterized by involuntary rhythmic fluctuations in limb swing frequency and trunk contraction amplitude within the analysis window, such as high-frequency, irregular jitters. Position information is a set of spatial coordinates of high-density regions within the obstacle distribution map, used to correlate the locations of abnormal behavior segments. Abnormal behavior segments are action sequences that simultaneously meet the requirements of low matching, significant periodic differences, and spatial overlap with high-density regions.

[0161] In this embodiment of the present application, analysis windows with a matching degree below the dynamic threshold are screened to extract the temporal fluctuation characteristics of the limb swing frequency and trunk contraction amplitude. Periodic components are detected through Fourier transform. If the fundamental frequency deviates from the normal range and the harmonic energy is dispersed, it is determined that the periodic difference is significant. The timestamps of the abnormal windows are matched with the spatial coordinates of the high-density areas in the obstacle distribution map, and segments with temporal and spatial overlap are screened and marked as position-related anomalies.

[0162] 504. Correlate and compare the abnormal behavior segments with the spatial distribution changes of the crowd density in the multimodal data, exclude non-stress action data generated by the pet's autonomous behavior, and generate the spatiotemporally correlated abnormal behavior segment set.

[0163] In step 504, the non-stress action data is a sequence of actions generated by the pet's autonomous behavior (e.g., chasing flying insects or sniffing the ground), where the changes in crowd density distribution do not overlap with the behavior segments in time and space. The spatiotemporally correlated abnormal behavior segments are the remaining abnormal segments after excluding the non-stress data, and they meet the strong correlation between abnormal behavior and sudden changes in crowd density in terms of temporal synchronization and spatial coverage.

[0164] In this embodiment, the marked location-correlated abnormal segments are spatiotemporally compared with the crowd density distribution in the multimodal data. If the crowd density in the grid cell where the abnormal segment occurs does not change significantly during the period of time, it is determined to be an autonomous action of the pet and is removed. Finally, the abnormal segments that spatiotemporally overlap with the area with increased crowd density are retained to generate a collection of spatiotemporally correlated abnormal behavior segments.

[0165] Here's a specific example:

[0166] In the park trail scenario, the system detected a sudden increase in obstacle density at the entrance and shortened the analysis window interval to 3 seconds. Within a certain window, the pet's trunk contraction amplitude fluctuated irregularly, with a similarity of 0.55 to a "normal walking" pattern, which was below the dynamic threshold of 0.6. The system extracted the sudden increase in the swing frequency in this window and, combined with the high-density coordinates at the entrance, marked it as an anomalous segment. After comparing it with crowd density data, similar segments away from crowd areas were excluded to generate the final anomalous set.

[0167] In summary, steps 501 to 504 synchronize behavioral anomaly detection with environmental pressure fluctuations by dynamically adjusting the analysis window and matching threshold, reducing missed detections or false alarms caused by fixed parameters. Combined with the correlation screening of spatial coordinates in high-density areas, stress behaviors and autonomous movements can be effectively distinguished. In the park trail scenario, the system accurately captures the abnormal contraction and shaking of pets when there are dense crowds at the entrance, eliminates the interference of running in open areas, provides high-confidence input for subsequent stress level predictions, and significantly improves the spatiotemporal correlation and environmental adaptability of behavioral analysis.

[0168] To improve the accuracy of identifying the emotional state of a pet and the adaptability of soothing strategies, in some embodiments, the steps of step 202 include identifying the emotional state of the target pet based on the natural language description, combined with the body movement timing data and the noise decibel level, and adding emotional soothing suggestions corresponding to the emotional state to the translation instruction, including:

[0169] 601. Perform word segmentation processing on the natural language description, extract context-related features, and generate a sentiment feature vector, wherein the sentiment feature vector includes word segmentation weight distribution and semantic dependency strength;

[0170] In step 601, the natural language description refers to information obtained through speech or text. Word segmentation is the process of breaking continuous natural language text into individual words. Contextual features refer to the semantic relationships between words. The sentiment feature vector is a mathematical representation of sentiment-related information extracted from the natural language description, such as the distribution of word segmentation weights and the strength of semantic dependencies.

[0171] In this embodiment, natural language processing techniques are first applied to segment natural language descriptions, and the importance weight of each word is calculated using methods such as word frequency and inverse document frequency. Next, a sentiment feature vector is generated by analyzing the sentence structure and the relationship between words. This vector not only reflects the emotional tendency in the text, but also includes the degree of semantic dependence between each word.

[0172] 602. Establish a two-dimensional feature space with body movement amplitude and movement frequency as axes, construct a dynamic trajectory point set based on the emotional feature vector and the target pet's body movement time series data within a set time window, and divide the emotional state into emotional intensity levels according to the distribution density of the dynamic trajectory point set in the two-dimensional feature space;

[0173] In step 602, the two-dimensional feature space is a quantitative analysis space constructed with the limb movement amplitude and movement frequency as coordinate axes. The limb movement amplitude is the maximum displacement of the joint movement, the movement frequency is the number of repetitions of the periodic limb movement per unit time, the dynamic trajectory point set is a set of time series data points of the target pet's movement amplitude and frequency within a set time window, the distribution density is the number of points in a unit area of ​​the trajectory point set in the two-dimensional space, and the emotion intensity level is a quantitative indicator of the pet's emotions divided by the distribution density.

