Mouse gait analysis method based on following shooting

Abstract

The invention discloses a following shooting-based mouse gait analysis method, which comprises the following steps of: obtaining an initial video image of a mouse activity area, and separating a mouse movement area through a preset background model to obtain a first image; calculating a motion vector of the mouse according to the first image to obtain motion trend data; adjusting the shooting angle and the focal length of the camera according to the motion trend data to obtain a second image; according to the second image, adjusting image brightness distribution, eliminating the influence of uneven illumination, and obtaining a third image; based on the third image, extracting contour features of the mouse, and obtaining gait feature data in combination with a pre-established gait model; according to the gait feature data, updating a tracking path of the camera to obtain a fourth image; enhancing the image definition according to the fourth image to obtain a fifth image; based on the fifth image, key joint positions of the mouse are recognized through a deep learning model, and a gait analysis report is generated in combination with gait feature data.

Description

Technical Field

[0001] The present invention belongs to the field of machine vision and deep learning technology, and in particular relates to a mouse gait analysis method based on follow-up shooting. Background Art

[0002] In the mouse gait analysis method based on follow-up shooting, a key technical problem is how to continuously and stably capture the video images of the mouse walking or running when the mouse is free to move. Due to the small size of mice, fast movement speed, and unpredictable activity path, the traditional fixed-position shooting method is difficult to meet the needs of continuous tracking shooting. In addition, in the environment where mice move freely, the background is complex and changeable, and the lighting conditions are uneven, which will have an adverse effect on the quality of video images, and then affect the accuracy of subsequent image processing and gait analysis. At the same time, due to the large range of mouse activities, how to achieve large-scale tracking and shooting of mice while ensuring the quality of video images is also an urgent problem to be solved. Therefore, how to design a flexible and intelligent follow-up shooting system that can adapt to the characteristics of free movement of mice and achieve stable, clear and comprehensive video image acquisition is a key technical challenge faced by the mouse gait analysis method based on follow-up shooting. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a mouse gait analysis method based on follow-up shooting, which realizes accurate analysis and long-term monitoring of mouse behavior.

[0004] To achieve the above object, the present invention provides a mouse gait analysis method based on follow-up shooting, comprising: acquiring an initial video image of the mouse activity area, separating the mouse movement area through a preset background model, and obtaining a first image;

[0005] According to the first image, calculating the motion vector of the mouse, determining the moving direction and speed of the mouse, and obtaining motion trend data;

[0006] According to the motion trend data, adjusting the shooting angle and focal length of the camera, if the moving speed of the mouse exceeds a preset threshold, starting the fast tracking mode to obtain a second image;

[0007] According to the second image, adjusting the image brightness distribution to eliminate the influence of uneven illumination, so as to obtain a third image;

[0008] Extracting the contour features of the mouse based on the third image, and judging the gait state of the mouse in combination with a pre-established gait model to obtain gait feature data;

[0009] According to the gait feature data, the tracking path of the camera is updated, and if the activity path of the mouse deviates from the preset range, the shooting area is replanned to obtain a fourth image;

[0010] According to the fourth image, enhancing image clarity and eliminating background noise to obtain a fifth image;

[0011] Based on the fifth image, the key joint positions of the mouse are identified through a deep learning model, and a gait analysis report is generated in combination with the gait feature data.

[0012] Technical effect of the present invention: The present invention discloses a mouse gait analysis method based on follow-up shooting, which analyzes the mouse movement trend through background separation and optical flow method, and adjusts the camera parameters in real time to maintain stable tracking. Adaptive illumination compensation and multi-frame fusion technology are used to improve image quality, and edge detection and deep learning models are combined to extract mouse gait characteristics. The present invention can continuously and stably capture mouse activities in complex environments, automatically adjust the shooting range and parameters, and effectively solve the problems of traditional methods such as tracking difficulties and unstable image quality in free activity environments. Through the coordinated application of intelligent algorithms and multiple image processing technologies, the present invention realizes accurate analysis and long-term monitoring of mouse behavior, providing reliable data support and technical guarantees for related research. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0014] Figure 1 Schematic diagram of the process of the mouse gait analysis method based on follow-up shooting according to an embodiment of the present invention;

[0015] Figure 2 A schematic diagram of a process of obtaining a first image according to an embodiment of the present invention;

