AI-based high-precision mouse direction control method, device, and mouse

Abstract

The present invention relates to the field of direction control technology, and in particular to an AI-based high-precision mouse direction control method, device, and mouse. The method comprises the following steps: obtaining basic mouse position data; tracking the mouse's movement path based on the basic mouse position data to generate mouse movement path data; performing two-dimensional vector decomposition of the mouse movement path data on the path point plane to generate the mouse's plane movement horizontal coordinate and the mouse's plane movement vertical coordinate; obtaining mouse display resolution data; and calculating the mouse's direction change angle and the number of direction reversals based on the mouse's plane movement horizontal coordinate and the mouse's plane movement vertical coordinate. The present invention effectively solves the problems of path error, insufficient jitter recognition, and unstable response in traditional mouse direction control through multi-sensor data fusion and AI adaptive compensation, thereby achieving high-precision mouse direction control.

Description

Technical Field

[0001] The present invention relates to the field of direction control technology, and in particular to an AI-based high-precision mouse direction control method, device, and mouse. Background Art

[0002] Traditional mouse control relies on mechanical and optical sensors. While achieving high sensitivity, it still suffers from issues such as insufficient precision, delayed response, and susceptibility to interference in complex environments. Leveraging deep learning, computer vision, and sensor fusion technologies, researchers are exploring methods to use AI models to predict and correct mouse motion trajectories in real time, thereby improving the accuracy and stability of directional control. Early research focused on gesture recognition and trajectory prediction based on image recognition, using convolutional neural networks (CNNs) to extract hand or mouse motion features and achieve rough direction determination. Subsequently, recurrent neural networks (RNNs) and long short-term memory networks (LSTMs) were introduced to capture the time series characteristics of mouse motion, significantly improving the continuity and accuracy of trajectory prediction. However, currently, the display of the mouse cursor is affected by screen resolution during movement, which can lead to misjudgment of mouse direction due to screen resolution and rendering artifacts. Furthermore, current mouse directional control cannot accurately capture the user's actual control intentions, resulting in low control stability and accuracy. Summary of the Invention

[0003] Based on this, it is necessary to provide an AI-based high-precision mouse direction control method, device and mouse to solve at least one of the above technical problems.

[0004] To achieve the above objectives, a high-precision mouse direction control method based on AI is provided, the method comprising the following steps:

[0005] Step S1: Acquire basic mouse position data; track the mouse movement path based on the basic mouse position data to generate mouse movement path data; decompose the mouse movement path data into two-dimensional vectors of the path points to generate the horizontal coordinate and vertical coordinate of the mouse movement plane;

[0006] Step S2: obtaining mouse display resolution data; calculating the direction change angle and the number of direction reversals of the mouse according to the horizontal coordinate and the vertical coordinate of the mouse plane movement; performing movement edge jaggies display feature extraction on the mouse display resolution data to obtain mouse movement jaggies change feature data; performing mouse jitter detection on the mouse movement jaggies change feature data using the direction change angle and the number of direction reversals to obtain a mouse jitter detection result;

[0007] Step S3: collecting user pressing data and electromyographic signals of the user's metacarpophalangeal joints using the built-in pressure sensor and electromyographic sensor of the mouse according to the mouse jitter detection result; performing time window stability analysis on the user pressing data and the electromyographic signals of the user's metacarpophalangeal joints to obtain user mouse control data;

[0008] Step S4: Adaptively compensate the mouse movement direction of the mouse jitter detection result using the user mouse control data to perform high-precision mouse direction control based on AI.

[0009] This invention uses two-dimensional vector decomposition of path points to achieve high-resolution analysis of minute movement trajectories, significantly improving mouse control accuracy, which is particularly critical in high-resolution or drawing scenarios. It uses direction change angles and the number of direction reversals to sensitively capture user operation trajectories, enabling comprehensive perception and understanding of complex mouse motion behaviors. By integrating jagged edge features with directional data, it effectively detects mouse jitter caused by involuntary factors and eliminates erroneous input. By acquiring signals from the mouse's built-in pressure sensors and electromyographic sensors, the system improves its ability to distinguish between "conscious" and "unconscious" movements based on the user's hand muscle activity and pressing habits. An AI model learns from different user operation patterns to achieve adaptive directional compensation tailored to different physiological characteristics, meeting personalized human-computer interaction needs. The system is particularly suitable for users with Parkinson's disease and post-stroke sequelae, effectively compensating for involuntary mouse jitter during control and achieving a smoother input experience. Even in complex usage environments such as vibrating desktops and low-stability supports, it dynamically corrects erroneous motion paths to ensure operational stability. Through adaptive analysis of display resolution data, the system can adjust the control sensitivity according to the current display environment to achieve pixel-level positioning capability. Based on the stability analysis of the time window, it effectively captures the changes in the behavioral characteristics of users in continuous operations, so that the system can dynamically adapt to the operation rhythm. Multi-dimensional fusion data (position, speed, direction, physiological signals) forms a portrait of user control behavior, improving the judgment accuracy of the system under abnormal conditions. Through the multimodal AI learning model that combines jitter detection with electromyographic signal input, it promotes the transformation of traditional peripherals into more intelligent human-computer interaction devices. Therefore, the present invention effectively solves the problems of path error, insufficient jitter recognition and unstable response in traditional mouse direction control through multi-sensor data fusion and AI adaptive compensation, and realizes high-precision mouse direction control.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: using the user device input interface to sample the mouse pointer position in real time, capturing the screen coordinate value in each frame or time slice to obtain basic mouse position data;

[0012] Step S12: performing time series reconstruction, missing point interpolation and path fitting analysis on the basic position data of the mouse, extracting a continuous path sequence of the mouse motion trajectory, and generating mouse movement path data;

[0013] Step S13: performing vector processing on the coordinate difference between any two consecutive points in the mouse movement path data to form a displacement vector set between adjacent points, and generating two-dimensional plane vector data of the path;

[0014] Step S14: Decompose the horizontal and vertical coordinates of the two-dimensional plane vector data of the path to generate the horizontal coordinate and the vertical coordinate of the mouse plane movement.

[0015] By sampling the mouse pointer position in real time, this method captures user mouse movements with frame-level or time-slice-level accuracy, ensuring the integrity and precision of the underlying position data and providing high-quality input for subsequent analysis. Time series reconstruction and missing point interpolation techniques effectively compensate for data loss caused by sampling intervals or system delays. Combined with path fitting, trajectory continuity is enhanced, making the generated mouse movement path data more accurate than the actual operation trajectory. Vectorizing adjacent points in the path yields a set of displacement vectors that comprehensively describe the mouse's motion trends and velocity variations in a two-dimensional plane, providing fundamental support for higher-level applications such as behavioral pattern recognition and operational intent analysis. Decomposing the path vector into horizontal and vertical coordinates creates independent horizontal and vertical coordinate data, facilitating the assessment of fine-grained features such as horizontal and vertical stability, tendency, and behavioral preferences. The entire step S1 design is highly versatile and adaptable to a wide range of interactive devices with pointer operations (such as computers, tablets, and smart whiteboards). It can also be combined with mouse click and scroll events to create a richer user interaction behavior modeling system.

[0016] Preferably, step S2 includes the following steps:

[0017] Step S21: Automatically detecting the resolution parameters of the display connected to the user terminal and calibrating the pixel density to obtain mouse display resolution data;

[0018] Step S22: performing coordinate sequence difference and arc cosine calculation on the horizontal coordinate and vertical coordinate of the mouse plane movement to generate the mouse direction change angle;

[0019] Step S23: performing signed gradient analysis on the direction change angle, detecting the positive and negative changes of the continuous direction vectors, counting the number of direction reversals, and generating the direction reversal count;

[0020] Step S24: analyzing the minimum discernible unit of the mouse display resolution data, and extracting the edge zigzag movement pattern of the mouse plane movement horizontal coordinate and the mouse plane movement vertical coordinate according to the minimum discernible unit, to generate mouse movement zigzag change feature data;

[0021] Step S25: performing abnormal movement fitting analysis on the mouse movement sawtooth change feature data based on the mouse direction change angle and the number of direction reversals to generate a mouse jitter detection result.