[0174] In an embodiment of the present application, an inertial sensor worn on the pet's trunk collects limb movement time series data, extracts the trunk contraction amplitude and limb swing frequency at each time point, and fuses the emotional feature vector, trunk contraction amplitude, and limb swing frequency to construct a dynamic trajectory point set. This is then mapped to a two-dimensional coordinate system to form a dynamic trajectory point set. A density-based spatial clustering algorithm is used to calculate the distribution density of trajectory points in different regions. The density threshold is adjusted based on the part-of-speech weight distribution of the emotional feature vector to divide the area into low-density, medium-density, and high-density areas. A density threshold is set based on historical data to classify the emotional state into different levels of emotional intensity.

[0175] 603. Extract the voiceprint pulse intervals whose duration exceeds the threshold in the noise decibel sequence, perform weighted coupling between the energy integral value of each voiceprint pulse interval and the motion frequency of the dynamic trajectory point set, and generate an emotional state determination coefficient;

[0176] In step 603, the voiceprint pulse interval is a continuous high-energy period in the noise decibel sequence whose duration exceeds a preset threshold. The energy integral value is the sum of the energy accumulated over time in the sound pressure level within the voiceprint pulse interval. The weighted coupling is the calculation process of fusing the voiceprint energy integral value with the motion frequency of the dynamic trajectory point set according to the weight coefficient. The emotional state determination coefficient is a comprehensive indicator that quantifies the correlation between acoustic interference and motion characteristics.

[0177] In this embodiment, a sliding window detection is performed on the noise decibel sequence to identify periods of time when the decibel level continuously exceeds a threshold. The voiceprint signal for this period is extracted and subjected to a fast Fourier transform to calculate the frequency domain energy integral. Simultaneously, the average motion frequency of the dynamic trajectory points within the same time period is extracted. The voiceprint energy integral and the motion frequency are linearly weighted according to a preset weight coefficient to generate an emotional state determination coefficient.

[0178] 604. Constructing a bidirectional mapping table between emotion labels and body movement patterns based on the frequency domain harmonic components of the voiceprint pulse interval. When the emotion state determination coefficient exceeds the dynamic baseline, using the context-related features to correct the confidence level of the emotion label corresponding to the current body movement pattern.

[0179] In step 604, the frequency domain harmonic component is the energy distribution feature of the integer multiple fundamental frequency in the voiceprint pulse spectrum extracted by Fourier transform, the bidirectional mapping table is a lookup table that records the correspondence between the harmonic component and the pet's limb movement pattern, the emotion label is the classification result of the pet's emotional state, and the confidence level is the credibility of the current emotion label.

[0180] In the embodiment of the present application, harmonic analysis is performed on the spectrum of the voiceprint pulse interval to extract the three harmonic components with the highest energy proportion and their frequency values. A pre-built two-way mapping table is queried to match the harmonic components with the body movement patterns recorded in the historical data (such as high-frequency harmonics associated with the ear-back movement). When the emotional state determination coefficient exceeds the dynamic baseline, the confidence level of the emotional label is adjusted based on the context-related features extracted from the natural language description and the harmonic energy distribution ratio. For example, when the energy proportion of a certain harmonic exceeds half, the confidence level is increased by 0.1.

[0181] 605. When the confidence of the corrected emotion tag exceeds a preset threshold, an interactive instruction template that is positively correlated with the emotion intensity level is matched from the emotion soothing strategy library, and the interactive instruction template is semantically aligned with the context of the translation instruction to output a translation instruction with additional emotion soothing suggestions.

[0182] In step 605, the emotional soothing suggestions are a set of intervention strategies that match the corrected emotion labels. The additional process involves logically integrating the suggested text with the path hints in the translated instructions. The emotional soothing strategy library is a set of predefined interactive instruction templates designed to help alleviate negative emotions in pets. Contextual alignment involves matching the content of the interactive instruction templates with the context of the translated instructions to ensure the relevance and applicability of the suggestions.

[0183] In an embodiment of the present application, once the confidence level of the corrected emotion tag reaches a preset threshold, the system selects the interactive instruction template with the highest positive rate of historical interactive feedback from the emotional soothing strategy library, semantically aligns it with the current context information, and outputs a suggested text with emotional soothing suggestions attached. The suggested text is inserted into the specified location of the translation instruction, such as appending a soothing action description after the path prompt to ensure semantic coherence. For example, the suggestion of "whispering the pet's name and keeping the leash loose" is matched to the "fear" tag and appended to the end of the instruction.

[0184] Here's a specific example:

[0185] In a park trail scenario, the system detected a sudden increase in the pet's trunk contraction from 5cm to 15cm in a crowded area, and a decrease in movement frequency from 5 to 2 times per second, generating a high-density trajectory point set. The system also simultaneously extracted the energy integral of the continuous whistle soundprint pulses, using a weighted determination coefficient of 0.75. Spectral analysis of the soundprint revealed a dominant 500Hz harmonic. A bidirectional mapping table associated the "fear" label and adjusted the confidence level from 0.7 to 0.8. Finally, a translated command was generated, such as "A crowd has been detected ahead. A left turn is recommended. Please gently soothe the pet and loosen the leash."

[0186] In summary, steps 601 to 605 achieve a refined classification of emotional intensity levels by quantifying body movement features in a two-dimensional feature space and combining this with a weighted coupling analysis of voiceprint pulse energy and motion frequency. A dynamic confidence correction mechanism based on frequency domain harmonic components and a bidirectional mapping table improves the reliability of emotion labels. The system accurately identifies pets' fear states caused by complex interference and generates instructions that integrate obstacle avoidance paths with personalized soothing strategies. This significantly enhances the efficiency and effectiveness of human-pet interactions, ultimately reducing the risk of pets experiencing stress and losing control, and improving the safety of outdoor activities.