[0016] Figure 3 A schematic diagram of a process for obtaining motion trend data according to an embodiment of the present invention;

[0017] Figure 4 A schematic diagram of a process of obtaining a second image according to an embodiment of the present invention;

[0018] Figure 5 A schematic diagram of a process for obtaining gait feature data according to an embodiment of the present invention;

[0019] Figure 6 A schematic diagram of a process for generating a gait analysis report according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0021] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0022] like Figure 1 As shown, this embodiment provides a mouse gait analysis method based on follow-up shooting, including: acquiring an initial video image of the mouse activity area, separating the mouse movement area through a preset background model, and obtaining a first image;

[0023] According to the first image, calculating the motion vector of the mouse, determining the moving direction and speed of the mouse, and obtaining motion trend data;

[0024] According to the motion trend data, adjusting the shooting angle and focal length of the camera, if the moving speed of the mouse exceeds a preset threshold, starting the fast tracking mode to obtain a second image;

[0025] According to the second image, adjusting the image brightness distribution to eliminate the influence of uneven illumination, so as to obtain a third image;

[0026] Extracting the outline features of the mouse based on the third image, and judging the gait state of the mouse in combination with a pre-established gait model to obtain gait feature data;

[0027] According to the gait feature data, the tracking path of the camera is updated, and if the activity path of the mouse deviates from the preset range, the shooting area is replanned to obtain a fourth image;

[0028] According to the fourth image, enhancing image clarity and eliminating background noise to obtain a fifth image;

[0029] Based on the fifth image, the key joint positions of the mouse are identified through a deep learning model, and a gait analysis report is generated in combination with the gait feature data.

[0030] Further, such as Figure 2 As shown, obtaining the first image includes:

[0031] S1, obtaining an initial video image sequence of the mouse activity area, and performing background separation processing using a preset background model on each frame of the initial video image sequence to obtain a binary image sequence of the mouse movement area;

[0032] S2, based on the motion region binary image sequence, removing the noise region in the image to obtain a first denoised image sequence;

[0033] S3. Based on the first image sequence, extract the complete motion area of ​​the mouse to obtain a second image sequence;

[0034] S4. Based on the second image sequence, using a convolutional neural network model to identify and classify the movement posture of the mouse to obtain a mouse movement posture category sequence;

[0035] S5. Based on the mouse motion posture category sequence, the behavior pattern of the mouse is modeled by a hidden Markov model to obtain a state transition matrix of the mouse behavior pattern;

[0036] S6, collecting the target mouse activity video image, repeating the processing from S1 to S4, and obtaining the target mouse movement posture category sequence;

[0037] S7, inputting the target mouse motion posture category sequence into the hidden Markov model of S5 for decoding to obtain a first image.

[0038] Further, such as Figure 3 As shown, obtaining motion trend data includes:

[0039] Based on the first image, preprocessing the first image to obtain a preprocessed image;

[0040] Based on the preprocessed image, calculating the motion vectors of pixels between adjacent frames to obtain pixel-level motion information;

[0041] According to the pixel-level motion information, a mouse target in the image is identified, and the position and size of the mouse are determined;

[0042] Track the mouse target in multiple consecutive frames of images, obtain the mouse's motion trajectory, and calculate the mouse's motion trajectory and speed information;

[0043] According to the movement trajectory and speed information of the mouse, the movement trend of the mouse is predicted to obtain the movement trend data of the mouse in the future period of time, and the movement trend data includes position coordinates and moving speed.

[0044] In a specific application example, assuming that the position of the mouse at time t is (100, 150) and the speed is (5, 3), the Kalman filter can predict the position at time t+1 and make corrections based on the actual observation results to obtain a more accurate trajectory estimate. Based on the obtained trajectory data, the direction and speed of the mouse can be calculated. For example, the instantaneous speed can be obtained by calculating the difference between adjacent positions, and the direction can be determined by the angle of the displacement vector. This information provides a basis for subsequent behavioral analysis. Kalman filtering is not only used for tracking, but also for predicting future movement trends. By establishing a motion model, such as a uniform motion or uniformly accelerated motion model, the filter can extrapolate the possible position of the mouse in the next few frames. This prediction is particularly useful for dealing with occlusion or fast motion scenes. Statistical analysis of the acquired motion data can obtain features such as average speed and acceleration. For example, calculating the mean and standard deviation of the speed over a period of time can reflect the overall intensity and fluctuation of the mouse's activities. These statistical features provide quantitative indicators for studying mouse behavior patterns. Finally, by comparing the analysis results with prior knowledge, abnormal situations can be discovered in a timely manner. For example, if mice are observed to maintain high-speed movements for a long time or repeatedly wandering in the same location, it may indicate abnormal behavior or problems with the experimental environment. This abnormality detection mechanism helps ensure the reliability of experimental data and timely intervention of potential health problems.

