Evaluation method and system for police martial arts fighting movements based on staggered time pulses

Abstract

The present invention discloses a police martial arts combat action evaluation method and system based on staggered time pulses, belonging to the technical field of action behavior analysis. The method solves the problem that existing methods cannot timely capture subtle dynamic changes of joints when processing complex motion sequences. The method comprises extracting key skeleton points from a target action video, converting serialized motion data into a discrete pulse signal set through a pulse function, processing the pulse signal set based on the staggered time function, and performing police martial arts combat action recognition and training evaluation on the target action video based on the pulse feature set using a multidimensional time series evaluation model. The present invention introduces a staggered time function to analyze the temporal relationship of skeleton points, and uses a feature adaptive extraction algorithm based on a flexible trigger mechanism of a water tank to ensure the temporal sequence and integrity of feature extraction, thereby improving the accuracy and robustness of motion behavior evaluation, thereby providing a scientific and effective evaluation tool for action training and evaluation.

Description

Technical Field

[0001] The present invention belongs to the technical field of motion behavior analysis, and in particular relates to a method and system for evaluating police martial arts fighting movements based on staggered time pulses. Background Art

[0002] In martial arts training, evaluating the accuracy of combat movements is crucial. Currently, traditional evaluation methods rely primarily on coach observation and video playback. Coach observation is an intuitive evaluation method, in which the coach makes subjective judgments based on experience. Although this method is simple and direct, the lack of quantitative standards makes the accuracy and consistency of the evaluation results difficult to guarantee, and it is also subject to the coach's personal ability and attention. In contrast, video playback records training footage and analyzes the details of the movements frame by frame. This method can capture more detailed information, but it also suffers from reliance on manual analysis, low efficiency, and lack of real-time feedback, making it difficult to meet the needs of rapid evaluation.

[0003] With the advancement of technology, motion recognition methods based on sensors and optical capture have been gradually applied to training evaluation. These methods utilize modern technologies to quantitatively analyze motion data, significantly improving evaluation accuracy and scientific validity. Sensor-based motion capture methods, for example, use wearable devices such as accelerometers and gyroscopes to collect real-time human motion data and analyze motion parameters using biomechanical models. While this method can capture motion details with high accuracy, the expensive and wearable sensor equipment interferes with the naturalness of training. Furthermore, the high technical requirements for equipment installation and commissioning, as well as the operator's skill level, limit its practical application in high-intensity training scenarios. Optical motion capture methods, on the other hand, employ reflective markers and a multi-camera system to capture human motion trajectories, generate skeletal models, and perform detailed analysis. This method is widely used in scientific research and can accurately capture motion trajectories and postures. However, its high equipment cost, complex environmental requirements, and reliance on specific optical conditions make it difficult to scale in real-world training scenarios. Furthermore, optical capture systems are susceptible to external interference in dynamic scenes, such as light changes or motion outside the camera's range, further limiting their applicability.

[0004] In recent years, video analysis techniques based on skeleton data have gradually become mainstream in the field of action recognition. These methods use algorithms to extract key skeletal points from videos and combine them with rule-based inference or keyframe extraction techniques to analyze the action. While significantly more automated than traditional methods, they still face numerous technical bottlenecks. First, existing methods often rely on preset empirical thresholds or simple rule-based algorithms to identify keyframes within an action. While these methods may perform well in static or simple scenes, they lack robustness and adaptability in complex motion environments or for action sequences with significant individual variation, limiting the accuracy and stability of recognition for highly dynamic actions. Second, due to the high dimensionality of skeleton data, feature extraction faces a trade-off between data dimensionality and computational efficiency. To reduce complexity, researchers often simplify the process by reducing dimensionality or selecting key points. However, these methods can easily lose important motion information, compromising the integrity of feature extraction and overall recognition performance. Furthermore, police martial arts combat maneuvers often involve complex dynamic behaviors, with significant temporal staggering between skeleton points. For example, different limbs have distinct timing differences when performing continuous movements. Existing motion recognition methods typically rely on synchronous processing or simple temporal modeling, making it difficult to accurately characterize the dynamic temporal relationships between moving parts. This limitation directly leads to insufficient feature mining for complex fighting movements, making it impossible to meet the practical needs of high-precision motion assessment.