[0022] By automatically detecting the display's resolution and pixel density, this method effectively adapts to various user terminal devices, avoiding coordinate errors caused by screen parameter differences and providing accurate mouse display resolution data for subsequent feature extraction. Using coordinate differencing and arccosine calculation, it accurately reflects the angle of direction change during mouse movement, providing important parameters for assessing the stability and continuity of user operations. Using signed gradient analysis to count the number of direction reversals helps identify patterns of frequent direction changes during user operations and is particularly suitable for detecting abnormal input behaviors such as tremors and misoperations. Extracting zigzag motion features in the horizontal and vertical coordinates based on the display device's smallest discernible unit effectively captures extremely subtle direction deviations or edge wiggles, laying the foundation for micro-jitter detection. Combining the direction change angle and the number of direction reversals to perform anomaly fitting analysis on the zigzag motion features accurately identifies mouse jitter caused by hand instability, device anomalies, or abnormal input, achieving highly sensitive mouse jitter detection. This overall process can be used not only for input stability assessment but also as a key technical module for identifying user operating habits and security risks (such as automated scripting and input recognition for the elderly or individuals with disabilities).

[0023] Preferably, step S25 includes the following steps:

[0024] Step S251: Calculate the first-order derivative of the mouse direction change angle according to the timestamp, extract the angle change rate, identify the sudden direction change behavior, and generate angular velocity change feature data;

[0025] Step S252: Counting the number of direction reversals per unit time and performing sliding window clustering to extract short-term high-frequency reversal features and generate direction reversal density distribution data;

[0026] Step S253: Calculating the frequency domain energy distribution of the mouse movement sawtooth change feature data;

[0027] Step S254: normalize the angular velocity change characteristic data, direction reversal density distribution data, and frequency domain energy distribution and fit them into a jitter judgment logic curve, and perform curve change detection on the jitter judgment logic curve based on a preset jitter change threshold to obtain a mouse jitter detection result.

[0028] This method calculates the first-order derivative of the mouse's direction change angle to extract angular velocity characteristics. This method can sensitively capture sudden directional changes during user operations, providing key support for detecting sudden or nonlinear operational characteristics. Cluster analysis of direction reversals within a sliding window effectively identifies high-frequency direction reversals within a unit time, generating direction reversal density distribution data and improving the accuracy of identifying short-term micro-jitter or periodic abnormal input. Frequency domain energy distribution calculation reveals the periodic noise components in mouse motion from a spectral perspective, making it particularly suitable for identifying persistent oscillatory input or hardware-level signal noise interference. The three core features (angular velocity change, reversal density, and frequency domain energy) are normalized and fused to construct a logical fitting curve, enhancing the overall modeling capability of jitter behavior. Threshold determination is then used to implement a stable and reliable jitter detection mechanism. By utilizing a normalized characteristic curve and an adjustable threshold mechanism, the system can flexibly adjust the recognition sensitivity based on different user habits, device types, or application scenarios, achieving adaptive jitter detection. The combined use of behavioral characteristic curves and frequency characteristic analysis can not only identify anomalies caused by hardware (such as mouse failure, poor contact), but also reflect potential risk factors such as changes in user status (such as hand fatigue, nerve tremors), providing basic data support for improving the quality of human-computer interaction and health monitoring.

[0029] Preferably, step S3 includes the following steps:

[0030] Step S31: collecting user pressing data and myoelectric signals of the user's metacarpophalangeal joints using the built-in pressure sensor and myoelectric sensor of the mouse according to the mouse jitter detection result;

[0031] Step S32: performing spatial heat zone clustering on the user's pressing data to generate a user palm contact heat map; extracting isotherms from the palm contact heat map, and performing regional gradient analysis on the user palm contact heat map based on the isotherms to generate user static grip posture feature data;

[0032] Step S33: performing Fourier transform on the user's pressing pressure data to generate user's pressing frequency domain data; calculating the peak spacing of the user's pressing frequency domain data, and performing user pressing tremor recognition on the user's pressing pressure data based on the peak spacing, thereby generating periodic tremor pattern data;

[0033] Step S34: performing synchronous window registration on the periodic tremor pattern data and the electromyographic signal to generate interference fitting correction data; performing time window stability analysis on the user's pressing data and the electromyographic signal using the interference fitting correction data to generate multi-channel stability evaluation data;

[0034] Step S35: Perform joint behavior modeling using the static gripping posture feature data and the multi-channel stability evaluation data to generate user mouse control data.

[0035] By leveraging the mouse's built-in pressure and electromyographic sensors, this invention can collect real-time physiological feedback from the user while using the mouse, extending recognition from mechanical behavior to biological signals and significantly broadening the dimensions of user state perception. Spatial hotspot clustering and isotherm analysis using palm contact heatmaps help accurately restore the actual contact pattern and force distribution between the user's palm and the mouse, providing a precise basis for analyzing the user's static grip and usage habits. By extracting frequency features through Fourier transform of user pressure data and identifying periodic tremor behavior based on peak spacing, this method effectively captures characteristics such as subtle tremors and compulsive repetitive actions, making it particularly suitable for medical assistance (such as Parkinson's disease screening) or user fatigue detection. By synchronously windowing periodic tremor patterns with electromyographic signals, an interference fitting correction model is constructed to effectively eliminate or compensate for unintended input caused by muscle fatigue, false touches, neural fluctuations, and other factors, enhancing the reliability of input stability analysis. Static grip characteristics are combined with multi-channel stability indicators to generate user-characterized mouse control data, which can be used for adaptive control parameter adjustment, precise auxiliary system triggering, or customized human-computer interaction optimization strategies.

[0036] Preferably, step S35 includes the following steps:

[0037] Step S351: Perform multi-dimensional thermal distribution gradient reconstruction on the static holding posture feature data to generate posture thermal flow tensor data; perform frequency-amplitude-phase composite domain nonlinear deconstruction on the multi-channel stability evaluation data to generate stability hybrid feature tensor data;

[0038] Step S352: Using a tensor cross-gating network, the posture heat flow tensor data and the stability hybrid feature tensor data are cross-fused into a three-modal tensor to generate a user thermal muscle coupling behavior feature map, where the three modalities include thermal inertia, myoelectric displacement, and behavioral intention.

[0039] Step S353: Decoding and modeling the user's thermal-muscle coupling behavior characteristic map to generate a nonlinear control state map;

[0040] Step S354: Match the user's current posture control state according to the nonlinear control state mapping diagram to generate user mouse control data.

[0041] The present invention reconstructs the thermal distribution gradient of static gripping posture feature data to form posture heat flow tensor data. At the same time, it performs composite nonlinear deconstruction of multi-channel stability evaluation data in frequency domain, amplitude domain, and phase domain to construct stability hybrid feature tensor data, which significantly improves the spatial expression and dynamic change capabilities of complex behavioral states. A tensor cross-gating network is constructed to integrate thermal inertia (dynamic thermal distribution of posture), electromyographic displacement (physiological activity changes) and behavioral intention. Figure 3 The modalities are cross-coupled to generate a user thermal muscle coupling behavior feature map with causal correlation characteristics, realizing the first high-dimensional unified representation of the user's hand control behavior. Through the behavioral intention decoding modeling of the feature map, it is possible to automatically learn the implicit intention pattern from the thermal muscle coupling behavior map, construct a nonlinear control state mapping map, and realize intelligent analysis of personalized user intentions with strong adaptability and high generalization ability. Based on the above mapping map, the user's current posture and control intention are dynamically matched to generate highly adaptive user mouse control data, which significantly improves the flexibility and precision of the input device in the human-computer interface and is suitable for a variety of interactive scenarios (such as games, drawing, remote control, etc.).

[0042] Preferably, step S4 includes the following steps:

[0043] Step S41: performing time-series dynamic direction fitting analysis on the user mouse control data to generate high-resolution direction control trajectory data; performing perturbation pattern reconstruction on the mouse jitter detection result to generate direction error interference data;

[0044] Step S42: performing nonlinear error inversion analysis on the high-resolution directional control trajectory data and the directional error interference data to generate compensation vector field data;

[0045] Step S43: analyzing the mouse control state of the user's mouse control data; performing multi-branch path prediction on the mouse direction control state based on the compensation vector field data to generate direction decision fusion data;

[0046] Step S44: performing adaptive compensation for the mouse direction based on the direction decision fusion data, generating an AI direction control response instruction, and performing an AI-based high-precision control operation for the mouse direction.

