Equation competition performance prediction system based on multi-modal monitoring

Abstract

The invention discloses an equestrian competition performance prediction system based on multi-modal monitoring, and relates to the technical field of equestrian sport monitoring and intelligent prediction. The system comprises an acquisition module, a processing module, a prediction module, a display module and a data closed-loop optimization mechanism. The acquisition module is used for acquiring horse video and physiological and environmental multi-modal data through multiple types of equipment; the processing module is used for realizing data fusion processing, gait behavior recognition and facial expression emotion evaluation; the prediction module outputs a competitive performance quantitative score and risk early warning based on a machine learning model; the display module provides a visual interaction interface; and the data closed-loop optimization mechanism realizes model iterative optimization through competition result return. According to the method, the problems of single monitoring dimension, data splitting, lack of prediction capability and the like in the prior art are solved, comprehensive perception of horse states and accurate prediction of competitive performance are realized, the scientific level of training management is improved, and the method has wide practical prospects and industrialization values.

Description

A Multimodal Monitoring-Based Equestrian Performance Prediction System Technical Field

[0001] This invention relates to the field of equestrian sports monitoring and intelligent prediction technology, specifically to an equestrian performance prediction system based on multimodal monitoring. Background Technology

[0002] With the development of equestrian sports and the increasing professionalization of horse racing, scientific management and training have gradually become key aspects of improving equine performance. Traditional equine training and health assessment mainly rely on the visual inspection and subjective judgment of experienced trainers or veterinarians, which has certain limitations. In recent years, Internet of Things (IoT) technology, wearable devices, and artificial intelligence have been gradually introduced into the equestrian industry, forming a preliminary "smart equestrian" system. Devices such as smart saddle pads, heart rate monitoring belts, and motion trackers have appeared on the market to collect data such as heart rate, cadence, speed, and stress during horse exercise. In addition, some studies have begun to explore using video analysis of gait or thermal imaging to detect areas of muscle strain for health monitoring or injury early warning.

[0003] However, existing technologies generally suffer from the following defects and shortcomings: most systems only cover some physiological or motor parameters of horses, failing to provide a holistic assessment of their condition, resulting in a single monitoring dimension; visual information and physiological data are not effectively integrated, lacking comprehensive indicators to support the judgment of training effects and competitive state, leading to a disconnect between data collection and analysis; there is a lack of research on "predictive management," and a lack of models and mechanisms to infer competition performance from training data, making it impossible to achieve a quantitative assessment of the horse's competitive potential; visual technology is mostly used for gait recognition, without addressing the recognition of emotional factors such as facial expressions, making it difficult to assess the horse's psychological state and its impact on competitive performance; most systems focus on health monitoring or lameness warnings, neglecting the identification of the horse's "peak performance" and its value in assisting pre-race decision-making.

[0004] Therefore, there is an urgent need for an intelligent system that can integrate multimodal data collection, possess behavioral recognition and emotion analysis capabilities, and be able to predict the future competitive performance of horses based on historical training records and physiological trends, in order to fill the gaps in existing equestrian technology in predictive analysis and decision support. Summary of the Invention

[0005] This invention aims to address the problems of data silos, one-sided state recognition, and lack of accurate performance prediction in existing equestrian training and competition prediction systems, and proposes an equestrian competition performance prediction system based on multimodal monitoring and algorithm analysis.

[0006] A multimodal monitoring-based equestrian performance prediction system includes: a data acquisition module for acquiring multimodal data of horses, including video image data, physiological index data, and environmental parameter data; a processing module communicatively connected to the acquisition module for fusing the multimodal data and performing facial expression recognition, gait and behavior recognition, and physiological state analysis based on the fused data; a prediction module communicatively connected to the processing module for predicting the horse's competitive performance and generating risk warnings based on the output of the processing module using a machine learning model; a display module for visually displaying the output results of the prediction module; and a data closed-loop optimization mechanism for incrementally learning and optimizing the machine learning model of the prediction module based on the horse's actual competition results.