[0005] Chinese patent CN118718339A discloses a motion assistance system and method. This system uses a motion evaluation unit to evaluate video information related to the user's movements, calculate a motion score, and output a feature map representing the contribution of joints. This method primarily relies on video analysis for motion evaluation, but it can be limited when processing complex motion sequences and may be unable to capture subtle dynamic changes in joints.

[0006] Chinese patent CN118737376A discloses an information processing device, an information processing method, a program recording medium, and a learning model. These devices, information processing methods, program recording medium, and learning models are capable of outputting age-appropriate guidance reports for each subject and assisting the subject's movements. The information processing device comprises: a subject information acquisition unit that acquires subject information including age information; a motion evaluation information acquisition unit that acquires motion evaluation information related to the subject's motion evaluation; and a storage unit that stores the learned model. However, when processing complex motion sequences, these methods fail to capture subtle dynamic changes in joints in a timely manner, nor can they provide fine-grained motion analysis. Existing methods are also inefficient and lack robustness, adaptability, accuracy, and stability. Furthermore, the high-dimensional nature of human skeletal point data makes it difficult to balance data dimensionality and computational efficiency when extracting features. Complex movements also manifest as significant temporal staggered relationships between skeletal points in different locations. Existing motion recognition methods often overlook this characteristic. To address these issues, we propose a method and system for evaluating police martial arts combat movements based on staggered temporal pulses. Summary of the Invention

[0007] The purpose of the present invention is to address the shortcomings of the existing technology and provide a police martial arts combat action evaluation method and system based on staggered time pulses. This solves the problem that the existing methods are unable to capture the subtle dynamic changes of joints in a timely manner when processing complex motion sequences, and are unable to provide more fine-grained motion analysis. At the same time, the existing methods are inefficient and lack robustness, adaptability, accuracy and stability. In addition, the high-dimensional characteristics of human skeletal point data make it difficult for the existing technology to balance data dimensions and computational efficiency when extracting features. Complex action behaviors are also manifested as obvious staggered time relationships between skeletal points in different parts. Existing action recognition methods often ignore this feature. The present invention improves accuracy through data preprocessing, enhances reliability through filtering, generates pulse signals to emphasize timing characteristics, analyzes the timing relationship of skeletal points through staggered time functions, and uses a feature adaptive extraction algorithm of the flexible triggering mechanism of the water tank to ensure the timing and integrity of feature extraction.

[0008] The present invention is achieved by providing a police martial arts combat action evaluation method based on staggered time pulses, the method comprising:

[0009] Obtain the target action video, extract key skeleton points from the target action video, number the key skeleton points, and cluster the corresponding coded skeleton points based on the anatomical features of the region;

[0010] Obtain a serialized cluster center point sequence, convert the serialized motion data into a discrete pulse signal set through a pulse function, and perform serial processing on the pulse signal set;

[0011] Loading the pulse signal set, adjusting the weight of the pulse signal set based on the staggered time function, and placing the weighted pulse signal set into the flexible trigger mechanism for processing to obtain the pulse feature set;

[0012] Construct and train a multidimensional time series evaluation model, take the pulse feature set as input, execute the multidimensional time series evaluation model, and perform police martial arts fighting action recognition and training evaluation on the target action video based on the pulse feature set, and output the action recognition and training evaluation results of the target action video.

[0013] The method for extracting key skeleton points from a target action video includes:

[0014] Loading at least one set of target action videos, and converting the target action videos frame by frame into at least one set of action images;

[0015] The human pose estimation algorithm BlazePose is used to detect key skeleton points in each frame of action images. The key skeleton points of the human body are detected and numbered. Each detected skeleton point is assigned a fixed number i.