[0047] This invention utilizes time-series dynamic direction fitting technology to convert the user's raw mouse control data into high-resolution direction control trajectory data. It also models and reconstructs the perturbation behavior induced by jitter detection to generate direction error interference data, effectively capturing non-ideal input behavior and improving the perception of subtle errors. Using nonlinear error inversion analysis, the original control trajectory is compared with the error data to generate compensation vector field data. This dynamically provides feedback on the discrepancy between the user's control intent and actual action, establishing a personalized and adaptive error self-compensation mechanism. The system further analyzes the current mouse control state and, combined with the compensation vector field, performs multi-branch path prediction and direction fusion to generate directional decision fusion data, enhancing the system's forward-looking judgment of complex movement trends. This makes it suitable for high-frequency operations and complex interactive scenarios (such as CAD drawing and esports). Based on the directional decision fusion data, AI directional control response commands are generated, enabling real-time adaptive direction control compensation, significantly reducing trajectory deviation caused by physiological jitter, inertial deviation, or operational error, achieving a smoother and more natural human-computer interaction process. The overall process supports trajectory fine-tuning and direction optimization through data analysis and intelligent algorithms without relying on external hardware devices, greatly improving the expressiveness and stability of mouse control in scenarios such as high-precision operations, assisted input, and remote control.

[0048] Preferably, step S43 includes the following steps:

[0049] Step S431: identifying path branch points of the compensation vector field data;

[0050] Step S432: constructing a path candidate graph for the mouse control state based on the path branch points to generate direction evolution graph data;

[0051] Step S433: performing path feasibility prediction modeling on the direction evolution graph data to generate path prediction confidence matrix data;

[0052] Step S434: Perform confidence-weighted aggregation analysis on the path prediction confidence matrix data to generate direction decision fusion vector data.

[0053] By identifying the path branch points in the compensation vector field, the present invention can accurately locate the behavioral divergence or operation turning points in the user control process, capture the various evolution possibilities of potential control intentions, and enhance the sensitivity of behavioral reasoning. Based on the path branch points, a path candidate graph is constructed to generate directional evolution map data, which can simulate the extension trend of user operations in multiple directions and provide high-dimensional visualization prediction model support for complex dynamic control. By performing path feasibility modeling on the directional evolution map, path prediction confidence matrix data is generated, and a probability and confidence evaluation mechanism is introduced to provide a quantitative reference for path selection and reduce the probability of incorrect control path selection. Confidence-weighted aggregation analysis is performed based on the confidence matrix, and multi-path direction prediction results are integrated to generate directional decision fusion vector data to avoid misleading a single path and improve the stability and adaptability of the final control instruction.

[0054] In this specification, an AI-based high-precision mouse direction control device is provided, which is used to execute the above-mentioned AI-based high-precision mouse direction control method. The AI-based high-precision mouse direction control device includes:

[0055] The mouse movement path analysis module is used to obtain the basic position data of the mouse; based on the basic position data of the mouse, the mouse movement path is tracked to generate the mouse movement path data; the mouse movement path data is decomposed into a two-dimensional vector of the path point plane to generate the horizontal coordinate and the vertical coordinate of the mouse plane movement;

[0056] The mouse display module is used to obtain mouse display resolution data; calculate the direction change angle and the number of direction reversals of the mouse based on the horizontal coordinate and the vertical coordinate of the mouse plane movement; extract the moving edge jagged display features of the mouse display resolution data to obtain mouse movement jagged change feature data; perform mouse jitter detection on the mouse movement jagged change feature data based on the direction change angle and the number of direction reversals to obtain a mouse jitter detection result;

[0057] The mouse control module is used to collect user pressure data and electromyographic signals of the user's metacarpophalangeal joints using the mouse's built-in pressure sensor and electromyographic sensor based on the mouse jitter detection results; and perform time window stability analysis on the user pressure data and electromyographic signals of the user's metacarpophalangeal joints to obtain user mouse control data;

[0058] The adaptive adjustment module is used to adaptively compensate the mouse movement direction of the mouse jitter detection results based on the user's mouse control data, so as to perform high-precision control operations of the mouse direction based on AI.

[0059] The present invention provides a mouse comprising a pressure sensor array and an electromyographic sensor array, which are connected to a microcontroller and are used to execute the above-mentioned AI-based high-precision mouse direction control method.

[0060] The beneficial effect of the present invention is that the mouse movement path analysis module realizes high-precision capture of mouse movement and continuous trajectory reconstruction by real-time sampling of mouse basic position data and tracking of movement path, providing a solid foundation for subsequent analysis. By decomposing the path point into two-dimensional vectors, the horizontal and vertical coordinate motion data are separated, providing a detailed numerical basis for the subsequent calculation of the direction change angle and the number of direction reversals, so that the dynamic changes of the motion trajectory are quantified and systematized. The mouse display module automatically obtains the display resolution and extracts the edge jagged change characteristics in combination with the mouse movement data, adapts to the pixel density of different display devices, enhances the system's adaptability to hardware environment differences, and improves the accuracy of detection. Combined with the direction change angle, the number of direction reversals and the jagged change characteristics, the system can accurately detect mouse jitter, promptly identify unstable mouse operation behavior, and ensure the stability and smoothness of user input. The mouse control module uses pressure sensors and electromyography sensors to collect user presses and electromyography signals, combined with time window stability analysis, to achieve a deep understanding of the user's operation intentions and improve the naturalness and responsiveness of mouse control. The adaptive adjustment module performs adaptive directional compensation based on user control data and jitter detection results, intelligently correcting directional errors. This enables high-precision, low-latency mouse directional control, significantly improving user experience and operational efficiency. Therefore, through multi-sensor data fusion and AI adaptive compensation, this invention effectively solves the problems of path error, insufficient jitter recognition, and unstable response in traditional mouse directional control, achieving high-precision mouse directional control. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A flowchart of a method for high-precision control of mouse direction based on AI;

[0062] Figure 2 for Figure 1 Detailed implementation steps of step S2 in FIG.

[0063] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0064] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0065] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0066] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0067] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0068] To achieve this, please refer to Figures 1 to 3 , a high-precision mouse direction control method based on AI, the method comprising the following steps:

[0069] Step S1: Acquire basic mouse position data; track the mouse movement path based on the basic mouse position data to generate mouse movement path data; decompose the mouse movement path data into two-dimensional vectors of the path points to generate the horizontal coordinate and vertical coordinate of the mouse movement plane;

[0070] Step S2: obtaining mouse display resolution data; calculating the direction change angle and the number of direction reversals of the mouse according to the horizontal coordinate and the vertical coordinate of the mouse plane movement; performing movement edge jaggies display feature extraction on the mouse display resolution data to obtain mouse movement jaggies change feature data; performing mouse jitter detection on the mouse movement jaggies change feature data using the direction change angle and the number of direction reversals to obtain a mouse jitter detection result;

[0071] Step S3: collecting user pressing data and electromyographic signals of the user's metacarpophalangeal joints using the built-in pressure sensor and electromyographic sensor of the mouse according to the mouse jitter detection result; performing time window stability analysis on the user pressing data and the electromyographic signals of the user's metacarpophalangeal joints to obtain user mouse control data;

[0072] Step S4: Adaptively compensate the mouse movement direction of the mouse jitter detection result using the user mouse control data to perform high-precision mouse direction control based on AI.