[0045] Further, such as Figure 4 As shown, obtaining the second image includes:

[0046] Based on the position coordinates in the obtained motion trend data, the motion trajectory of the mouse is predicted to obtain the predicted position of the mouse at the next moment;

[0047] According to the predicted position of the mouse, the camera's shooting angle is controlled by the servo motor so that the camera's shooting center coincides with the predicted position of the mouse, and the displacement of the mouse between consecutive frames is calculated to obtain the mouse's real-time moving speed;

[0048] A video frame in a fast tracking mode is acquired through a camera, and image enhancement and denoising are performed on the video frame to obtain a second image.

[0049] Specific application examples: Obtaining mouse movement trend data is the basis of behavioral analysis. Position coordinates are usually in pixels, such as (320,240) means that the mouse is in the center of the image. The movement speed can be expressed in pixels / second, such as 10 pixels / second. These data provide an important basis for subsequent analysis. The Kalman filter algorithm plays a key role in predicting the mouse's movement trajectory. The algorithm combines previous states and current measurements to estimate the optimal prediction value. Assuming that the current position of the mouse is (100,100) and the speed is 5 pixels / second, the algorithm may predict that the position in the next second will be (105,105). This prediction helps the camera to adjust in time to keep the mouse in the center of the picture. The servo motor controls the camera shooting angle, which is the key to achieving accurate tracking. Depending on the predicted position, the motor may need to adjust the angle by about 0.5 degrees. This small adjustment ensures the stability of the picture while keeping the mouse in the center of the field of view. The optical flow method calculates the mouse displacement, which is the core technology of real-time tracking. By comparing the changes in pixels in consecutive frames, the direction and speed of the mouse's movement can be obtained. If a horizontal displacement of 20 pixels is found and the frame rate is 30fps, the horizontal speed of the mouse can be calculated to be about 600 pixels / second. The threshold setting for starting the fast tracking mode is crucial. Assuming that the normal movement speed is 5 pixels / frame, the threshold can be set to 15 pixels / frame. Once the speed is detected to exceed this threshold, the system immediately switches to the fast tracking mode to improve the sampling rate and tracking accuracy. Dynamic adjustment of the camera focus is the key to ensuring image clarity. When moving quickly, the focal length may need to be adjusted from 50mm to 35mm to expand the field of view. This adjustment not only ensures that the mouse is always in the center of the picture, but also captures more surrounding environment information, which is conducive to a comprehensive analysis of mouse behavior. Image enhancement and denoising are essential for obtaining a high-quality second image. Gaussian filtering can be used to remove noise, and then histogram equalization can be used to increase contrast. These processes make the mouse outline clearer, which is convenient for subsequent behavioral analysis, such as posture recognition or specific action detection. Through the comprehensive application of this series of technologies, the movement trajectory and behavioral characteristics of the mouse can be accurately captured.

[0050] Further, obtaining the third image includes:

[0051] Based on the brightness distribution information of the second image, determining an area in the image where uneven illumination exists;

[0052] According to the uneven lighting area, dynamically adjust the brightness compensation coefficient in the area;

[0053] Adaptively adjusting the pixel brightness value of the uneven illumination area according to the compensation coefficient to eliminate the influence of the uneven illumination;

[0054] Smoothing the adjusted pixel brightness values ​​to eliminate sudden brightness changes and obtain an image area with uniform illumination;

[0055] The processed uneven illumination area is fused with the uniform illumination area in the original image to obtain an image with uniform illumination as a whole, and the image with uniform illumination as a whole is subjected to contrast enhancement processing to obtain a third image.