[0007] Furthermore, the acquisition module includes: a video acquisition unit, which includes a high-definition camera and an infrared thermal imager deployed in the stable and training track area; a physiological sensing unit, which includes a heart rate strap, a pressure-sensing saddle pad, and a body temperature patch that can be worn on the horse; an environmental data acquisition unit, which includes a temperature and humidity sensor, a barometric pressure sensor, and a noise sensor; a GPS positioning module for monitoring the horse's movement trajectory; and a training load sensing module integrated into the training equipment.

[0008] Furthermore, the processing module includes: a data fusion processing unit for cleaning, temporal synchronization, and feature extraction of multi-source heterogeneous data from the acquisition module; a facial expression recognition unit for assessing the horse's emotional state by extracting facial feature points based on the video image data; a gait and behavior recognition unit for identifying the horse's gait rhythm, symmetry, and behavior category based on the video image data through key point detection and skeleton modeling; and a physiological state analysis unit for assessing the horse's training load and recovery state based on the physiological index data through temporal analysis methods.

[0009] Furthermore, the prediction module includes: a state assessment model, used to comprehensively assess the horse's real-time competitive state based on the output of the processing module; and a competitive performance prediction model, employing a multi-layer neural network model or a time-series regression model, used to output a quantitative competition performance prediction score and risk warning level based on the output of the state assessment model, the horse's historical training data, and psychological state indicators.

[0010] Furthermore, the data closed-loop optimization mechanism is configured to: use the actual race results of the horses as a supervision signal, compare them with the prediction results of the prediction module, and iteratively update the parameters of the machine learning model by optimizing the loss function to achieve incremental learning of the model.

[0011] Furthermore, the processing module and the prediction module are deployed on the same main control processor, which adopts an ARM architecture and integrates a wireless communication module to receive data from the acquisition module in real time and perform edge computing, with a data processing delay of no more than 500ms.

[0012] Furthermore, the display module specifically includes a mobile application and a web page, used to display horse status trend charts, emotion recognition results, and competitive performance prediction scorecards, and supports the generation and export of training reports containing the prediction scores and risk warnings.

[0013] Furthermore, the video acquisition unit, physiological sensing unit, and environmental data acquisition unit are connected to the processing module via wired or wireless communication methods, wherein the wireless communication methods include Bluetooth or LoRa.

[0014] This technical solution integrates four dimensions: video recognition, sensor analysis, intelligent prediction, and interactive feedback. It features strong systematicity, good adaptability, high prediction accuracy, and flexible deployment. It can significantly improve the scientific level of horse training and management and the efficiency of competition strategy formulation, and has significant industry promotion value.

[0015] The beneficial effects of this invention are: 1. Multimodal fusion enhances the comprehensiveness of assessment: The system integrates multi-source data from video recognition, thermal imaging, physiological sensors, and environmental sensors, covering horse behavior, physical characteristics, emotions, and environmental dimensions. Compared to traditional single-point sensors, it has higher data integrity and accuracy, achieving a comprehensive assessment of the horse's condition; 2. Emotion recognition fills a technological gap: Utilizing facial expression feature points, the system assesses emotions, identifying states such as anxiety, alertness, and fatigue in real time, providing trainers with key behavioral references and solving the problem that existing systems cannot identify the horse's psychological state; 3. Accurate prediction assists scientific decision-making: An AI-based training-performance correlation model is constructed to quantitatively score the horse's competitive state before the race, assisting in the formulation of scientific horse selection and competition strategies. 4. Enhanced Model Evolution Capability through Closed-Loop Feedback: Model self-learning and performance optimization are achieved through feedback of competition results, improving the long-term accuracy and adaptability of the prediction model. Unlike traditional static model systems, it possesses continuous optimization capabilities. 5. Modular Design Enhances Practicality and Scalability: Users can intuitively view status trends, risk warnings, and model suggestions through a graphical terminal, enhancing ease of use. The modular hardware design can flexibly adapt to different horse farms or competition scenarios, reducing deployment costs and facilitating industry promotion. 6. Ensuring Horse Health and Improving Training Efficiency: Real-time quantification of training load and recovery indicators warns of overtraining risks and reduces horse injury rates. Simultaneously, it accurately identifies high-efficiency training intervals, improving overall training efficiency and horse lifespan. Attached Figure Description