[0187] In order to solve the problem of accuracy and timing correlation in detecting sudden environmental interference events, in some embodiments, the parameters in the parameter set whose fluctuation amplitude exceeds a preset threshold are accumulated in time series in step 401 to generate a parameter sequence reflecting the sudden environmental interference event, including:

[0188] 701. Establish a dynamic fluctuation threshold in time slices. Based on the historical change trend of noise decibels over time and the historical spatial distribution of crowd density in the parameter set, set the range exceeding two standard deviations of the median of the historical change trend or historical distribution change as the initial trigger boundary.

[0189] In step 701, the time slice is to divide the continuous time axis into analysis units of fixed length, which is used to discretize the time series data. The dynamic fluctuation threshold is the parameter anomaly judgment boundary that changes with the time slice, and is dynamically adjusted according to historical data to adapt to environmental changes. The historical change trend of noise decibels over time refers to the statistical distribution range within the historical time period, including statistics such as maximum value, minimum value and median. Two times the standard deviation of the median is the statistical fluctuation range calculated based on historical data, which is used to eliminate occasional noise interference. The initial trigger boundary is the parameter value range that exceeds two times the standard deviation of the median, which serves as the starting condition for interference event detection. The historical distribution change of crowd density in space refers to the change in the number of people in a certain area in the past.

[0190] In the embodiment of the present application, data on noise levels and crowd density in the environment are first collected over a period of time. Statistical methods are then used to calculate the historical trends and distributions of these parameters, particularly the median and standard deviation. Next, a dynamic fluctuation threshold, expressed in time slices, is set based on the above calculation results, and the range exceeding two standard deviations of the median of the historical trend or distribution change is set as the initial trigger boundary. When the monitored data exceeds this boundary, a significant fluctuation is considered to have occurred.

[0191] 702. Calculate the parameter value fluctuation difference of adjacent time slices. When the difference direction of three consecutive time slices is consistent and the fluctuation amplitude exceeds the initial trigger boundary, lock the current time slice to the interference event end mark.

[0192] In step 702, parameter value fluctuation differences are the increase or decrease in the value of the same parameter within adjacent time slices, reflecting the instantaneous intensity of the parameter change. Difference direction consistency refers to the consistent direction of parameter changes across multiple consecutive time slices, such as a continuous increase or decrease. The interference event end marker is the point in time when the parameter value falls below the trigger boundary, defining the event's time span.

[0193] In the embodiment of the present application, the parameter difference values ​​of adjacent time slices are calculated in time slice order, and the difference direction of three consecutive time slices is detected to see if it is consistent and the amplitude exceeds the initial trigger boundary. If the conditions are met, the event start time slice is locked, and continuous monitoring is performed until the parameter value falls below the boundary, and the end time slice is recorded to determine the event period. For example, if the noise decibel values ​​of three consecutive time slices (5 seconds each) are 75 decibels, 85 decibels, and 95 decibels respectively, and the difference direction is positive and exceeds the trigger boundary (80 decibels), then it is determined that the conditions are met.

[0194] 703. Based on the locked time slice, extract the frequency domain energy surge interval of the noise decibel and the sudden change direction of the crowd density spatial gradient, and segment the overlapping areas in the time dimension into independent interference segments, and accumulate interference segments with temporal continuity or overlapping spatial coverage.

[0195] In step 703, the frequency domain energy surge interval is a set of frequency bands in the acoustic signal extracted by spectral analysis whose energy is significantly higher than the fundamental frequency, reflecting a specific type of acoustic interference. The spatial gradient of crowd density is the rate and direction of change of crowd density within a unit distance, characterizing the diffusion characteristics of crowd movement. The time dimension overlapping area is the interval where the frequency domain energy surge and the crowd density mutation completely or partially overlap on the time axis. The independent interference segment is the minimum continuous time period separated from the overlapping area, and has the characteristics of temporal continuity or spatial coverage overlap.

[0196] In the embodiment of the present application, a fast Fourier transform is performed on the locked time slice to extract the frequency band and time period of the frequency domain energy surge. The gradient change direction and rate of the crowd density in space are calculated synchronously, the acoustic surge period and the density mutation period are aligned on the time axis, and the interval where the two completely overlap is intercepted as an independent interference segment. For example, if the acoustic surge period is 10:10-10:30 and the crowd diffusion period is 10:15-10:40, the overlapping interval of 10:15-10:30 is intercepted as an independent interference segment (duration 15 minutes).

[0197] 704. Perform parameter weighting on the accumulated interference segments, wherein the duration of the frequency domain energy sudden increase interval and the spatial diffusion speed of the crowd density spatial gradient are used as weight factors to generate a parameter sequence reflecting the sudden interference event in the environment.

[0198] In step 704, the duration weighting factor is the importance of the total time span of the frequency domain energy surge interval in the weighted calculation, reflecting the impact strength of the temporal dimension. The spatial diffusion rate weighting factor is the importance of the rate of change in the crowd density gradient direction in the weighted calculation, reflecting the impact range of the spatial dimension. The parameter sequence is a time-series feature set that records the onset time, peak intensity, and spatial range of the interference event and is used to quantify the overall intensity of the event.