[0073] This invention uses two-dimensional vector decomposition of path points to achieve high-resolution analysis of minute movement trajectories, significantly improving mouse control accuracy, which is particularly critical in high-resolution or drawing scenarios. It uses direction change angles and the number of direction reversals to sensitively capture user operation trajectories, enabling comprehensive perception and understanding of complex mouse motion behaviors. By integrating jagged edge features with directional data, it effectively detects mouse jitter caused by involuntary factors and eliminates erroneous input. By acquiring signals from the mouse's built-in pressure sensors and electromyographic sensors, the system improves its ability to distinguish between "conscious" and "unconscious" movements based on the user's hand muscle activity and pressing habits. An AI model learns from different user operation patterns to achieve adaptive directional compensation tailored to different physiological characteristics, meeting personalized human-computer interaction needs. The system is particularly suitable for users with Parkinson's disease and post-stroke sequelae, effectively compensating for involuntary mouse jitter during control and achieving a smoother input experience. Even in complex usage environments such as vibrating desktops and low-stability supports, it dynamically corrects erroneous motion paths to ensure operational stability. Through adaptive analysis of display resolution data, the system can adjust the control sensitivity according to the current display environment to achieve pixel-level positioning capability. Based on the stability analysis of the time window, it effectively captures the changes in the behavioral characteristics of users in continuous operations, so that the system can dynamically adapt to the operation rhythm. Multi-dimensional fusion data (position, speed, direction, physiological signals) forms a portrait of user control behavior, improving the judgment accuracy of the system under abnormal conditions. Through the multimodal AI learning model that combines jitter detection with electromyographic signal input, it promotes the transformation of traditional peripherals into more intelligent human-computer interaction devices. Therefore, the present invention effectively solves the problems of path error, insufficient jitter recognition and unstable response in traditional mouse direction control through multi-sensor data fusion and AI adaptive compensation, and realizes high-precision mouse direction control.

[0074] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a flow chart of a method for high-precision control of the mouse direction based on AI according to the present invention. In this example, the method for high-precision control of the mouse direction based on AI includes the following steps:

[0075] Step S1: Acquire basic mouse position data; track the mouse movement path based on the basic mouse position data to generate mouse movement path data; decompose the mouse movement path data into two-dimensional vectors of the path points to generate the horizontal coordinate and vertical coordinate of the mouse movement plane;

[0076] Step S2: obtaining mouse display resolution data; calculating the direction change angle and the number of direction reversals of the mouse according to the horizontal coordinate and the vertical coordinate of the mouse plane movement; performing movement edge jaggies display feature extraction on the mouse display resolution data to obtain mouse movement jaggies change feature data; performing mouse jitter detection on the mouse movement jaggies change feature data using the direction change angle and the number of direction reversals to obtain a mouse jitter detection result;

[0077] Step S3: collecting user pressing data and electromyographic signals of the user's metacarpophalangeal joints using the built-in pressure sensor and electromyographic sensor of the mouse according to the mouse jitter detection result; performing time window stability analysis on the user pressing data and the electromyographic signals of the user's metacarpophalangeal joints to obtain user mouse control data;

[0078] Step S4: Adaptively compensate the mouse movement direction of the mouse jitter detection result using the user mouse control data to perform high-precision mouse direction control based on AI.

[0079] In this embodiment of the present invention, a high-precision motion capture system samples the basic position of the mouse in real time, with a sampling frequency of 1kHz and a spatial resolution of less than 0.01mm, to obtain the original coordinate sequence of the mouse on a two-dimensional plane. The collected basic position data is imported into MATLAB R2024a software, and a custom script is used to track the mouse movement path in real time, generating movement path data with a timestamp. Subsequently, within the same environment, the MATLAB function library is used to perform two-dimensional vector decomposition of the path points, extracting the horizontal and vertical coordinate data of the mouse's planar movement. The decomposition results are output in a 0.5ms time window. The pixel resolution of the current display device (e.g., 1920×1080px) is obtained, and an adaptive filtering algorithm is used in MATLAB based on the horizontal and vertical coordinate data of the mouse's planar movement to calculate the direction change angle and the number of direction reversals at each sampling point. Simultaneously, the OpenCV library is used to extract aliasing features from the edge area of ​​the mouse cursor trajectory, and the aliasing edge density and amplitude are calculated to generate aliasing feature data for mouse movement. The direction change angle and the number of direction reversals are then input into the discrimination logic. When the sawtooth feature and sudden change in direction occur simultaneously, the mouse jitter detection result (a binary label of 0 / 1) is output. Based on the mouse jitter detection results, a micro pressure sensor and surface electromyography (sEMG) sensor module pre-installed inside the mouse collect user pressure torque data and metacarpophalangeal joint electromyography (EMG) signals in real time. The collected signals are transmitted to the LabVIEW 2023 environment via a USB interface. After preprocessing with a 1000Hz sampling rate and a 20–450Hz bandpass filter, stability analysis is performed within a 200ms sliding time window. The signal mean and variance within the window are calculated, and the user mouse control data is output. A hybrid C++ / Python AI control framework uses user mouse control data as input. A trained convolutional neural network model predicts the optimal movement direction in real time. Based on the prediction, the original jitter detection results are adaptively compensated for. The compensation algorithm runs on an edge computing platform equipped with an NVIDIA Jetson Xavier GPU, achieving a latency of less than 2ms and enabling high-precision AI-based mouse direction control.

[0080] Preferably, step S1 includes the following steps:

[0081] Step S11: using the user device input interface to sample the mouse pointer position in real time, capturing the screen coordinate value in each frame or time slice to obtain basic mouse position data;

[0082] Step S12: performing time series reconstruction, missing point interpolation and path fitting analysis on the basic position data of the mouse, extracting a continuous path sequence of the mouse motion trajectory, and generating mouse movement path data;

[0083] Step S13: performing vector processing on the coordinate difference between any two consecutive points in the mouse movement path data to form a displacement vector set between adjacent points, and generating two-dimensional plane vector data of the path;

[0084] Step S14: Decompose the horizontal and vertical coordinates of the two-dimensional plane vector data of the path to generate the horizontal coordinate and the vertical coordinate of the mouse plane movement.

[0085] In an embodiment of the present invention, the mouse pointer position is sampled in real time at a high frequency via the built-in input interface of a user device (e.g., a PC or laptop) by calling an operating system API function (e.g., GetCursorPos on Windows or CGEventCreate on macOS). The sampling frequency can be set to at least 240 frames per second. For each frame or time slice, the two-dimensional (x, y) coordinates of the current mouse pointer on the screen are captured and the corresponding timestamp is recorded, forming a time-stamped set of basic mouse position data. The basic mouse position data obtained in step S11 is first interpolated on the time axis to fill in missing points caused by system delays or sampling interference. The interpolation method can be linear or spline interpolation. Path fitting analysis is then performed on the complete time series data. A smooth path is fitted using the least squares method to remove noise points and reconstruct the continuous mouse movement trajectory at each moment. Ultimately, mouse movement path data containing the sequence of path points is generated. The difference between the coordinates of any two consecutive sampling points (xi, yi) and (xi+1, yi+1) in the mouse movement path data obtained in step S12 is calculated to define the two-dimensional plane displacement vector Vi = (xi+1-xi, yi+1-yi) between each pair of points. The entire path sequence is processed sequentially to obtain a displacement vector set consisting of multiple plane vectors for subsequent motion pattern analysis and direction recognition. Each two-dimensional vector Vi generated in step S13 is decomposed to extract its lateral displacement (Δx=xi+1-xi) and longitudinal displacement (Δy=yi+1-yi). Two parallel arrays are constructed to record the changes in the horizontal and vertical plane coordinates of the mouse movement, respectively. The generated horizontal and vertical coordinates of the mouse movement serve as the basic input for subsequent jitter detection and adaptive compensation analysis.

[0086] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0087] Step S21: Automatically detecting the resolution parameters of the display connected to the user terminal and calibrating the pixel density to obtain mouse display resolution data;

[0088] Step S22: performing coordinate sequence difference and arc cosine calculation on the horizontal coordinate and vertical coordinate of the mouse plane movement to generate the mouse direction change angle;

[0089] Step S23: performing signed gradient analysis on the direction change angle, detecting the positive and negative changes of the continuous direction vectors, counting the number of direction reversals, and generating the direction reversal count;

[0090] Step S24: analyzing the minimum discernible unit of the mouse display resolution data, and extracting the edge zigzag movement pattern of the mouse plane movement horizontal coordinate and the mouse plane movement vertical coordinate according to the minimum discernible unit, to generate mouse movement zigzag change feature data;

[0091] Step S25: performing abnormal movement fitting analysis on the mouse movement sawtooth change feature data based on the mouse direction change angle and the number of direction reversals to generate a mouse jitter detection result.