[0016] Figure 1 is a schematic block diagram of the overall system structure of the present invention; Figure 2 is a schematic diagram of the layout of the stable and track monitoring camera and thermal imaging of the present invention; Figure 3 is a schematic diagram of the installation of the wearable sensor (heart rate belt, pressure pad, body temperature patch) of the present invention; Figure 4 is a schematic diagram of the UI of the training status and competition scoring interface of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] As shown in Figures 1 to 4: As shown in Figure 1, this system includes a data acquisition module, a processing module, a prediction module, a display module, and a data closed-loop optimization mechanism. The data acquisition module collects multimodal data of horses through various sensors and camera devices; the processing module fuses and analyzes the data; the prediction module outputs a prediction of athletic performance based on the analysis results; the display module provides a visual interactive interface; and the data closed-loop optimization mechanism optimizes the prediction model through feedback from competition results.

[0019] The processing module is deployed on the main control processor, adopts an ARM architecture, and integrates a wireless communication module to achieve real-time data reception and edge computing. The data processing latency does not exceed 500ms. The processing module includes a data fusion processing unit, a facial expression recognition unit, a gait and behavior recognition unit, and a physiological state analysis unit. The processing module is communicatively connected to the acquisition module and is used to fuse the multimodal data. Based on the fused data, it performs facial emotion recognition and gait symmetry analysis of horses using a computer vision model, and assesses physiological state using a time-series model.

[0020] The prediction module includes a condition assessment model and a performance prediction model. The condition assessment model processes the output of the module to evaluate the horse's real-time condition; the performance prediction model, based on the condition assessment results, historical data, and psychological indicators, outputs a quantitative score and risk warning.

[0021] The display module provides a visual interface via mobile application or web page, as shown in Figure 4. The interface includes a basic horse information area, a status trend graph module, an expression / emotion recognition panel, a performance prediction scorecard, and a report export toolbar.

[0022] The data closed-loop optimization mechanism uses the actual competition results as a supervision signal, compares them with the prediction results, and iteratively updates the model parameters by optimizing the loss function to achieve incremental learning.

[0023] 1. The specific implementation and deployment of the data acquisition module are shown in Figures 2 and 3: The data acquisition module completes the synchronous acquisition of multimodal data of horses through visual devices deployed in fixed locations and sensor devices worn on the horses. The deployment of the data acquisition device is as follows: 1.1 Fixed visual monitoring equipment deployment in the stable area: A high-definition wide-angle camera is installed directly above the entrance of each stable to continuously monitor the daily behaviors of horses such as entering and exiting, drinking, standing, and lying down.

[0024] An infrared thermal imager is installed in the center of the ceiling inside the stable, with a field of view covering the entire interior space of the stable, to monitor the temperature distribution and thermal zone changes of the horses' bodies.

[0025] Training track area: Four to six high-definition cameras are evenly deployed along the inner side of the track to cover the movement trajectory of the horses throughout the training process from multiple angles.

[0026] High-frame-rate thermal imaging instruments were deployed at the start and finish lines of the track to capture changes in the horses' body surface temperature before and after high-intensity exercise, providing data for fatigue assessment.

[0027] Signal Connection and Integration: All cameras and thermal imaging devices are connected to the nearest data acquisition and control box via wired or wireless means, and ultimately integrated into the edge processing equipment. Image signals from the training track area are transmitted back to the main processing server via fiber optic cable. The devices are installed at staggered angles to ensure no blind spots in monitoring.

[0028] 1.2 Deployment of Wearable Physiological Sensing Devices Simultaneously, the following wearable sensors are used to record the horse's physiological data in real time: Heart Rate Belt: Worn on the horse's chest, secured by an elastic band and conforming to the skin, it is used to detect the heart rate during exercise in real time and obtain the heart rate variability (HRV) index to determine the intensity of training load, fatigue level, and recovery status.