[0199] In an embodiment of the present application, weight coefficients are assigned to duration and spatial diffusion speed according to preset rules, for example, the weight of duration is 0.6, and the weight of diffusion speed is 0.4. The duration and spatial diffusion speed of the independent interference segment are calculated to generate a weighted value of 9.2. The event start time, weighted peak intensity (9.2) and spatial impact range (50 meters eastward diffusion) are encoded as a structured parameter sequence, for example, "start time 10:15, peak intensity 9.2, impact range 50 meters eastward". The parameter sequence is associated with the environmental pressure characteristic vector and stored as input to the subsequent stress level prediction model.

[0200] Here's a specific example:

[0201] In the park trail scenario, based on historical data, the noise decibel trigger boundary is set at 40 decibels to 80 decibels, and the crowd density boundary is set at 1 person / square meter to 3 people / square meter. When the obstacle density in a certain area is detected to increase from 0.3 to 0.8 within 10 seconds, the time slice is shortened to 5 seconds to improve the detection sensitivity. The overlapping interval between the acoustic surge period and the crowd diffusion period is intercepted as an independent interference segment. The weighted value is calculated based on the duration weight of 0.6 and the diffusion speed weight of 0.4 to generate the parameter sequence "peak intensity 9.2, impact range 50 meters eastward" for subsequent path planning and stress level prediction.

[0202] In summary, steps 701 to 704 accurately capture real interference events and filter out sporadic noise through dynamic thresholds and continuous triggering. Based on frequency and spatial multidimensional feature segmentation and weighted aggregation, the intensity quantification and spatiotemporal characterization of complex interference events are achieved. By correlating acoustic spikes with crowd diffusion, a highly discriminative parameter sequence is generated, providing reliable input for environmental pressure feature fusion, significantly improving the accuracy of stress level prediction and path planning, and ultimately enhancing the overall effectiveness of pet behavior management in complex environments.

[0203] In order to solve the problem of decreased matching accuracy caused by dynamic environmental interference in the pet behavior recognition system, in some embodiments, the step 502 of calculating the degree of matching between the limb movement parameters in the current analysis window and the standard behavior patterns in the pet body language database based on the movement change characteristics of adjacent moments in the limb movement time series data includes:

[0204] 801. Decompose the continuous change sequence of limb swing frequency, trunk contraction amplitude, and head deflection angle in the limb movement time series data into motion trajectories of multiple joints, and extract the motion change features of each joint at adjacent moments;

[0205] In step 801, the motion trajectory refers to converting continuous limb motion data into a displacement sequence of multiple joints. Limb swing frequency is the number of periodic limb movements per unit time. Trunk contraction amplitude represents the extreme change in trunk length. Head deflection angle describes the three-axis rotation angle of the head coordinate system relative to the trunk coordinate system. Joints refer to the motion connection points with degrees of freedom in the pet's three-dimensional skeletal model. Continuous change sequences are a sequence of continuous data points of limb motion parameters arranged in chronological order, representing the changing trends of these parameters over time.

[0206] In this embodiment, a 3D pet motion model is first established using skeletal rigging technology. The raw data collected by the accelerometer and gyroscope are mapped to twelve key joints. A sliding window method is used to segment the continuously changing sequence in the time domain. A first-order difference algorithm is used to calculate the displacement vectors of each joint at adjacent moments. Coordinate transformation is then used to eliminate errors caused by differences in the installation positions of the inertial measurement units. Finally, the motion change characteristics of each joint at adjacent moments are output.

[0207] 802. Based on the gradient direction of the spatial density change in the obstacle distribution map, dynamically adjust the weight distribution of the joint points of the motion trajectory decomposition, and calculate the offset of the joint points of the limb parts corresponding to the high-density area;

[0208] In step 802, the obstacle distribution map is a two-dimensional rasterized environmental feature map constructed using lidar. The spatial density gradient represents the rate of change in the number of obstacles per unit distance. Joint weight assignment involves adjusting the calculation coefficient based on the degree of influence of the environmental region on limb movement. The offset reflects the coefficient of the degree to which the joint's motion trajectory is affected by the environment. The gradient direction refers to the direction of change in obstacle density in space, which is the vector direction from high-density areas to low-density areas.

[0209] In this embodiment, a simultaneous positioning and mapping technique is used to update the obstacle distribution map and calculate the obstacle density gradient vector for the area corresponding to the current window. Based on the angle between the gradient direction and the pet's movement direction, the weight coefficients of the limbs in front of the movement direction (such as the front leg joints) are dynamically adjusted, and the offset of the limb joint points is calculated. When a high-density obstacle area is detected ahead, the weight of the forelimb joints is increased to 2 times the baseline value, and the weight of the hindlimb joints is reduced to 0.7 times the baseline value. This enhances environmental adaptability through a weighted attention mechanism.

[0210] 803. Based on the offset calculation result, the cosine value of the angle between the motion change feature and the reference trajectory corresponding to the standard behavior pattern is counted in the analysis window. The cosine value of the angle is dynamically thresholded based on the temporal distribution characteristics of the visual interference parameter in the environmental pressure feature vector.

[0211] In step 803, the baseline trajectory is the theoretical motion path of the standard behavior pattern in three-dimensional space, and the cosine of the angle represents the directional similarity between the actual motion trajectory and the standard trajectory. Visual interference parameters include environmental factors that affect visual feature extraction, such as the rate of change of light intensity and the degree of motion blur. The analysis window is a fixed-duration data slice used for local feature extraction, with its length determined by the minimum period of the behavior pattern. The temporal distribution characteristic refers to the variation pattern of the visual interference parameters over time.