[0092] In an embodiment of the present invention, the resolution parameters (e.g., 1920×1080, 3840×2160, etc.) of the display connected to the user terminal are automatically obtained by calling the underlying system interface (such as the GetSystemMetrics function under the Windows platform or the CGDisplayPixelsWide / High under the Mac platform). In addition, by obtaining the physical size (in inches) and resolution of the display screen, the pixel density DPI (Dots Per Inch) is calculated to obtain the screen pixel offset value corresponding to the unit physical movement distance, and finally generate the mouse display resolution data for subsequent standardized judgment of the movement amount and the sawtooth amplitude. For the horizontal and vertical coordinates of the mouse plane movement obtained in step S14, the difference (Δx, Δy) between the coordinates of the two adjacent points is calculated respectively, and then the inverse cosine function (arccos) is used to calculate the angle θ between the two-dimensional displacement vector and the horizontal direction. The formula is as follows: ; The angle is output in the form of radians or degrees, and is recorded in sequence to form a sequence of mouse direction change angles for subsequent direction trend analysis. The direction angle sequence obtained in step S22 is subjected to sign gradient analysis. By calculating the difference Δθ between any two adjacent direction angles and judging the sign change (for example, from positive to negative or negative to positive), the direction reversal behavior is identified. Whenever a zero-crossing behavior of the sign from positive to negative or negative to positive is detected, it is regarded as a direction reversal. Finally, all reversals are counted to generate a direction reversal number index. Combined with the mouse display resolution data in step S21, the minimum discernible unit displacement under the current screen conditions (for example, the distance in millimeters corresponding to 1 pixel) is obtained. Analyze the horizontal and vertical coordinate sequences of the mouse plane movement to screen out those back-and-forth fluctuation behaviors with very small amplitudes (close to the minimum discernible unit) in space but high frequencies. The judgment criteria are as follows: If the continuous horizontal or vertical movement values ​​exhibit a frequent jump sequence such as "+1, -1, +1, -1" or "+2, -2, +2, -2"; and the amplitude does not exceed 2-3 times the minimum discernible unit, it is considered to be sawtooth oscillation. The position, amplitude, and frequency of these oscillation segments are extracted to generate mouse movement sawtooth variation feature data, which serves as an important basis for subsequent jitter detection. The direction change angle generated in step S22 and the number of direction reversals in step S23 are used as dynamic behavior features and are integrated with the sawtooth variation feature data extracted in step S24 for analysis. Methods such as sliding window linear regression, Fourier frequency decomposition, and wavelet energy analysis are used to detect the presence of nonlinear high-frequency fluctuation trends. Mouse jitter is preliminarily determined to occur if any of the following conditions are met: the number of direction reversals exceeds a set threshold (e.g., more than 20 reversals per second); the sawtooth oscillation characteristics are significant, with the jitter amplitude ranging from 1 to 3 pixels but at a very high frequency; and the direction change angle fluctuates violently, with an unusually high frequency of angle jumps greater than 90 degrees. The resulting mouse jitter detection results can be used for intelligent behavior recognition, graphics editing assistance, game auxiliary control, or system-level input anomaly recognition mechanism.

[0093] Preferably, step S25 includes the following steps:

[0094] Step S251: Calculate the first-order derivative of the mouse direction change angle according to the timestamp, extract the angle change rate, identify the sudden direction change behavior, and generate angular velocity change feature data;

[0095] Step S252: Counting the number of direction reversals per unit time and performing sliding window clustering to extract short-term high-frequency reversal features and generate direction reversal density distribution data;

[0096] Step S253: Calculating the frequency domain energy distribution of the mouse movement sawtooth change feature data;

[0097] Step S254: normalize the angular velocity change characteristic data, direction reversal density distribution data, and frequency domain energy distribution and fit them into a jitter judgment logic curve, and perform curve change detection on the jitter judgment logic curve based on a preset jitter change threshold to obtain a mouse jitter detection result.

[0098] In the embodiment of the present invention, the mouse direction change angle sequence (θ1, θ2, ..., θ n ) to calculate the first-order derivative, that is: ;in is the angular velocity of the i-th time slice, is the corresponding timestamp. By calculating the angular velocity change sequence of all directional angle point pairs, the angular velocity mutation point, peak position and average change rate are identified, and the angular velocity change feature data is extracted to determine whether the user has frequent and high-amplitude direction jumps within a unit time. If multiple angular velocity mutation points appear in a row, it indicates that there is uncontrolled or oscillatory offset behavior in the mouse movement. The number of direction reversals extracted in step S23 is time-normalized, and the number of reversals is counted within a certain time window (such as 100ms, 200ms, 500ms). Use the sliding window mechanism to segment the entire time series and construct the reversal density value per unit time: ; Subsequently, a cluster analysis algorithm (such as DBSCAN or K-Means) is used to classify the reversal density peaks to identify whether there are short-term high-frequency reversal areas, thereby generating direction reversal density distribution data for subsequent pattern recognition and fitting. For the mouse movement sawtooth change feature data extracted in step S24, this step introduces Fourier transform (FFT) or wavelet transform to perform frequency domain analysis on the sawtooth waveform signal: window processing is performed on the sawtooth movement sequence; the time domain sawtooth sequence is converted to the frequency domain using fast Fourier transform (FFT); the power spectral density (PSD) of each frequency component is calculated; the energy concentration area, main frequency range and amplitude distribution are analyzed. Finally, the energy distribution data of the sequence in the frequency domain is output to identify whether there are spectral characteristics of high-frequency jitter behavior, providing a vibration frequency basis for fitting analysis. The three types of feature data extracted in steps S251-S253, namely angular velocity change feature data, direction reversal density distribution data, and frequency domain energy distribution data, are sequentially normalized. Then, with time as the independent variable, the normalized data are comprehensively fitted into a jitter determination logic curve, for example, using a weighted logic function. Finally, the curve is tested for changes based on a set jitter threshold (e.g., detecting time periods continuously exceeding the threshold, areas where the fluctuation amplitude exceeds the standard, etc.) to determine whether the "mouse jitter" condition is met. If there are continuous time periods where the points on the jitter determination logic curve are greater than the jitter threshold, the output is "mouse jitter present"; otherwise, the output is "no jitter detected." The final mouse jitter detection result is output and can be used for interface calls, prompting users to adjust their operating methods, or as part of the system's auxiliary control logic.

[0099] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0100] Step S31: collecting user pressing data and myoelectric signals of the user's metacarpophalangeal joints using the built-in pressure sensor and myoelectric sensor of the mouse according to the mouse jitter detection result;

[0101] Step S32: performing spatial heat zone clustering on the user's pressing data to generate a user palm contact heat map; extracting isotherms from the palm contact heat map, and performing regional gradient analysis on the user palm contact heat map based on the isotherms to generate user static grip posture feature data;

[0102] Step S33: performing Fourier transform on the user's pressing pressure data to generate user's pressing frequency domain data; calculating the peak spacing of the user's pressing frequency domain data, and performing user pressing tremor recognition on the user's pressing pressure data based on the peak spacing, thereby generating periodic tremor pattern data;

[0103] Step S34: performing synchronous window registration on the periodic tremor pattern data and the electromyographic signal to generate interference fitting correction data; performing time window stability analysis on the user's pressing data and the electromyographic signal using the interference fitting correction data to generate multi-channel stability evaluation data;

[0104] Step S35: Perform joint behavior modeling using the static gripping posture feature data and the multi-channel stability evaluation data to generate user mouse control data.