[0029] Pressure-sensing saddle pad: Embedded under the saddle, in direct contact with the horse's back, it is used to detect the pressure distribution in different areas of the saddle pad, analyze whether the pressure on the saddle is even, prevent pressure sores and discomfort, and help identify changes in the horse's posture during training.

[0030] Temperature monitoring patch: This flexible adhesive patch is applied to the flank or shoulder area of ​​the horse to continuously monitor its body surface temperature. This data can be fused with thermal imaging data to identify abnormal temperature rises (such as inflammation or fatigue) or temperature fluctuations, providing early warning of performance abnormalities.

[0031] 1.3 Auxiliary motion data acquisition equipment In addition, the acquisition module also includes a GPS positioning module for monitoring the horse's movement trajectory, and a training load sensing module integrated into training equipment (such as horse tack or training equipment) for quantifying and recording the external load during the exercise process.

[0032] 1.4 The terminal deploys a main control processor, integrating an edge computing chip and a wireless communication module, capable of real-time data reception, time-series matching, feature extraction, and intermediate analysis calculations; the analysis and prediction terminal is a computing platform running training status scoring and performance prediction algorithms. The main control processor is deployed in the racecourse control center, while the analysis and prediction terminal can be deployed on an edge server or in the cloud, accessible to users via an app or webpage.

[0033] 2. The configuration requirements for the acquisition module in this application are as follows: Video acquisition unit: including high-definition cameras (resolution ≥ 1080P, frame rate ≥ 30fps) and infrared thermal imagers (temperature measurement range -20℃~150℃, temperature measurement accuracy ±0.5℃) deployed in the stable and training track areas; a high-definition wide-angle camera is installed directly above the stable entrance, and an infrared thermal imager is installed in the center of the internal ceiling; 4-6 cameras are evenly distributed along the inner side of the training track. The system includes a high-definition camera system, with high-frame-rate thermal imaging devices deployed at both the starting and ending points, arranged at staggered angles to achieve blind-spot-free monitoring; a physiological sensing unit comprising a wearable heart rate monitor (sampling frequency 1Hz~10Hz), a pressure-sensitive saddle pad (sampling frequency 5Hz~20Hz), and a body temperature patch (temperature range 30℃~42℃, accuracy ±0.1℃); the heart rate monitor is secured to the horse's chest with an elastic band, the pressure-sensitive saddle pad is embedded in the bottom of the saddle, and the body temperature patch is flexibly adhered to the horse's flank or shoulder; an environmental data acquisition unit including a temperature and humidity sensor (temperature accuracy ±0.3℃, humidity accuracy ±2% RH), a barometric pressure sensor (accuracy ±1hPa), and a noise sensor (measurement range 30dB~130dB); and an auxiliary acquisition module including a GPS for monitoring the horse's movement trajectory. The system comprises a positioning module (positioning accuracy ≤ 5m) and a training load sensing module integrated into the training equipment (sampling frequency 10Hz); 2. The system operation flow of this invention is as follows: Before training: The trainer puts the sensing device on the horse and checks the connection and data acquisition normality; During training: The system is started, and the acquisition device records video, physiological, environmental and movement trajectory data in real time and transmits it to the processing module; Data processing: The processing module completes data cleaning, synchronization, normalization and feature extraction to achieve emotion recognition, gait analysis and physiological state assessment; Prediction output: The prediction module calls the model to generate a competitive performance score and risk warning report; Result display: Users view the results through a visual interface, and the data is archived to the analysis terminal; Closed-loop optimization: The actual results are fed back after the competition for model parameter updates.

[0034] The main components of the visualization interface are as follows (see Figure 4 for details): Horse basic information area: displays basic information such as horse number, name, breed, age, and trainer's name; displays background information such as the start and end time of the current training cycle, training frequency, and stable number.

[0035] Status Trend Module: Line graphs show the changes of key indicators over time during training, such as average heart rate, cadence, body temperature, training load index (TLI), recovery index (RI), etc.; different physiological or behavioral parameter curves can be viewed, and comparisons can be made by day / week / month; threshold reminders can be set: when an indicator exceeds the safe range, it will be highlighted in red or a warning icon will be displayed.