[0212] In this embodiment, a dynamic time warping algorithm is used to align the time axes of the actual trajectory with the reference trajectory, calculating the motion direction vector of each joint point in each frame of data. A cosine similarity algorithm is used to calculate the cosine value of the angle between vectors. A time-varying threshold function is constructed by combining the current illumination sensor data and the image clarity score of the visual interference parameter. When the illumination suddenly changes beyond the threshold, anomalies caused by overexposure or shadows are eliminated through adaptive filtering.

[0213] 804. The truncated cosine values ​​of the included angle are weighted and summed according to the joint weights to generate a local matching degree. The local matching degree is phase-compensated based on the periodic difference between the limb swing frequency in the analysis window and the standard behavior pattern in the pet body language database to obtain the matching degree of the current analysis window.

[0214] In step 804, phase compensation involves time-domain correction of the matching degree based on motion frequency differences. Periodic differences are calculated by extracting the fundamental frequency component through Fourier transform. Weighted summation involves linearly superimposing local similarities based on joint importance coefficients. Standard behavior patterns refer to the three-dimensional motion trajectory and frequency feature templates of typical pet behaviors pre-stored in the database.

[0215] In this embodiment, a weighted summation formula is used to aggregate the corrected similarities of the 12 joint points, with the weight coefficients derived from the dynamic allocation results in step 802. A fast Fourier transform is performed on the limb swing frequency in the current window, and a cross-correlation operation is performed with the standard frequency in the database to obtain a phase offset. Linear interpolation is used to phase-align the local matching degrees, ultimately outputting a standardized matching degree between 0 and 1.

[0216] Here's a specific example:

[0217] On a park trail, a pet encountered alternating terrain between cobblestones and lawns during a morning run. When entering a high-density obstacle zone on the cobblestones, the system detected an automatic increase in the weight of the forelimb joints. The pet's front legs raised to avoid the gravel, resulting in a cosine value of 0.7 matching the forelimb trajectory in the "alert walk" mode in the database. However, due to the swaying shadows of trees, the trigger threshold for the visual interference parameter was lowered to 0.65. After phase compensation, the final match reached 0.85, accurately identifying alert behavior. On the lawn, the detected swaying frequency had a phase difference of 0.3 compared to the "joyful run" mode; after compensation, the match improved to 0.92.

[0218] In summary, steps 801 to 804 significantly improve the accuracy of behavior recognition in complex environments through multi-dimensional feature fusion and dynamic parameter adjustment. First, the joint trajectory decomposition combined with the weight distribution mechanism of environmental perception effectively highlights key motion features affected by terrain. Second, phase compensation technology eliminates systematic errors caused by differences in movement rhythm, ensuring recognition consistency at different movement speeds. At the same time, the weight distribution and threshold adjustment mechanism supports feature adaptation for pets of different body shapes and movement habits, significantly improving generalization capabilities to meet the interaction needs in complex dynamic scenes.

[0219] In order to further improve the accuracy of predicting the stress level of a pet in a complex outdoor environment, in some embodiments, determining the predicted result of the stress level of the target pet in the complex outdoor environment by mapping the association between the environmental pressure feature vector and the abnormal behavior segment in step 104 includes:

[0220] 901. Constructing a spatiotemporal alignment relationship between the environmental pressure feature vector and the abnormal behavior segment, and extracting a joint feature set of the environmental pressure feature vector and the abnormal behavior segment within an overlapping time window;

[0221] In step 901, the environmental stress feature vector contains various physical and chemical parameters of the target pet's environment that may trigger a stress response, such as temperature, humidity, noise level, and air quality. Abnormal behavioral segments refer to time periods that show significant deviations from normal behavior, as identified by analyzing the pet's behavioral patterns. Spatiotemporal alignment involves chronologically matching these environmental data with abnormal behavioral segments. Overlapping time windows select time periods where both environmental changes and abnormal behaviors occur for more precise analysis.

[0222] In this embodiment, data from various sensors is first collected to construct an environmental pressure feature vector. Video analysis techniques are then used to identify and extract segments of abnormal pet behavior. An algorithm then determines the overlap between these two segments within the same time window and extracts a joint feature set from these segments. These features may include, but are not limited to, the rate of change of environmental factors and the responsiveness of the pet's behavior. The result is a feature set that reflects the interaction between the environment and the pet's behavior within a specific time period, such as a sudden change in temperature and the pet's agitated movements during a specific period.

[0223] 902. Perform nonlinear correlation analysis on the joint feature set, capture the coupling relationship between the environmental pressure feature vector and the abnormal behavior segment through a multi-layer feature interaction network, and generate a coupling strength coefficient representing the strength of the correlation between the environment and the behavior;

[0224] In step 902, a joint feature set is formed from the data obtained in the previous step, revealing the potential connection between environmental factors and pet behavior. Nonlinear association analysis is a statistical method used to explore complex interactions between variables. The multi-layer feature interaction network is a deep learning architecture specifically designed to handle deep interactions between high-dimensional data. The coupling strength coefficient is a quantitative indicator generated during this process, which is used to indicate the close relationship between environmental factors and pet behavior.

[0225] In this embodiment, a pre-trained multi-layer feature interaction network is used to process the joint feature set. This network adjusts its internal parameters by learning from a large amount of historical data, thereby accurately calculating the coupling strength coefficient. This process involves adjusting the network weights to minimize prediction error and assessing the impact of environmental changes on pet behavior based on the output coupling strength coefficient.