[0016] Among them, the human pose estimation algorithm BlazePose performs key bone point detection on each frame of action image, analyzes the human body contour and joint points in the image, and estimates the coordinates of each bone point in three-dimensional space. Assume that at time t, the coordinates of bone point number i are:

[0017] p i (t)=(x i (t),y i (t),z i (t))

[0018] Among them, (x i (t),y i (t),z i (t)) represents the coordinates of the skeleton point i in three-dimensional space;

[0019] Obtain the key skeleton points detected by the human body, and filter and reduce the noise of the key skeleton points detected by the human body based on a low-pass filter;

[0020] Filtered coordinates Calculated by the following formula:

[0021]

[0022] When the corresponding encoded bone points are clustered based on the anatomical features of the region, the clustering result is expressed as:

[0023]

[0024] Among them, each cluster C k Represents a group of related bone points, reflecting their common features during movement.

[0025] The method of converting serialized motion data into a discrete pulse signal set by using a pulse function comprises:

[0026] Load the cluster center point sequence c k (t), calculate the speed and pulse the cluster center sequence to extract the temporal characteristics of the action;

[0027] Let v k (t) is the cluster center c k (t) The velocity at time t, defined as:

[0028]

[0029] Wherein, Δt is the time interval between adjacent sampling moments;

[0030] Preset a speed threshold function θ k , used to generate a pulse signal;

[0031] The speed of the cluster center exceeds the preset speed threshold function θ k When it is considered to be in an active state at that moment, the corresponding pulse signal s is generated. k (t):

[0032]

[0033] Among them, s k (t) is the pulse signal of the kth cluster at time t;

[0034] Integrate at least one set of pulse signals s k (t) obtaining a discrete pulse signal set;

[0035] Load the pulse signal set and serialize the pulse signals of each cluster center point to obtain the pulse signals in the form of time series:

[0036] S k ={s k (1),s k (2),…,s k (T)}

[0037] Where T is the time length of the entire sequence;

[0038] Define the weighted fusion pulse signal S f (t):

[0039]

[0040] Among them, w k Represents the weight of the kth cluster, which is used to balance the contribution of different clusters to highlight the cluster that occupies a dominant position in the action characteristics. In this process, a unified pulse sequence S representing the overall action is obtained by weighted fusion of the pulse signals of different clusters. f (t).

[0041] The method of adjusting the weight of the pulse signal set based on the peak-shifting time function and placing the weight-adjusted pulse signal set into the flexible trigger mechanism for processing includes:

[0042] Load the pulse signal set, identify the movement sequence between multiple skeleton points in the pulse signal set based on the movement time difference between the skeleton points based on the staggered time function, capture the dynamic characteristics of the target action, and identify the occurrence of coherent actions by analyzing the time series characteristics between multiple skeleton point categories;

[0043] Among them, the peak time function τ ij (t) is defined as:

[0044] τ ij (t) = exp(-α|(t j -t i )-Δt ij |)

[0045] Among them, t i and t j Cluster C i and C j At the time stamp of moment t, α is the parameter for controlling the attenuation of the function. By exponentially attenuating the deviation of the time series of the cluster center point, when the time difference is greater than the preset delay Δt ij The closer the deviation is, the larger the function value is, indicating that the action timing is in line with expectations; on the contrary, if the deviation is large, the function value tends to zero, indicating that the action timing is abnormal;

[0046] The function τ ij (t) The output is used as a dynamic weight and acts on the pulse signal S f (t):

[0047]

[0048] in is the adjusted pulse signal, which is calculated by the peak time weight τ ij (t) Dynamically enhances the time difference feature.

[0049] Load the pulse signal set processed by the peak time function Combined with the flexible trigger mechanism based on the water tank, the pulse signal set features are adaptively extracted and the pulse feature set is output.

[0050] The pulse signal set processed by the staggered time function is loaded, and the features of the pulse signal set are adaptively extracted by combining the flexible trigger mechanism based on the water tank, and the pulse feature set is output.