[0056] Further, such as Figure 5 As shown, obtaining gait feature data includes:

[0057] Preprocessing the third image to obtain a preprocessed image;

[0058] Analyze the preprocessed image, extract the contour features of the mouse in the image, and obtain a binary image of the mouse contour;

[0059] According to the pre-established gait model, the extracted mouse contour features are analyzed, and the current mouse gait state is determined through model matching and feature comparison;

[0060] If the mouse gait state conforms to the preset gait pattern, the current gait feature data is saved as input for subsequent analysis; otherwise, the next frame of image is processed;

[0061] According to the stored gait feature data, similar gait features are grouped into one category to obtain feature sets under different gait states;

[0062] The feature sets under different gait states are trained, a gait state classifier is established, the gait state of mice is tracked and analyzed in real time, and continuous gait feature data is obtained.

[0063] Further, obtaining the fourth image includes:

[0064] According to the gait feature data, the next movement trajectory of the mouse is predicted, and the tracking path of the camera is updated in real time;

[0065] Compare the mouse's activity path with the preset range to determine whether the mouse deviates from the preset range. If the mouse deviates from the preset range, the camera is triggered to re-plan the shooting area.

[0066] Perform semantic segmentation on the mouse activity area to determine the optimal shooting area;

[0067] Pass the optimal shooting area parameters to the camera, adjust the camera parameters, and reacquire the shooting area;

[0068] A fourth image is acquired according to the reacquired parameters of the photographing area.

[0069] Specifically, the outline and posture information of the mouse can be obtained by preprocessing, edge detection and feature extraction of the image. For example, the Canny edge detection algorithm can accurately depict the outline of the mouse body, while the skeleton extraction algorithm can identify the position of the joints. These feature data lay the foundation for subsequent analysis. The Kalman filter algorithm performs well in predicting the trajectory of mouse movement. By combining historical position data and current observations, the algorithm can effectively estimate the position and speed of the mouse at the next moment. Assuming that the current position of the mouse is (10,15) and the speed is (2,1), the Kalman filter may predict the position at the next moment to be (12,16). This prediction ability enables the camera to adjust in advance to ensure that the mouse is always kept in the center of the field of view. The setting of the preset range is crucial for experimental control. The experimental area can be divided into multiple sub-areas, such as the central area, the edge area, etc. When the mouse position coordinates exceed the preset range, the system will trigger an alarm. For example, if the preset range is a square area of ​​20×20cm, and the mouse moves to the position of (22,18), the system will determine that it deviates from the preset range. Deep learning algorithms play an important role in optimizing the shooting area. Semantic segmentation networks such as U-Net can accurately identify mice, backgrounds, and other objects in images. By analyzing the segmentation results, the system can determine the main activity area of ​​the mice and adjust the camera parameters accordingly. For example, if the mice are found to be frequently active in the upper right corner of the image, the system will automatically adjust the camera focus and angle to place this area in the center of the shot. Camera parameter adjustment is the key to achieving accurate tracking. Based on the calculated optimal shooting area, the system may need to adjust parameters such as focal length, aperture, and shutter speed. For example, when the mouse moves quickly, the shutter speed may need to be increased to 1 / 500 second to capture a clear image; in low light conditions, the aperture may need to be increased to f / 2.8 to obtain sufficient exposure. The advantage of this system lies in its adaptability and precision. Through real-time analysis and adjustment, it can always accurately track and record the behavior of mice. This not only improves the quality and efficiency of data collection, but also captures subtle behavioral changes that may be overlooked by manual observation.

[0070] Further, obtaining the fifth image includes:

[0071] Extracting feature information of each frame of the image according to the fourth image and constructing a feature matrix;

[0072] According to the characteristic matrix, determining the main characteristic components, removing the secondary components and background noise;

[0073] Reconstructing an image matrix according to the main characteristic components to obtain a preliminary fused image;

[0074] According to the preliminary fused image, performing image enhancement processing in the frequency domain by using wavelet transform to obtain an enhanced frequency domain image;

[0075] Reconstructing the enhanced frequency domain image by using inverse wavelet transform to obtain a reconstructed time domain image;

[0076] According to the reconstructed time domain image, residual noise is removed to obtain a fifth image.