[0226] 903. Divide the target pet into stress level threshold intervals in the outdoor complex environment based on the numerical distribution range of the coupling strength coefficient;

[0227] In step 903, the numerical distribution range of the coupling strength coefficient represents the value interval of all calculated coupling strength coefficients. The stress level threshold interval is set based on this numerical distribution, with the purpose of distinguishing different degrees of stress states based on different coupling strength coefficients.

[0228] In the embodiment of the present application, all the collected coupling strength coefficients are first statistically analyzed to understand their distribution characteristics. By applying a clustering algorithm, the data is divided into different groups according to the numerical value of the coupling strength coefficient. Each group represents a different degree of stress response possibility. Then, based on the results of these groupings, a specific stress level threshold interval is set. For example, a lower coupling strength coefficient may correspond to a low stress level, while a higher coefficient indicates a high stress level.

[0229] 904. Dynamically assign a confidence weight of the stress level according to the location of the coupling strength coefficient within the stress level threshold range, and modify the confidence weight based on the same environmental pressure in historical stress events to obtain a stress level prediction result.

[0230] In step 904, the confidence weight is a parameter that measures the confidence level that the current coupling strength coefficient falls within a certain stress level range. The same environmental pressure in historical stress events means that the current confidence weight is adjusted based on past pet stress cases that occurred under similar circumstances.

[0231] In the embodiment of the present application, the current coupling strength coefficient is first determined based on the stress level threshold interval obtained in the previous step, and a preliminary confidence weight is assigned accordingly. The system then searches the historical database for records of stress events that have occurred in similar environments in the past. By analyzing these records, the similarity between the current situation and historical cases is evaluated, and the initially assigned confidence weight is adjusted accordingly. Finally, the weighted average value obtained after comprehensively considering the above factors is the final stress level prediction result.

[0232] Here's a specific example:

[0233] In the park trail scenario, the environmental stress feature vector shows a noise decibel gradient of 25 dB / s between 10:10 and 10:25, an obstacle density change rate of 0.5 / s, a limb swing frequency offset of -3 Hz in the behavioral anomaly segment, and a 6 cm difference in trunk contraction amplitude. A multi-layer feature interaction network generated a coupling strength coefficient of 0.65. Based on the historical threshold range, the stress level was classified as medium (0.3-0.7). The confidence weight was adjusted to 0.78 based on the historical co-occurrence frequency of 80%, and the output was "medium stress level, confidence level 0.78."

[0234] In summary, steps 901 to 904 precisely quantify the coupled relationship between environmental stress and behavioral abnormalities through spatiotemporal alignment and nonlinear correlation analysis, avoiding the limitations of single-parameter analysis. Dynamic thresholding and confidence correction mechanisms, combined with historical data, optimize the reliability of prediction results. The system effectively identifies the combined effects of noise and obstacle density on pets' contraction behavior, outputting a medium stress level and a high confidence weight, providing high-precision input for the subsequent generation of personalized routing prompts and soothing strategies.

[0235] Figure 2 The present invention provides a structural diagram of a pet status intelligent translation system based on multimodal data, such as Figure 2 As shown, the system includes:

[0236] An acquisition module 21 acquires multimodal data of the complex outdoor environment in which the target pet is located, wherein the multimodal data includes noise decibels, crowd density, and three-dimensional spatial data captured by a laser radar;

[0237] An encoding module 22 generates an obstacle distribution map based on the three-dimensional spatial data, and jointly encodes the multimodal data and the obstacle distribution map to generate an environmental pressure feature vector;

[0238] The extraction module 23 synchronously collects the target pet's body movement time series data, and performs pattern matching on the body movement time series data with a preset pet body language database to extract abnormal behavior segments that are temporally and spatially associated with the outdoor complex environment;

[0239] A mapping module 24 determines a predicted result of the stress level of the target pet in the outdoor complex environment by mapping the association between the environmental pressure feature vector and the abnormal behavior segment;

[0240] The generating module 25 generates a translation instruction according to the stress level prediction result, wherein the translation instruction includes an avoidance path prompt for the high-density area in the obstacle distribution map.

[0241] Figure 2 The pet status intelligent translation system based on multimodal data can be executed Figure 1 The implementation principles and technical effects of the multimodal data-based intelligent translation method for pet status described in the illustrated embodiment will not be elaborated upon. The specific manner in which each module and unit performs operations in the multimodal data-based intelligent translation system for pet status described in the aforementioned embodiment has been described in detail in the related embodiments and will not be elaborated upon here.

[0242] In one possible design, Figure 2The pet status intelligent translation system based on multimodal data of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0243] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0244] The processing component 32 is used for the above Figure 1 The embodiment provides a pet status intelligent translation method based on multimodal data.

[0245] The processing component 32 may include one or more processors to execute computer instructions to perform all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0246] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0247] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0248] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0249] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0250] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0251] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment provides a pet status intelligent translation method based on multimodal data.