[0036] Expression / Emotion Recognition Panel: Displays the recent emotional state trends of horses analyzed by the AI ​​model, such as tension, relaxation, alertness, and fatigue; uses simple expression icons with text descriptions to assist in training decisions and adjustments to training intensity; Competitive Performance Prediction Scorecard: Displays the system's prediction of the competition potential corresponding to the current training cycle (e.g., score 85 / 100, good condition); displays the stability assessment of the condition (e.g., "moderate fluctuation, it is recommended to continue training for 1 week"); provides risk level prompts (e.g., mild fatigue, abnormal stress distribution needs to be observed, etc.); Report and Export Toolbar: Users can choose to export training status summary reports, prediction score reports, charts in PDF or CSV format; supports data synchronization to the horse farm management system or printing paper reports for archiving.

[0037] 3. The data processing method of this invention is as follows: Data fusion processing: cleaning, timing synchronization and feature extraction are performed. Based on the edge processor clock, each node is synchronized through the NTP protocol, and software offset compensation is performed on the video stream (fixed delay of about 100ms) and Bluetooth sensor data (delay of 20-50ms) to achieve an alignment accuracy of ≤10ms.

[0038] Facial expression recognition: Emotions are assessed by extracting facial feature points. A key point detection model based on deep learning (such as a variant of the HRNet architecture) is used to locate feature points such as ears, eyes, nose, and mouth. The time sequence of feature point coordinates is then input into a recurrent neural network (such as LSTM) for classification.

[0039] Gait and behavior recognition: Keypoint detection and skeleton modeling are used to identify keypoints. 2D / 3D pose estimation algorithms (such as OpenPose or VideoPose3D) are used to obtain keypoints. Spatiotemporal graph convolutional network (ST-GCN) is used to model the spatiotemporal relationship of keypoints to identify behaviors.

[0040] State assessment model: Based on heart rate variability data, the time-domain index RMSSD and / or frequency-domain index high-frequency power reflecting parasympathetic nerve activity are calculated, and combined with the dynamic changes of the above indexes before and after training, the training load and recovery status of horses are quantitatively assessed.

[0041] Competitive performance prediction model: The prediction model is a multi-task learning model that uses a machine learning model for prediction. It includes a shared feature encoding network (such as a multi-layer fully connected network) and parallel regression output layers (for scoring) and classification output layers (for risk level).

[0042] The algorithm flow of the above data processing method is a preferred embodiment of the present invention. Those skilled in the art can make equivalent substitutions for the algorithm structure or model form without departing from the overall technical concept of the present invention.

[0043] Example 1 uses this system to monitor the training of 10 competition horses (aged 4-5 years, weighing 440-480 kg) at a horse farm. The core objective is to accurately identify abnormal horse conditions, quantify competitive potential, select the best horses for competition, and optimize training programs accordingly to improve overall competitiveness in the competition through multimodal monitoring of the system.

[0044] I. Data Acquisition Deployment (Covering 10 Horses in Full Dimensions of Monitoring) Video Acquisition Unit: Each of the 10 dedicated stables in the stable area is equipped with one 1080P high-definition camera (30fps) + one infrared thermal imager (temperature measurement accuracy ±0.5℃); Six high-definition cameras are deployed on the inner side of the training track (1600 meters), and one high-frame-rate thermal imager (60fps) is deployed at the start and finish lines to achieve blind-spot-free motion capture and body surface temperature monitoring of the 10 horses throughout the training process.

[0045] Physiological sensing unit: Each horse is fixedly equipped with a chest heart rate strap (sampling frequency 8Hz), a saddle-embedded pressure-sensing pad (sampling frequency 15Hz), and a flexible body temperature patch on the side abdomen (accuracy ±0.1℃) to collect heart rate, pressure distribution, and body temperature data in real time.

[0046] Movement trajectory and environmental data collection: Each horse's harness is equipped with a GPS positioning module (positioning accuracy ≤5m), and three sets of temperature and humidity (accuracy ±0.3℃ / ±2%RH), air pressure (±1hPa), and noise (30-130dB) sensors are deployed around the training track to collect movement trajectory and environmental parameters simultaneously.