[0077] Specifically, image sequence feature extraction is the basis of multi-frame image fusion. By calculating the grayscale histogram of each frame image, edge detection results, etc., a feature matrix can be constructed. For example, for a set of mouse motion image sequences, the features such as the mouse outline and position in each frame are extracted to form a feature matrix describing the mouse motion state. Principal component analysis is used for dimensionality reduction and denoising. By calculating the covariance matrix of the feature matrix, solving the eigenvalues ​​and eigenvectors, the main feature components can be determined. In the analysis of mouse motion, it may be found that the first three principal components account for more than 90% of the total variance. These principal components may correspond to key information such as the position, velocity and acceleration of the mouse. Image reconstruction is to map the reduced-dimensional data back to the original space. Reconstructing the image matrix using the main feature components can obtain a preliminary fused image that removes secondary information and background noise. This step helps to highlight the key features of mouse motion, such as trajectory and posture changes. Wavelet transform enhances image clarity in the frequency domain. By decomposing the image into sub-bands of different scales and directions, high-frequency details can be enhanced in a targeted manner. For mouse motion images, edge and texture information can be enhanced to improve the clarity of the mouse outline. The inverse wavelet transform converts the enhanced frequency domain information back to the time domain. This step combines the enhancement effects of each scale to generate a time domain image with higher clarity. In mouse motion analysis, this may be manifested as clearer hair texture and limb contours. The non-local mean filtering algorithm searches for similar regions in the image and uses the weighted average of these regions to reduce noise. This method can effectively retain image details while removing residual noise. For mouse motion images, it can remove interference information in the background while retaining motion details. The final result of the multi-frame image fusion processing is output as the fifth image. This process combines multiple steps such as feature extraction, dimensionality reduction and denoising, frequency domain enhancement, and spatial domain filtering to generate an image that is both clear and low-noise. For mouse motion analysis, the final fused image should be able to clearly show the mouse's motion trajectory, posture changes, and tiny behavioral details, providing high-quality data support for subsequent behavioral analysis. The significance of this series of processing steps is that through the synergy of multiple image processing techniques, useful information can be extracted and retained to the maximum extent, while noise and irrelevant information can be effectively suppressed. This not only improves the visual quality of the image, but more importantly, it enhances the effective information contained in the image, laying a solid foundation for subsequent image analysis and understanding tasks. In mouse behavior research, this high-quality image processing result can significantly improve the accuracy of behavior recognition and help researchers more accurately analyze the movement patterns and behavioral characteristics of mice.

[0078] Further, such as Figure 6 As shown, the gait analysis report generated includes:

[0079] Inputting the fifth image into a pre-established deep learning model, performing feature extraction and key joint position recognition on the fifth image, and obtaining coordinate position information of the key joints of the mouse;

[0080] Acquiring pre-collected mouse gait characteristic data, and establishing a gait analysis model according to the gait characteristic data;

[0081] The coordinate position information of the key joints of the mouse is input into the gait analysis model, and the gait of the mouse is analyzed in combination with the pre-established gait feature data to generate a gait analysis report of the mouse.

[0082] Specifically, deep learning models play a key role in mouse gait analysis. This model usually adopts a convolutional neural network structure, which can accurately identify the key joint positions in mouse images after being trained with a large amount of labeled data. For example, a typical deep learning model may identify key points such as the head, shoulder, elbow, wrist, hip, knee and ankle of the mouse. The coordinate information of these key points provides basic data for subsequent gait analysis. The collection of gait feature data is a prerequisite for establishing a gait analysis model. Researchers usually collect a large amount of gait data from healthy mice and mice with different diseases. These data include parameters such as step length, step frequency, gait cycle, and support phase duration. By performing statistical analysis on these data, a reference model for normal and abnormal gait can be established. The gait analysis model uses machine learning algorithms, such as support vector machines or random forests, to compare and analyze the joint coordinate information output by the deep learning model with pre-established gait feature data. For example, the model may calculate indicators such as the mouse's step length ratio (the ratio of the stride length of the front and rear limbs), gait symmetry (the consistency of left and right limb movements), etc. These indicators are compared with the reference model to determine whether the mouse's gait is normal. Preset threshold conditions play an important role in gait analysis. For example, the stride ratio of normal mice is usually between 0.8 and 1.2. If the analysis results show that the stride ratio exceeds this range, the system may determine it as an abnormal gait. Similarly, a gait symmetry index below 0.9 may be considered abnormal. If multiple indicators exceed the preset threshold, the system will generate a detailed gait analysis report. Gait analysis reports usually contain information from multiple aspects. The first is basic gait parameters, such as stride length, stride frequency, etc. The second is more in-depth analysis results, such as the movement trajectory of each joint, changes in joint angles, etc. The report may also include comparative analysis with normal reference values, as well as possible speculation on the cause of the abnormality. For example, if the stride length of the mouse's hind limbs is found to be significantly shortened, while the forelimbs are normal, the report may suggest that this may be related to spinal cord injury or neurodegenerative diseases. This method based on deep learning and gait analysis models has many technical advantages. First, it improves the accuracy and objectivity of the analysis and reduces human errors. Secondly, it can capture subtle gait changes that are difficult to detect with the naked eye, which helps in early diagnosis of diseases. In addition, this method is highly efficient and repeatable, and can quickly process large amounts of data, providing strong support for drug screening and disease research. By continuously accumulating data and optimizing models, the accuracy and scope of this method will be further improved, bringing more breakthroughs to mouse-related medical research.