[0051] The flexible trigger mechanism based on the water tank uses a nested water tank model to process the pulse signal set;

[0052] When the nested water tank model processes the pulse signal set, the small water tank state function Q is defined k (t), the tank state function Q k (t) represents the cumulative value of the small tank at time t, and the tank state function Q k The accumulation rules for (t) are as follows:

[0053]

[0054] in, Represents the adjusted cluster C after adjusting the peak time function k Pulse signal input, λ t is the linear time decay coefficient of the water tank, which is used to control the automatic emptying speed of the water tank;

[0055] To ensure that the data in the small tank represents the complete action, a cumulative threshold η is defined k , when Q k (t)≥η k When , the action is considered to be completed and the features are output:

[0056]

[0057] After the feature is exported, the small sink will be reset:

[0058] Q k (t) = 0

[0059] If there is not enough pulse signal input within a period of time, the accumulated value in the small water tank will gradually decrease over time.

[0060] The nesting mechanism of the nested sink model is expressed as:

[0061] Multiple small tanks are nested in a large tank, and the state function Q of the large tank is D (t) The accumulation rules are:

[0062]

[0063] Among them, I(Q k (t)≥η k) is the small sink output labeling function. When the small sink outputs a complete action, the value is 1, otherwise it is 0; γ t is the linear attenuation coefficient of the large water tank, and the cumulative threshold of the large water tank is η D , when Q D (t)≥η D When , the whole set of actions is considered completed, the whole set of action features are output, and the large sink is reset:

[0064] Q D (t) = 0

[0065] When the accumulated value of the sink reaches the threshold, the output feature can be expressed as the overall feature F(t) of the skeleton point serialization pulse:

[0066] F(t)={S output (t)|Q D (t)≥η D}

[0067] Among them, F(t) represents the feature output at time t, S output (t) is the serialized pulse signal after being processed by the peak time staggering function and the water tank mechanism.

[0068] The multidimensional time series evaluation model is a pulse neural network, which consists of an input layer, multiple hidden layers, a pooling layer and an output layer. The input layer of the network receives the overall feature F(t) of the pulse feature set;

[0069] The input layer accepts the overall feature F(t) as input, which is used to capture the motion pattern of the skeleton points in the time series;

[0070] Each hidden layer corresponds to a specific skeleton point category. The hidden layer processes information through a pulse neuron model. Using the pulse neuron model, the membrane potential u of each neuron is i (t) According to the input feature accumulation, the update rule is:

[0071] u i (t) = u i (t-1)+w i F(t)-τu i (t-1)

[0072] Among them, w i is the connection weight, and τ is the time constant that controls the decay of membrane potential.

[0073] The neuron pulse emission mechanism of the multidimensional time series evaluation model is:

[0074] When the membrane potential u i (t) exceeds the threshold θ iWhen , the neuron will output a pulse signal, and the output signal is defined as:

[0075]

[0076] Here, the output signal o(t) represents the pulse emission state of the neuron at time t, which is used to reflect the occurrence of motor behavior;

[0077] A pooling layer is set between the hidden layer and the output layer. The pooling layer is used to integrate the output features from multiple hidden neurons and reduce the dimensionality of the data. The pooling layer uses max pooling to select the maximum value from the output of multiple spiking neurons to ensure that the most significant features are focused on in action recognition.

[0078] The pooling operation of the pooling layer is defined as:

[0079] P(t)=max(o1(t),o2(t),...,o n (t))

[0080] Among them, P(t) is the output of the pooling layer, o n (t) is the pulse output of the nth neuron in the hidden layer;

[0081] The output layer is responsible for converting the pulse signal of the hidden layer into the final action recognition result and outputting the feature The calculation method is:

[0082]

[0083] Among them, w oi is the connection weight between the output layer neurons and the hidden layer neurons, and N is the number of hidden layer neurons.

[0084] The cross entropy loss function is used to optimize the performance of the spiking neural network during training of the multidimensional time series evaluation model. The loss function of the multidimensional time series evaluation model is defined as:

[0085]

[0086] Adjust the connection weights w in the network through the back-propagation algorithm oi , to minimize losses.