[0047] Data transmission and processing: All acquisition devices are connected to the ARM architecture main control processor via Bluetooth + LoRa wireless communication. The data processing latency is 410ms. The processing module completes multi-source data cleaning, time synchronization (alignment accuracy ≤10ms) and feature extraction.

[0048] II. Results of the Analysis Phase (Accurate Identification of Abnormal States in 3 Horses) Horse No. 3 Gait Symmetry Deviation: Through ST-GCN spatiotemporal graph convolutional network analysis in the processing module, there was a significant difference in gait rhythm between the left and right hind limbs of Horse No. 3 during training. The gait symmetry was only 82% (normal competitive horses ≥90%), and the movement trajectory deviation of key joints reached 1.2cm, which was judged to be an imbalance in gait force.

[0049] Horse No. 5 exhibited significant anxiety: Based on facial feature point extraction from video images (HRNet architecture model) and LSTM recurrent neural network classification, Horse No. 5 showed 45% anxiety, 20% alertness, and only 15% relaxation during training, significantly higher than other horses (average anxiety ≤20%), indicating insufficient psychological stability.

[0050] Horse No. 7's heart rate recovery was abnormal: According to the time-series analysis method of the physiological state analysis unit, after each 60-minute high-intensity training session, Horse No. 7's resting heart rate (average 65 bpm before training) took 3.5 hours to recover to below 80 bpm, while the average recovery time for other horses was 1.5-2 hours, indicating that its cardiovascular load tolerance and recovery ability were relatively weak.

[0051] The remaining 7 horses: Analysis showed that horses No. 1, 2, 4, 6, 8, 9, and 10 had a gait symmetry of ≥90%, a mood state mainly of "relaxation and soothing" (accounting for ≥70%), and physiological indicators (heart rate, body temperature, stress distribution) and recovery curves were all within the normal range.

[0052] III. Prediction Phase Output (Quantitative Scoring and Competition Recommendations) The prediction module uses a multi-layer neural network model, integrating the output of the processing module, the historical training data of 10 horses over the past 3 months, and psychological state indicators to output a quantitative performance score (out of 100 points) and risk warnings: Horse Number Performance Score Risk Level Competition Recommendations 1 78 points Low Risk Alternative Competition 2 75 points Low Risk Alternative Competition 3 72 points Medium Risk Postpone Competition, Optimize Gait 4 83 points Low Risk Alternative Competition 5 76 points Medium Risk Postpone Competition, Regulate Mood 6 85 points Low Risk Alternative Competition 7 74 points Medium Risk Postpone Competition, Improve Recovery Ability 8 89 points No Risk Priority Competition 9 81 points Low Risk Alternative Competition 10 87 points Low Risk Alternative Competition Horse No. 8 in the table has the highest score (points). The status assessment model shows that its "real-time competitive status is excellent". Its heart rate variability (RMSSD value 52ms), gait symmetry (95%), and emotional stability (relaxation + soothing ratio 85%) are the best among the 10 horses. There are no risk warnings. The system clearly marks it as "recommended to participate in the competition".

[0053] IV. Decision Feedback Measures (Targeted Training Optimization and Competition Selection) Adjustment of Training for Abnormal Horses: Horse No. 3: Add "Gait Correction Special Training" (30 minutes per day, focusing on strengthening the coordination of the left hind limb), and use thermal imaging to monitor the distribution of muscle power heat zones, and assess the improvement of gait symmetry weekly.

[0054] Horse #5: Reduce daily training time (from 60 minutes to 45 minutes), add 20 minutes of environmental adaptation and calming training before training (reduce venue noise, standardize trainer operating procedures), and track changes in the proportion of tension in real time through an emotion recognition panel.

[0055] Horse #7: Reduce training load index (TLI from 1.4 to 1.0), increase recovery period (training interval from 1 day to 1.5 days), supplement electrolytes and high-protein feed, and monitor the improvement of heart rate recovery curve after training.