[0083] The present invention discloses a mouse gait analysis method based on follow-up shooting, which analyzes the mouse movement trend through background separation and optical flow method, and adjusts the camera parameters in real time to maintain stable tracking. Adaptive illumination compensation and multi-frame fusion technology are used to improve image quality, and edge detection and deep learning models are combined to extract mouse gait characteristics. The present invention can continuously and stably capture mouse activities in complex environments, automatically adjust the shooting range and parameters, and effectively solve the problems of traditional methods such as tracking difficulties and unstable image quality in free activity environments. Through the coordinated application of intelligent algorithms and multiple image processing technologies, the present invention realizes accurate analysis and long-term monitoring of mouse behavior, providing reliable data support and technical guarantees for related research.

[0084] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A mouse gait analysis method based on follow-up shooting, characterized in that: include: Acquire an initial video image of the mouse activity area, separate the mouse movement area through a preset background model, and obtain a first image; According to the first image, calculating the motion vector of the mouse, determining the moving direction and speed of the mouse, and obtaining motion trend data; According to the motion trend data, adjusting the shooting angle and focal length of the camera, if the moving speed of the mouse exceeds a preset threshold, starting the fast tracking mode to obtain a second image; According to the second image, adjusting the image brightness distribution to eliminate the influence of uneven illumination, so as to obtain a third image; Extracting the outline features of the mouse based on the third image, and judging the gait state of the mouse in combination with a pre-established gait model to obtain gait feature data; According to the gait feature data, the tracking path of the camera is updated, and if the activity path of the mouse deviates from the preset range, the shooting area is replanned to obtain a fourth image; According to the fourth image, enhancing image clarity and eliminating background noise to obtain a fifth image; Based on the fifth image, the key joint positions of the mouse are identified through a deep learning model, and a gait analysis report is generated in combination with the gait feature data.

2. The mouse gait analysis method based on follow-up shooting as claimed in claim 1, characterized in that: Obtaining the first image includes: S1, obtaining an initial video image sequence of the mouse activity area, and performing background separation processing using a preset background model on each frame of the initial video image sequence to obtain a binary image sequence of the mouse movement area; S2, based on the motion region binary image sequence, removing the noise region in the image to obtain a first denoised image sequence; S3. Based on the first image sequence, extract the complete motion area of ​​the mouse to obtain a second image sequence; S4. Based on the second image sequence, using a convolutional neural network model to identify and classify the movement posture of the mouse to obtain a mouse movement posture category sequence; S5. Based on the mouse motion posture category sequence, the behavior pattern of the mouse is modeled by a hidden Markov model to obtain a state transition matrix of the mouse behavior pattern; S6, collecting the target mouse activity video image, repeating the processing from S1 to S4, and obtaining the target mouse movement posture category sequence; S7, inputting the target mouse motion posture category sequence into the hidden Markov model of S5 for decoding to obtain a first image.

3. The mouse gait analysis method based on follow-up shooting as claimed in claim 1, characterized in that: Obtaining sports trend data includes: Based on the first image, preprocessing the first image to obtain a preprocessed image; Based on the preprocessed image, calculating the motion vectors of pixels between adjacent frames to obtain pixel-level motion information; According to the pixel-level motion information, a mouse target in the image is identified, and the position and size of the mouse are determined; Track the mouse target in multiple consecutive frames of images, obtain the mouse's motion trajectory, and calculate the mouse's motion trajectory and speed information; According to the movement trajectory and speed information of the mouse, the movement trend of the mouse is predicted to obtain the movement trend data of the mouse in the future period of time, and the movement trend data includes position coordinates and moving speed.