[0087] On the other hand, the present invention further provides an action evaluation system based on off-peak time pulses, the action evaluation system based on off-peak time pulses comprising:

[0088] A data processing module is used to obtain a target action video, extract key skeleton points from the target action video, number the key skeleton points, and cluster the corresponding coded skeleton points based on the anatomical features of the region;

[0089] The pulse processing module converts the serialized motion data into a discrete pulse signal set through a pulse function and performs serial processing on the pulse signal set;

[0090] A feature extraction module is used to load a pulse signal set, adjust the weight of the pulse signal set based on the staggered time function, and process the weighted pulse signal set in a flexible trigger mechanism to obtain a pulse feature set;

[0091] The action evaluation module is used to build and train a multidimensional time series evaluation model. It takes the pulse feature set as input and executes the multidimensional time series evaluation model. The multidimensional time series evaluation model performs police martial arts fighting action recognition and training evaluation on the target action video based on the pulse feature set, and outputs the action recognition and training evaluation results of the target action video.

[0092] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0093] In the present invention, the temporal relationship of skeleton points is analyzed by introducing a staggered time function to judge actions and coherent actions according to the time series differences of different skeleton points and their categories. The feature adaptive extraction algorithm based on the flexible trigger mechanism of the water tank ensures the temporal and integrity of feature extraction. By constructing and training a multidimensional time series evaluation model based on a spiking neural network, the cross entropy loss function is used for training in the design of the multidimensional time series evaluation model, thereby optimizing the performance of the spiking neural network and improving the accuracy and robustness of the motor behavior evaluation, thereby providing a scientific and effective evaluation tool for action training and evaluation, greatly improving the practicality and scientific nature of training evaluation, and the lightweight design of the spiking neural network enables the system to perform real-time data processing and action recognition at a high speed and accuracy under conditions of limited hardware resources. It also has a strong temporal evaluation capability and can monitor and analyze action dynamics in real time, especially in training environments with high dynamic changes. It overcomes the problem that existing methods are unable to capture subtle dynamic changes of joints in a timely manner when processing complex motion sequences, and are unable to provide more fine-grained motion analysis. It also solves the problems of existing methods such as inefficiency, lack of robustness, adaptability, accuracy and stability, and the high-dimensional characteristics of human bone point data, which makes it difficult for existing technologies to balance data dimensionality and computational efficiency when extracting features.

[0094] In the present invention, serialized motion data is converted into a discrete pulse signal set through a pulse function. This pulse method converts the continuous motion of the time series into pulse data with specific time nodes, retains important timing characteristics during the motion process, and helps to capture dynamic changes in complex motion.

[0095] In the present invention, a low-pass filter is used to eliminate high-frequency components and retain the low-frequency trend information in the coordinates of the skeleton points. The filtering can effectively remove the jitter caused by posture estimation errors, and is also conducive to reflecting the basic form and change trend of human body movements, which is of great significance for accurately describing human body movements. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 The present invention provides a schematic diagram of the implementation process of the police martial arts fighting action evaluation method based on staggered time pulses.

[0097] Figure 2 The figure shows a schematic diagram of the implementation process of the method for extracting key skeleton points from the target action video.

[0098] Figure 3 A schematic diagram of the implementation flow of a method for converting serialized motion data into a discrete pulse signal set through a pulse function is shown.

[0099] Figure 4 The present invention shows a schematic diagram of the implementation process of a method for adjusting the weight of a pulse signal set based on a peak-shifting time function and placing the pulse signal set after weight adjustment into a flexible triggering mechanism for processing.

[0100] Figure 5 A logical diagram of the water tank-based flexible triggering mechanism is shown.

[0101] Figure 6 Shown is the identification flow chart of the multidimensional time series evaluation model.