[0105] In an embodiment of the present invention, a pressure sensor and an electromyography (EMG) sensor module are integrated into a mouse device to obtain physiological signals during contact between the user's palm and the mouse. The pressure sensor collects the user's continuous / instantaneous pressure on the mouse at the contact surface of the mouse housing (e.g., the palm rest and side button area). Muscle electrical activity in the user's palm and finger areas is collected to obtain EMG waveforms within the corresponding time period, reflecting physiological states such as hand muscle tension and contraction frequency. This data collection is triggered synchronously with the jitter detection results in step S25 to ensure temporal consistency between behavioral triggering and physiological signal acquisition. The user's pressure data over a period of time is mapped to a two-dimensional coordinate plane of the mouse contact area. A clustering algorithm (e.g., DBSCAN or MeanShift) is used to spatially cluster the contact point areas. This generates a palm contact heat map of the user's palm surface, with more concentrated hot zones indicating a higher degree of palm pressure concentration. The pressure contours of the heat map are calculated (isothermal extraction) to identify isobaric boundaries on the palm surface. The gradient direction and gradient amplitude of the isobaric zone are analyzed to extract the spatial structure of the palm pressure distribution when the user grips the palm. The heat zone shape (symmetry, offset center, etc.) is combined to generate static grip posture feature data for the user, which can be represented as a structural tensor or grip distribution matrix. A fast Fourier transform (FFT) or short-time Fourier transform (STFT) is performed on the pressure data sequence. The frequency domain energy spectrum is then determined to observe the presence of obvious periodic tremor characteristics (main peaks in the spectrum). The spacing between the main peaks in the frequency domain and the corresponding tremor period in the time domain are calculated. The presence of multiple evenly distributed main peaks in the frequency domain indicates typical tremor behavior (such as Parkinson's tremor). The tremor frequency, amplitude, and stability are combined to generate periodic tremor pattern data. Periodic tremor pattern data and electromyographic (EMG) signals are synchronized within time windows. Signal registration is established using a sliding window alignment algorithm (such as dynamic time warping (DTW)) or cross-correlation matching. Noise interference patterns corresponding to tremor are fitted in the EMG signals to generate interference-corrected data. The interference-corrected data are used to denoise the pressure and EMG signals. Signal stability (such as amplitude stability, variance consistency, and dominant frequency variation) is evaluated within each time window. The evaluation results from each channel are fused to generate multi-channel stability evaluation data for subsequent modeling. Two key data sources: static grip posture feature data (reflecting the relationship between the user's palm structure and mouse contact) and multi-channel stability evaluation data (reflecting the user's muscle stability control ability in this posture) are input into the behavioral modeling module. Using statistical modeling methods (such as Gaussian mixture models (GMMs)) or machine learning algorithms (such as random forests and lightweight neural networks), a user-customized mouse control model is constructed to learn the multivariate relationship between mouse jitter, posture, tremor, and stability. Output dynamic weight adjustment parameters (such as acceleration adjustment and sensitivity adaptation); can form a user mouse control data set for subsequent control optimization or input device parameter adjustment.

[0106] Preferably, step S35 includes the following steps:

[0107] Step S351: Perform multi-dimensional thermal distribution gradient reconstruction on the static holding posture feature data to generate posture thermal flow tensor data; perform frequency-amplitude-phase composite domain nonlinear deconstruction on the multi-channel stability evaluation data to generate stability hybrid feature tensor data;

[0108] Step S352: Using a tensor cross-gating network, the posture heat flow tensor data and the stability hybrid feature tensor data are cross-fused into a three-modal tensor to generate a user thermal muscle coupling behavior feature map, where the three modalities include thermal inertia, myoelectric displacement, and behavioral intention.

[0109] Step S353: Decoding and modeling the user's thermal-muscle coupling behavior characteristic map to generate a nonlinear control state map;

[0110] Step S354: Match the user's current posture control state according to the nonlinear control state mapping diagram to generate user mouse control data.

[0111] In this embodiment of the present invention, characteristic data of a user's static grip posture is acquired and multi-dimensional thermal distribution analysis is performed on this data. By constructing a gradient relationship for thermal distribution, a thermal flow data structure that reflects posture change trends is reconstructed. Simultaneously, stability assessment data from multiple channels, such as muscle tremor and hand fine-tuning, is collected and nonlinearly decomposed in terms of frequency, intensity, and phase to extract multiple stability features, which are then reconstructed into a composite feature data structure. A tensor-level cross-gated fusion method is used to deeply fuse the thermal flow data and stability feature data. During the fusion process, three key modalities are uniformly modeled and cross-aligned. These three modalities correspond to the user's thermal inertia characteristics, the deformation trend caused by muscle activity, and the potential behavioral intention expressed by the user during operation. Ultimately, a multi-layered graph is generated that comprehensively reflects the user's behavioral characteristics, representing the coupled relationship between thermal sensation, muscle response, and action intention. Subsequently, based on this constructed user feature graph, intent decoding is performed through a deep behavioral modeling mechanism. This decoding process no longer relies on linear inference models, but instead employs a nonlinear modeling strategy to gradually approximate the user's underlying control intent. During this process, the user's feature map is mapped into a multi-state control command space, thereby establishing a dynamic mapping relationship between user intent and control state. Finally, based on the established control state mapping relationship, the user's current hand gesture information is quickly matched to identify their current operational intent. Based on the matching results, specific control commands are generated and converted into mouse control data that the user can directly manipulate, achieving a precise and continuous human-computer interaction control experience.

[0112] Preferably, step S4 includes the following steps:

[0113] Step S41: performing time-series dynamic direction fitting analysis on the user mouse control data to generate high-resolution direction control trajectory data; performing perturbation pattern reconstruction on the mouse jitter detection result to generate direction error interference data;

[0114] Step S42: performing nonlinear error inversion analysis on the high-resolution directional control trajectory data and the directional error interference data to generate compensation vector field data;

[0115] Step S43: analyzing the mouse control state of the user's mouse control data; performing multi-branch path prediction on the mouse direction control state based on the compensation vector field data to generate direction decision fusion data;

[0116] Step S44: performing adaptive compensation for the mouse direction based on the direction decision fusion data, generating an AI direction control response instruction, and performing an AI-based high-precision control operation for the mouse direction.

[0117] In this embodiment of the present invention, dynamic directional fitting is performed on the mouse control data generated in step S1 (i.e., time-series movement path data). Using a locally weighted regression (e.g., LOESS) or sliding window vector fitting algorithm, high-precision modeling is performed on the directional change trends within continuous trajectory segments, thereby generating high-resolution directional control trajectory data. This data depicts the dynamic movement direction and subtle angular changes of the mouse pointer. Next, based on the horizontal and vertical coordinate data of the mouse movement extracted in the previous step, a perturbation analysis model is used to detect high-frequency jitter points (i.e., unintentional subtle movements) in the mouse movement path. Directional perturbations with characteristic frequencies are extracted using methods such as Fourier analysis or wavelet analysis, and their interference waveforms are reconstructed to generate directional error interference data describing abnormal directional perturbations. The high-resolution directional control trajectory data obtained in step S41 and the directional error interference data are subjected to nonlinear error inversion analysis. This process can be based on variational optimization methods (e.g., L-BFGS) or residual modeling of deep neural networks (e.g., RNN-ResNet architecture) to reconstruct the error spatial relationship between the ideal and actual directional trajectories. By comparing the ideal directional path with the error-perturbed path, global offset trends and local perturbation patterns are established. Compensation vector field data is generated based on this data, where each vector represents the directional fine-tuning that should be applied at the current path position to offset the error perturbation and achieve directional correction. The current user mouse control data is then used to identify whether it is in target tracking, drawing, text editing, or menu hovering. This state recognition can be performed using a multi-feature decision tree or a lightweight convolutional network. Based on the state recognition results, the compensation vector field data generated in step S42 is used to perform multi-branch path prediction for the current directional control state. Multiple corrected path branches are simulated (based on the current direction vector + compensation vector), and the cost functions of each branch path (such as target offset, curvature smoothness, and user intent match) are calculated. These are then fused into an optimal path solution to generate fusion directional decision data. Based on the fusion directional decision data obtained in step S43, the final directional adaptive compensation process is performed. This process can employ a dynamic weight adjustment mechanism to adaptively adjust the compensation effect in different states (e.g., providing stronger compensation in the drawing state and maintaining a sensitive response in the editing state). The compensated results will be converted into AI directional control response instructions, including digital commands for controlling the mouse movement direction, speed adjustment, and angle fine-tuning. These response instructions can be injected into the operating system interface through the system input simulation module to achieve AI-based high-precision mouse direction control tasks, significantly reducing directional offset and operation errors, and enhancing user experience and interaction efficiency.

[0118] Preferably, step S43 includes the following steps:

[0119] Step S431: identifying path branch points of the compensation vector field data;

[0120] Step S432: constructing a path candidate graph for the mouse control state based on the path branch points to generate direction evolution graph data;

[0121] Step S433: performing path feasibility prediction modeling on the direction evolution graph data to generate path prediction confidence matrix data;

[0122] Step S434: Perform confidence-weighted aggregation analysis on the path prediction confidence matrix data to generate direction decision fusion vector data.