[0252] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0253] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0254] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0255] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A pet status intelligent translation method based on multimodal data, characterized in that: include: Acquire multimodal data of the target pet's complex outdoor environment, including noise decibels, crowd density, and three-dimensional spatial data captured by a LiDAR sensor. generating an obstacle distribution map based on the three-dimensional spatial data, and jointly encoding the multimodal data and the obstacle distribution map to generate an environmental pressure feature vector; Synchronously collecting the target pet's body movement time series data, and performing pattern matching on the body movement time series data with a preset pet body language database to extract abnormal behavior segments that are temporally and spatially associated with the complex outdoor environment; Determining a predicted result of the stress level of the target pet in the outdoor complex environment by mapping the environmental pressure feature vector with the abnormal behavior segment; A translation instruction is generated according to the stress level prediction result, wherein the translation instruction includes an avoidance path prompt for a high-density area in the obstacle distribution map.

2. The pet status intelligent translation method based on multimodal data according to claim 1 is characterized in that: After generating the translation instruction according to the stress level prediction result, the method further includes: Building a large base model based on multimodal data, the large base model includes cross-modal analysis results of human speech, environmental sounds, and animal calls; Jointly labeling the cross-modal parsing results of the animal calls according to the environmental stress feature vector and the abnormal behavior segment to obtain a labeling result, wherein the labeling result includes a mapping relationship between dog breed category, emotion type and environmental stress; Extracting Mel-spectrum features of the annotation result using the large base model to obtain a spectrum vector, aligning the spectrum vector with the environmental pressure feature vector, and adjusting parameters of the large base model based on the alignment result to obtain an adjusted large base model; Constructing dog language reasoning logic based on the adjusted base large model, and using the dog language reasoning logic to adjust the language generation strategy of the translation instruction to generate a natural language description that matches the personality characteristics of the target pet; Based on the natural language description, combined with the body movement timing data and the noise decibel level, the emotional state of the target pet is identified, and emotional soothing suggestions corresponding to the emotional state are added to the translation instruction; The translation instruction with the emotional comfort suggestion is pushed to the user terminal, and the avoidance path prompt in the translation instruction is dynamically updated according to the changes in the high-density area in the obstacle distribution map.

3. The method according to claim 1, characterized in that The obstacle distribution map is constructed by quantifying the density of mobile obstacles and spatial occlusion coefficients within the target pet's activity range; Generating an obstacle distribution map based on the three-dimensional spatial data includes: Dividing the target pet's corresponding activity range into multiple spatial grid cells based on the three-dimensional spatial data, extracting the movement trajectory of obstacles in each spatial grid cell and their frequency of occurrence within a preset time period, and calculating the moving obstacle density and spatial occlusion coefficient of each spatial grid cell based on the change in the degree of line of sight occlusion between adjacent spatial grid cells; The mobile obstacle density and spatial occlusion coefficient of each spatial grid unit are mapped into a three-dimensional thermal map, which is superimposed to generate the obstacle distribution map.

4. The method according to claim 1, wherein The step of jointly encoding the multimodal data and the obstacle distribution map to generate an environmental pressure feature vector includes: Extracting a parameter set from the multimodal data that overlaps with a high-density area in the obstacle distribution map based on the temporal trend of noise decibels and the spatial distribution of crowd density, and performing time-series accumulation on parameters in the parameter set whose fluctuation amplitude exceeds a preset threshold to generate a parameter sequence reflecting sudden environmental interference events; Correlating and matching the mobile obstacle density and spatial occlusion coefficient with a parameter sequence, determining a diffusion path of the sudden environmental interference event in the obstacle distribution map, and adjusting the mobile obstacle density and spatial occlusion coefficient of each spatial grid unit according to the diffusion path; Based on the adjusted mobile obstacle density, spatial occlusion coefficient and the temporal accumulation results of the parameter sequence, the temporal fluctuation characteristics of the multimodal data and the spatial topological characteristics of the obstacle distribution map are tensor-spliced, and the splicing results are encoded through a convolutional neural network to generate an environmental pressure feature vector.

5. The method according to claim 1, wherein The limb movement time series data includes a continuous change sequence of limb swing frequency, trunk contraction amplitude and head deflection angle; The method of pattern matching the limb movement time series data with a preset pet body language database to extract abnormal behavior segments that are temporally and spatially associated with the outdoor complex environment includes: Dividing the limb movement time series data into analysis windows of variable length, and dynamically adjusting the sliding interval of the analysis windows according to the spatial density change of the obstacle distribution map in the outdoor complex environment; In each analysis window, based on the motion change characteristics of adjacent moments in the limb motion time series data, the degree of matching between the limb motion parameters in the current analysis window and the standard behavior patterns in the pet body language database is calculated, and the parameters related to visual interference in the environmental pressure feature vector are combined to generate dynamic matching conditions; Filtering the analysis windows whose matching degree is lower than a preset benchmark according to the dynamic matching condition, extracting the periodic differences between the limb swing frequency and the trunk contraction amplitude within the analysis window, and combining the location information of the high-density area in the obstacle distribution map to mark the abnormal behavior segments that are positionally associated with the outdoor complex environment; The abnormal behavior segments are correlated and compared with the spatial distribution changes of the crowd density in the multimodal data, and non-stress action data generated by the pet's autonomous behavior are excluded to generate the spatiotemporally correlated abnormal behavior segment set.