4. The mouse gait analysis method based on follow-up shooting as claimed in claim 1, characterized in that: Obtaining the second image includes: Based on the position coordinates in the obtained motion trend data, the motion trajectory of the mouse is predicted to obtain the predicted position of the mouse at the next moment; According to the predicted position of the mouse, the camera's shooting angle is controlled by the servo motor so that the camera's shooting center coincides with the predicted position of the mouse, and the displacement of the mouse between consecutive frames is calculated to obtain the mouse's real-time moving speed; A video frame in a fast tracking mode is acquired through a camera, and image enhancement and denoising are performed on the video frame to obtain a second image.

5. The mouse gait analysis method based on follow-up shooting as claimed in claim 1, characterized in that: Obtaining the third image includes: Based on the brightness distribution information of the second image, determining an area in the image where uneven illumination exists; According to the uneven lighting area, dynamically adjust the brightness compensation coefficient in the area; Adaptively adjusting the pixel brightness value of the uneven illumination area according to the compensation coefficient to eliminate the influence of the uneven illumination; Smoothing the adjusted pixel brightness values ​​to eliminate sudden brightness changes and obtain an image area with uniform illumination; The processed uneven illumination area is fused with the uniform illumination area in the original image to obtain an image with uniform illumination as a whole, and the image with uniform illumination as a whole is subjected to contrast enhancement processing to obtain a third image.

6. The mouse gait analysis method based on follow-up shooting as claimed in claim 1, characterized in that: The gait feature data obtained include: Preprocessing the third image to obtain a preprocessed image; Analyze the preprocessed image, extract the contour features of the mouse in the image, and obtain a binary image of the mouse contour; According to the pre-established gait model, the extracted mouse contour features are analyzed, and the current mouse gait state is determined through model matching and feature comparison; If the mouse gait state conforms to the preset gait pattern, the current gait feature data is saved as input for subsequent analysis; otherwise, the next frame of image is processed; According to the stored gait feature data, similar gait features are grouped into one category to obtain feature sets under different gait states; The feature sets under different gait states are trained, a gait state classifier is established, the gait state of mice is tracked and analyzed in real time, and continuous gait feature data is obtained.

7. The mouse gait analysis method based on follow-up shooting according to claim 1, characterized in that: Obtaining the fourth image comprises: According to the gait feature data, the next movement trajectory of the mouse is predicted, and the tracking path of the camera is updated in real time; Compare the mouse's activity path with the preset range to determine whether the mouse deviates from the preset range. If the mouse deviates from the preset range, the camera is triggered to re-plan the shooting area. Perform semantic segmentation on the mouse activity area to determine the optimal shooting area; Pass the optimal shooting area parameters to the camera, adjust the camera parameters, and reacquire the shooting area; A fourth image is acquired according to the reacquired parameters of the photographing area.

8. The mouse gait analysis method based on follow-up shooting as claimed in claim 1, characterized in that: Obtaining the fifth image comprises: Extracting feature information of each frame of the image according to the fourth image and constructing a feature matrix; According to the characteristic matrix, determining the main characteristic components, removing the secondary components and background noise; Reconstructing an image matrix according to the main characteristic components to obtain a preliminary fused image; According to the preliminary fused image, performing image enhancement processing in the frequency domain by using wavelet transform to obtain an enhanced frequency domain image; Reconstructing the enhanced frequency domain image by using inverse wavelet transform to obtain a reconstructed time domain image; According to the reconstructed time domain image, residual noise is removed to obtain a fifth image.

9. The mouse gait analysis method based on follow-up shooting as claimed in claim 1, characterized in that: Generated gait analysis report includes: Inputting the fifth image into a pre-established deep learning model, performing feature extraction and key joint position recognition on the fifth image, and obtaining coordinate position information of the key joints of the mouse; Acquiring pre-collected mouse gait characteristic data, and establishing a gait analysis model according to the gait characteristic data; The coordinate position information of the key joints of the mouse is input into the gait analysis model, and the gait of the mouse is analyzed in combination with the pre-established gait feature data to generate a gait analysis report of the mouse.

Images