[0123] In an embodiment of the present invention, potential path branch points in the current mouse movement trajectory are detected by utilizing compensation vector field data. Path branch points are key nodes in the spatial trajectory where multiple directional adjustment options exist, typically manifesting as locations where vector directions in the compensation vector field significantly diverge. The specific method includes calculating the gradient and divergence of the local vector field to identify regions with multimodal distributions of vector directions; utilizing a clustering algorithm (such as DBSCAN) to group adjacent vectors and distinguish between different path directions; and selecting a reasonable set of branch points incorporating time series continuity constraints. Based on the path branch points identified in step S431, a candidate path graph is constructed. This graph represents multiple mouse movement direction paths, with the path branch points as nodes and the candidate direction vectors connecting adjacent nodes as edges. The construction process includes connecting the path branch points in time series order to form a directed graph structure; assigning edge weights, adjusting the weights based on mouse control state characteristics (such as speed, acceleration, and the user's current operational intent); and generating complete directional evolution graph data, reflecting the diversity and dynamic evolution of directional paths. The path feasibility prediction model is applied to the direction evolution graph data from step S432 to evaluate the rationality and feasibility of each candidate path. Common methods include predicting path state transition probabilities based on a hidden Markov model (HMM) or a long short-term memory network (LSTM); comprehensively evaluating the path state transition probability by combining multi-dimensional features such as path curvature, path smoothness, and consistency with historical trajectories; and generating a path prediction confidence matrix covering all candidate paths, with each matrix element representing the prediction confidence of the corresponding path. More specifically, the path feasibility prediction model is constructed by collecting compensation vector field data, including the two-dimensional displacement vector of the mouse movement at each time point and its corresponding timestamp; simultaneously collecting user mouse control state data (such as auxiliary features such as speed, acceleration, and pressure sensor output); annotating path branch points and their subsequent true trajectory directions (for supervised learning). Missing values ​​and outliers are handled, and interpolation and smoothing filtering methods are used to ensure trajectory continuity. Data augmentation techniques are used to generate more path variation samples (such as adding noise and slightly deforming the trajectory) to improve model generalization. Multidimensional features are extracted from path branch points and adjacent trajectories: geometric features (current direction vector, curvature, trajectory change rate); time series features (speed, acceleration, time interval); and user behavior features (pressure intensity, electromyography). A sequence feature vector is constructed as model input. According to step S432, the path branch points are treated as graph nodes, and the edges represent candidate directional paths. Each edge is assigned a preliminary weight, such as path smoothness and consistency with historical trajectories. The graph structure data is encoded into a form acceptable to a graph neural network (GNN), with node and edge features serving as input.This approach combines a graph neural network (GNN) with a time series model (such as LSTM) to balance path structure and temporal dynamics. The GNN module processes spatial dependencies in the candidate path graph and learns high-order interaction features between nodes and edges. The LSTM module processes time series features to capture temporal variations in trajectories. The model's input layer receives node and edge feature matrices; the GNN layer uses a multi-layer graph convolutional network (GCN or GraphSAGE) to extract spatial features; the sequence layer uses an LSTM or GRU network to receive GNN outputs and time series features; the fully connected layer outputs a feasibility score (confidence) for each path; and the Softmax layer normalizes the score to a probability distribution. A cross-entropy loss function is used for supervised learning to train path classification (feasible vs. infeasible paths). A path order constraint loss can be optionally added to ensure that the predicted path conforms to trajectory continuity. Training is performed in batches using the Adam optimizer with an appropriate learning rate and regularization term. Accuracy and loss on the validation set are monitored to avoid overfitting, and an early stopping strategy is used to improve generalization. The system inputs real-time compensation vector field data and mouse control state to construct a candidate path graph. The trained model is used to calculate the feasibility confidence score for each candidate path. A path prediction confidence matrix is ​​output, with each matrix element corresponding to the confidence score of each path. The direction vectors of the candidate paths are weighted and summed according to the confidence matrix. Normalization is performed to avoid weight imbalance, generating the final fusion direction decision vector data, which serves as input for the subsequent adaptive compensation module. The path feasibility prediction model is combined with the mouse control state recognition module to improve overall robustness. This is combined with the real-time data acquisition module to achieve online prediction. Deployed on an edge computing platform or local device, the system ensures low-latency response. Based on the path prediction confidence matrix obtained in step S433, confidence-weighted aggregation is performed to fuse the prediction results of each path and form a unified direction decision output. This implementation includes: multiplying the direction vectors of all candidate paths by their corresponding confidence scores as weights and performing a weighted summation. Through normalization, the final fusion direction decision vector data is generated, which accurately reflects the optimal adjustment direction for the current mouse direction. This fusion vector serves as the key input for subsequent direction compensation and control response steps.

[0124] In this specification, an AI-based high-precision mouse direction control device is provided, which is used to execute the above-mentioned AI-based high-precision mouse direction control method. The AI-based high-precision mouse direction control device includes:

[0125] The mouse movement path analysis module is used to obtain the basic position data of the mouse; based on the basic position data of the mouse, the mouse movement path is tracked to generate the mouse movement path data; the mouse movement path data is decomposed into a two-dimensional vector of the path point plane to generate the horizontal coordinate and the vertical coordinate of the mouse plane movement;

[0126] The mouse display module is used to obtain mouse display resolution data; calculate the direction change angle and the number of direction reversals of the mouse based on the horizontal coordinate and the vertical coordinate of the mouse plane movement; extract the moving edge jagged display features of the mouse display resolution data to obtain mouse movement jagged change feature data; perform mouse jitter detection on the mouse movement jagged change feature data based on the direction change angle and the number of direction reversals to obtain a mouse jitter detection result;

[0127] The mouse control module is used to collect user pressure data and electromyographic signals of the user's metacarpophalangeal joints using the mouse's built-in pressure sensor and electromyographic sensor based on the mouse jitter detection results; and perform time window stability analysis on the user pressure data and electromyographic signals of the user's metacarpophalangeal joints to obtain user mouse control data;

[0128] The adaptive adjustment module is used to adaptively compensate the mouse movement direction of the mouse jitter detection results based on the user's mouse control data, so as to perform high-precision control operations of the mouse direction based on AI.

[0129] The present invention provides a mouse comprising a pressure sensor array and an electromyographic sensor array, which are connected to a microcontroller and are used to execute the above-mentioned AI-based high-precision mouse direction control method.

[0130] The beneficial effect of the present invention is that the mouse movement path analysis module realizes high-precision capture of mouse movement and continuous trajectory reconstruction by real-time sampling of mouse basic position data and tracking of movement path, providing a solid foundation for subsequent analysis. By decomposing the path point into two-dimensional vectors, the horizontal and vertical coordinate motion data are separated, providing a detailed numerical basis for the subsequent calculation of the direction change angle and the number of direction reversals, so that the dynamic changes of the motion trajectory are quantified and systematized. The mouse display module automatically obtains the display resolution and extracts the edge jagged change characteristics in combination with the mouse movement data, adapts to the pixel density of different display devices, enhances the system's adaptability to hardware environment differences, and improves the accuracy of detection. Combined with the direction change angle, the number of direction reversals and the jagged change characteristics, the system can accurately detect mouse jitter, promptly identify unstable mouse operation behavior, and ensure the stability and smoothness of user input. The mouse control module uses pressure sensors and electromyography sensors to collect user presses and electromyography signals, combined with time window stability analysis, to achieve a deep understanding of the user's operation intentions and improve the naturalness and responsiveness of mouse control. The adaptive adjustment module performs adaptive directional compensation based on user control data and jitter detection results, intelligently correcting directional errors. This enables high-precision, low-latency mouse directional control, significantly improving user experience and operational efficiency. Therefore, through multi-sensor data fusion and AI adaptive compensation, this invention effectively solves the problems of path error, insufficient jitter recognition, and unstable response in traditional mouse directional control, achieving high-precision mouse directional control.