6. The method according to claim 2, characterized in that The identifying the emotional state of the target pet based on the natural language description, combined with the body movement timing data and the noise decibels, and adding emotional soothing suggestions corresponding to the emotional state to the translation instruction include: Performing word segmentation processing on the natural language description, extracting context-related features, and generating a sentiment feature vector, wherein the sentiment feature vector includes word segmentation weight distribution and semantic dependency strength; Establishing a two-dimensional feature space with body movement amplitude and movement frequency as axes, constructing a dynamic trajectory point set based on the emotional feature vector and the target pet's body movement time series data within a set time window, and dividing the emotional state into emotional intensity levels according to the distribution density of the dynamic trajectory point set in the two-dimensional feature space; Extracting voiceprint pulse intervals whose duration exceeds a threshold in the noise decibel sequence, weightedly coupling the energy integral value of each voiceprint pulse interval with the motion frequency of the dynamic trajectory point set to generate an emotional state determination coefficient; A bidirectional mapping table between emotion labels and body movement patterns is constructed based on the frequency domain harmonic components of the voiceprint pulse interval. When the emotion state determination coefficient exceeds the dynamic baseline, the context-related features are used to correct the confidence level of the emotion label corresponding to the current body movement pattern. When the confidence of the corrected emotion label exceeds a preset threshold, an interactive instruction template that is positively correlated with the emotion intensity level is matched from the emotion soothing strategy library, and the interactive instruction template is semantically aligned with the context of the translation instruction to output a translation instruction with additional emotion soothing suggestions.

7. The method according to claim 4, characterized in that The step of performing time-series accumulation on the parameters in the parameter set whose fluctuation amplitude exceeds a preset threshold to generate a parameter sequence reflecting the sudden environmental interference event includes: Establish a dynamic fluctuation threshold in time slices. Based on the historical trend of noise decibels over time and the historical spatial distribution of crowd density in the parameter set, set the range exceeding two standard deviations of the median of the historical trend or historical distribution change as the initial trigger boundary; Calculate the parameter value fluctuation difference of adjacent time slices. When the difference direction of three consecutive time slices is consistent and the fluctuation amplitude exceeds the initial trigger boundary, lock the current time slice to the end mark of the interference event; Based on the locked time slice, the frequency domain energy surge interval of noise decibels and the mutation direction of the spatial gradient of crowd density are extracted, and the overlapping areas in the time dimension are divided into independent interference segments, and interference segments with temporal continuity or overlapping spatial coverage are accumulated; The accumulated interference segments are parameter-weighted, wherein the duration of the frequency domain energy surge interval and the spatial diffusion speed of the crowd density spatial gradient are used as weight factors to generate a parameter sequence reflecting the sudden interference event in the environment.

8. The method according to claim 5, characterized in that The step of calculating the degree of matching between the limb movement parameters in the current analysis window and the standard behavior patterns in the pet body language database based on the movement change characteristics at adjacent moments in the limb movement time series data includes: Decomposing the continuous change sequence of limb swing frequency, trunk contraction amplitude and head deflection angle in the limb movement time series data into motion trajectories of multiple joints, and extracting the motion change characteristics of each joint at adjacent moments; Based on the gradient direction of the spatial density change in the obstacle distribution map, the weight distribution of the joint points of the motion trajectory decomposition is dynamically adjusted, and the offset of the joint points of the limb parts corresponding to the high-density areas is calculated; Based on the offset calculation result, the cosine value of the angle between the motion change feature and the reference trajectory corresponding to the standard behavior pattern is counted in the analysis window, and the cosine value of the angle is dynamically thresholded based on the time distribution characteristics of the visual interference parameter in the environmental pressure feature vector; The truncated cosine values ​​of the included angle are weighted and summed according to the joint weights to generate a local matching degree. The local matching degree is then phase-compensated based on the periodic difference between the limb swing frequency within the analysis window and the standard behavior pattern in the pet body language database to obtain the matching degree of the current analysis window.

9. The method according to claim 1, characterized in that The step of determining the predicted result of the stress level of the target pet in the outdoor complex environment by associating the environmental pressure feature vector with the abnormal behavior segment includes: Constructing a spatiotemporal alignment relationship between the environmental pressure feature vector and the abnormal behavior segment, and extracting a joint feature set of the environmental pressure feature vector and the abnormal behavior segment within an overlapping time window; Performing nonlinear correlation analysis on the joint feature set, capturing the coupling relationship between the environmental pressure feature vector and the behavioral abnormality segment through a multi-layer feature interaction network, and generating a coupling strength coefficient representing the strength of the correlation between the environment and the behavior; Based on the numerical distribution range of the coupling strength coefficient, dividing the stress level threshold interval of the target pet in the outdoor complex environment; According to the landing point of the coupling strength coefficient in the stress level threshold range, the confidence weight of the stress level is dynamically allocated, and the confidence weight is corrected in combination with the same environmental pressure in historical stress events to obtain a stress level prediction result.

10. A pet status intelligent translation method based on multimodal data, characterized in that: include: An acquisition module acquires multimodal data of the complex outdoor environment in which the target pet is located, wherein the multimodal data includes noise decibels, crowd density, and three-dimensional spatial data captured by a lidar; an encoding module, which generates an obstacle distribution map based on the three-dimensional spatial data, and jointly encodes the multimodal data and the obstacle distribution map to generate an environmental pressure feature vector; An extraction module synchronously collects the target pet's body movement time series data, performs pattern matching on the body movement time series data with a preset pet body language database, and extracts abnormal behavior segments that are temporally and spatially associated with the outdoor complex environment; a mapping module, which determines a predicted result of the stress level of the target pet in the outdoor complex environment by associating the environmental pressure feature vector with the abnormal behavior segment; A generation module generates a translation instruction according to the stress level prediction result, wherein the translation instruction includes an avoidance path prompt for a high-density area in the obstacle distribution map.

Images