[0056] Horses selected for the competition: Based on the system's prediction results, horse number 8 was selected as the main horse for the competition, and horse number 10 was selected as the backup (scoring 87 points, stable condition).

[0057] V. Internal test of closed-loop optimization effect training results: In the 1600-meter test competition organized within the racecourse, horse number 8 won the 1600-meter race. The actual competitive performance completely matched the "optimal state" predicted by the system, and the consistency between the predicted score and the race result reached 98%.

[0058] Model optimization: The subsequent race / training data of horse No. 8 and the other 9 horses (a total of 120 valid samples) are fed back to the data closed-loop optimization mechanism. The system uses the actual results as the supervision signal to optimize the loss function and iteratively update the machine learning model parameters to achieve incremental learning.

[0059] Performance improvements: After optimization, the system's accuracy in predicting the competitive performance of racing horses has increased by 8%, the response speed for gait abnormality recognition has accelerated by 15%, and the accuracy of emotion state classification has improved to 92%, providing more reliable technical support for subsequent race selection and training optimization.

[0060] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0061] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A performance prediction system for equestrian competitions based on multimodal monitoring, characterized in that... include: The data acquisition module is used to acquire multimodal data of horses, including video image data, physiological index data, and environmental parameter data. The processing module, communicatively connected to the acquisition module, is used to fuse the multimodal data and perform facial expression recognition, gait and behavior recognition, and physiological state analysis based on the fused data. The prediction module, also communicatively connected to the processing module, is used to predict the horse's competitive performance and generate risk warnings based on the output of the processing module using a machine learning model. Display module: used to visualize the output results of the prediction module; Data closed-loop optimization mechanism: used to incrementally learn and optimize the machine learning model of the prediction module based on the actual race results of the horses.

2. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that: The video acquisition unit includes high-definition cameras and infrared thermal imagers deployed in the stables and training track areas; The physiological sensing unit includes a heart rate monitor, a pressure-sensitive saddle pad, and a body temperature patch that can be worn on the horse; the environmental data acquisition unit includes a temperature and humidity sensor, a barometric pressure sensor, and a noise sensor.

3. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that: The data acquisition module also includes a GPS positioning module for monitoring the movement trajectory of horses and a training load sensing module integrated into the training equipment.

4. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that: The processing module includes a data fusion processing unit for cleaning, time-series synchronization, and feature extraction of multi-source heterogeneous data from the acquisition module; and a facial expression recognition unit for assessing the emotional state of horses by extracting facial feature points based on the video image data. The gait and behavior recognition unit is used to identify the gait rhythm, symmetry, and behavior category of a horse based on the video image data through key point detection and skeleton modeling.

5. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that: The physiological state analysis unit is used to evaluate the training load and recovery status of horses based on the physiological index data using time-series analysis methods.

6. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that: The prediction module includes a state assessment model, which is used to comprehensively assess the horse's real-time competitive state based on the output of the processing module; and a competitive performance prediction model, which uses a multi-layer neural network model or a time-series regression model, to output a quantitative competition performance prediction score and risk warning level based on the output of the state assessment model, the horse's historical training data and psychological state indicators.

7. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that... The data closed-loop optimization mechanism is configured to: use the actual race results of the horses as a supervision signal, compare them with the prediction results of the prediction module, and iteratively update the parameters of the machine learning model by optimizing the loss function to achieve incremental learning of the model.

8. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that: The processing module and the prediction module are deployed on the same main control processor. The main control processor adopts an ARM architecture and integrates a wireless communication module to receive data from the acquisition module in real time and perform edge computing. Its data processing latency does not exceed 500ms.

9. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that: The display module specifically includes a mobile application and a web page, used to display horse status trend charts, emotion recognition results, and competitive performance prediction scorecards, and supports the generation and export of training reports containing the prediction scores and risk warnings.

10. The equestrian performance prediction system based on multimodal monitoring according to claim 1, characterized in that: The video acquisition unit, physiological sensing unit, and environmental data acquisition unit are connected to the processing module via wired or wireless communication methods, including Bluetooth or LoRa.

Images