[0131] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0132] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A high-precision mouse direction control method based on AI, characterized in that: The following steps are involved: Step S1: Obtain basic mouse position data; Track the movement path of the mouse based on the basic position data of the mouse to generate mouse movement path data; Decomposing the mouse movement path data into a two-dimensional vector of the path points to generate the horizontal coordinate and vertical coordinate of the mouse movement plane; Step S2: Obtaining mouse display resolution data; Calculate the direction change angle and direction reversal times of the mouse according to the horizontal coordinate and vertical coordinate of the mouse plane movement; Extracting the moving edge jagged display features of the mouse display resolution data to obtain the mouse movement jagged change feature data; performing mouse jitter detection on the mouse movement jagged change feature data based on the direction change angle and the number of direction reversals to obtain the mouse jitter detection result; wherein step S2 includes the following steps: Step S21: Automatically detecting the resolution parameters of the display connected to the user terminal and calibrating the pixel density to obtain mouse display resolution data; Step S22: performing coordinate sequence difference and arc cosine calculation on the horizontal coordinate and vertical coordinate of the mouse plane movement to generate the mouse direction change angle; Step S23: performing signed gradient analysis on the direction change angle, detecting the positive and negative changes of the continuous direction vectors, counting the number of direction reversals, and generating the direction reversal count; Step S24: analyzing the minimum discernible unit of the mouse display resolution data, and extracting the edge zigzag movement pattern of the mouse plane movement horizontal coordinate and the mouse plane movement vertical coordinate according to the minimum discernible unit, to generate mouse movement zigzag change feature data; Step S25: performing abnormal movement fitting analysis on the mouse movement sawtooth change feature data based on the mouse direction change angle and the number of direction reversals to generate a mouse jitter detection result; Step S3: collecting user pressing data and electromyographic signals of the user's metacarpophalangeal joints using the built-in pressure sensor and electromyographic sensor of the mouse according to the mouse jitter detection result; performing time window stability analysis on the user pressing data and the electromyographic signals of the user's metacarpophalangeal joints to obtain user mouse control data; Step S4: Adaptively compensate the mouse movement direction of the mouse jitter detection result using the user mouse control data to perform high-precision mouse direction control based on AI.

2. The AI-based high-precision mouse direction control method according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: using the user device input interface to sample the mouse pointer position in real time, capturing the screen coordinate value in each frame or time slice to obtain basic mouse position data; Step S12: performing time series reconstruction, missing point interpolation and path fitting analysis on the basic position data of the mouse, extracting a continuous path sequence of the mouse motion trajectory, and generating mouse movement path data; Step S13: performing vector processing on the coordinate difference between any two consecutive points in the mouse movement path data to form a displacement vector set between adjacent points, and generating two-dimensional plane vector data of the path; Step S14: Decompose the horizontal and vertical coordinates of the two-dimensional plane vector data of the path to generate the horizontal coordinate and the vertical coordinate of the mouse plane movement.

3. The AI-based high-precision mouse direction control method according to claim 1, characterized in that: Step S25 includes the following steps: Step S251: Calculate the first-order derivative of the mouse direction change angle according to the timestamp, extract the angle change rate, identify the sudden direction change behavior, and generate angular velocity change feature data; Step S252: Counting the number of direction reversals per unit time and performing sliding window clustering to extract short-term high-frequency reversal features and generate direction reversal density distribution data; Step S253: Calculating the frequency domain energy distribution of the mouse movement sawtooth change feature data; Step S254: normalize the angular velocity change characteristic data, direction reversal density distribution data, and frequency domain energy distribution and fit them into a jitter judgment logic curve, and perform curve change detection on the jitter judgment logic curve based on a preset jitter change threshold to obtain a mouse jitter detection result.

4. The AI-based high-precision mouse direction control method according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: collecting user pressing data and myoelectric signals of the user's metacarpophalangeal joints using the built-in pressure sensor and myoelectric sensor of the mouse according to the mouse jitter detection result; Step S32: performing spatial heat zone clustering on the user's pressing data to generate a user palm contact heat map; extracting isotherms from the palm contact heat map, and performing regional gradient analysis on the user palm contact heat map based on the isotherms to generate user static grip posture feature data; Step S33: performing Fourier transform on the user's pressing pressure data to generate user's pressing frequency domain data; calculating the peak spacing of the user's pressing frequency domain data, and performing user pressing tremor recognition on the user's pressing pressure data based on the peak spacing, thereby generating periodic tremor pattern data; Step S34: performing synchronous window registration on the periodic tremor pattern data and the electromyographic signal to generate interference fitting correction data; performing time window stability analysis on the user's pressing data and the electromyographic signal using the interference fitting correction data to generate multi-channel stability evaluation data; Step S35: Perform joint behavior modeling using the static gripping posture feature data and the multi-channel stability evaluation data to generate user mouse control data.

5. The AI-based high-precision mouse direction control method according to claim 4, characterized in that: Step S35 includes the following steps: Step S351: Perform multi-dimensional thermal distribution gradient reconstruction on the static holding posture feature data to generate posture thermal flow tensor data; perform frequency-amplitude-phase composite domain nonlinear deconstruction on the multi-channel stability evaluation data to generate stability hybrid feature tensor data; Step S352: Using a tensor cross-gating network, the posture heat flow tensor data and the stability hybrid feature tensor data are cross-fused into a three-modal tensor to generate a user thermal muscle coupling behavior feature map, where the three modalities include thermal inertia, myoelectric displacement, and behavioral intention. Step S353: Decoding and modeling the user's thermal-muscle coupling behavior characteristic map to generate a nonlinear control state map; Step S354: Match the user's current posture control state according to the nonlinear control state mapping diagram to generate user mouse control data.

6. The AI-based high-precision mouse direction control method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing time-series dynamic direction fitting analysis on the user mouse control data to generate high-resolution direction control trajectory data; performing perturbation pattern reconstruction on the mouse jitter detection result to generate direction error interference data; Step S42: performing nonlinear error inversion analysis on the high-resolution directional control trajectory data and the directional error interference data to generate compensation vector field data; Step S43: analyzing the mouse control state of the user's mouse control data; performing multi-branch path prediction on the mouse direction control state based on the compensation vector field data to generate direction decision fusion data; Step S44: performing adaptive compensation for the mouse direction based on the direction decision fusion data, generating an AI direction control response instruction, and performing an AI-based high-precision control operation for the mouse direction.

7. The AI-based high-precision mouse direction control method according to claim 6, characterized in that: Step S43 includes the following steps: Step S431: identifying path branch points of the compensation vector field data; Step S432: constructing a path candidate graph for the mouse control state based on the path branch points to generate direction evolution graph data; Step S433: performing path feasibility prediction modeling on the direction evolution graph data to generate path prediction confidence matrix data; Step S434: Perform confidence-weighted aggregation analysis on the path prediction confidence matrix data to generate direction decision fusion vector data.

8. A high-precision mouse direction control device based on AI, characterized in that: For executing the AI-based high-precision mouse direction control method according to claim 1, the AI-based high-precision mouse direction control device comprises: The mouse movement path analysis module is used to obtain the basic position data of the mouse; based on the basic position data of the mouse, the mouse movement path is tracked to generate the mouse movement path data; the mouse movement path data is decomposed into a two-dimensional vector of the path point plane to generate the horizontal coordinate and the vertical coordinate of the mouse plane movement; The mouse display module is used to obtain mouse display resolution data; calculate the direction change angle and the number of direction reversals of the mouse based on the horizontal coordinate and the vertical coordinate of the mouse plane movement; extract the moving edge jagged display features of the mouse display resolution data to obtain mouse movement jagged change feature data; perform mouse jitter detection on the mouse movement jagged change feature data based on the direction change angle and the number of direction reversals to obtain a mouse jitter detection result; The mouse control module is used to collect user pressure data and electromyographic signals of the user's metacarpophalangeal joints using the mouse's built-in pressure sensor and electromyographic sensor based on the mouse jitter detection results; and perform time window stability analysis on the user pressure data and electromyographic signals of the user's metacarpophalangeal joints to obtain user mouse control data; The adaptive adjustment module is used to adaptively compensate the mouse movement direction of the mouse jitter detection results based on the user's mouse control data, so as to perform high-precision control operations of the mouse direction based on AI.

9. A mouse, characterized in that: The method comprises a pressure sensor array and an electromyographic sensor array, which are connected to a microcontroller and used to execute the AI-based high-precision mouse direction control method as described in any one of claims 1 to 7.

Images