[0102] Figure 7 It is a structural diagram of the action evaluation system based on staggered time pulses provided by the present invention. DETAILED DESCRIPTION

[0103] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0104] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0105] When processing complex motion sequences, existing methods are unable to capture subtle dynamic changes of joints in a timely manner, nor can they provide more fine-grained motion analysis. To address the above problems, we propose a police martial arts fighting action evaluation method and system based on staggered time pulses. When implementing the method, the target action video is first obtained, key skeleton points are extracted from the target action video, and the key skeleton points are numbered. Then, the serialized motion data is converted into a discrete pulse signal set through a pulse function, the pulse signal set is serialized, and the weight of the pulse signal set is adjusted based on the staggered time function. The pulse signal set with adjusted weights is placed in a flexible trigger mechanism for processing. Finally, a multidimensional time series evaluation model performs police martial arts fighting action recognition and training evaluation on the target action video based on the pulse feature set, and outputs the action recognition and training evaluation results of the target action video. In an embodiment of the present invention, the temporal relationship of skeletal points is analyzed by introducing a staggered time function to determine actions and continuous actions based on the time series differences of different skeletal points and their categories. A feature adaptive extraction algorithm based on a flexible trigger mechanism of a water tank ensures the temporal sequence and integrity of feature extraction. A multidimensional time series evaluation model based on a spiking neural network is constructed and trained. The cross-entropy loss function is used for training in the design of the multidimensional time series evaluation model to optimize the performance of the spiking neural network and improve the accuracy and robustness of motion behavior evaluation. This provides a scientific and effective evaluation tool for motion training and evaluation, greatly improving the practicality and scientific nature of training evaluation. The lightweight design of the spiking neural network enables the system to perform real-time data processing and motion recognition at a high speed and accuracy even under conditions of limited hardware resources. It also has a strong temporal evaluation capability and can monitor and analyze motion dynamics in real time, especially in training environments with high dynamic changes. It overcomes the problem that existing methods are unable to capture subtle dynamic changes of joints in a timely manner when processing complex motion sequences, and are unable to provide more fine-grained motion analysis. It also solves the problems of existing methods such as inefficiency, lack of robustness, adaptability, accuracy and stability, and the high-dimensional characteristics of human bone point data, which makes it difficult for existing technologies to balance data dimensionality and computational efficiency when extracting features.

[0106] It should be noted that the police martial arts combat action evaluation method based on staggered pulses in the embodiments of the present invention can be used in a variety of applications, including but not limited to competitive sports, medical rehabilitation, industrial production, intelligent monitoring, fitness training, and police martial arts training. For example, during police martial arts training, action evaluation is used to analyze and improve trainees' combat skills and tactical movements to enhance combat efficiency and survivability. The present invention also supports detailed analysis and evaluation of complex combat action processes. This method improves the accuracy and efficiency of police martial arts combat action analysis, providing a solid technical support and application foundation for evaluating police officers' combat training.

[0107] The embodiment of the present invention provides a method for evaluating police martial arts fighting movements based on staggered time pulses. Figure 1 The following is a schematic diagram of the implementation process of a method for evaluating police martial arts fighting movements based on staggered time pulses. The method specifically includes:

[0108] S10, obtaining a target action video, extracting key skeleton points from the target action video, numbering the key skeleton points, and clustering the corresponding coded skeleton points based on the anatomical features of the region;

[0109] In this embodiment, a video of the target action is captured using a camera, video recorder, or machine vision-based surveillance camera. Key skeletal points are extracted from the video and numbered. These skeletal points correspond to the body's major joints and motion nodes, effectively capturing dynamic information during movement. The extracted skeletal point data is then used as the basis for analysis.

[0110] S20, obtaining a serialized cluster center point sequence, converting the serialized motion data into a discrete pulse signal set through a pulse function, and performing serialization processing on the pulse signal set;

[0111] In an embodiment of the present invention, serialized motion data is converted into a discrete set of pulse signals through a pulse function. This pulse method converts the continuous motion of a time series into pulse data with specific time nodes, retains important timing characteristics during the motion process, and helps to capture dynamic changes in complex motion.

[0112] S30, loading a pulse signal set, adjusting the weight of the pulse signal set based on the peak-shifting time function, and placing the weight-adjusted pulse signal set into a flexible triggering mechanism for processing to obtain a pulse feature set;

[0113] S40, construct and train a multidimensional time series evaluation model, use the pulse feature set as input, execute the multidimensional time series evaluation model, the multidimensional time series evaluation model performs police martial arts fighting action recognition and training evaluation on the target action video based on the pulse feature set, and outputs the target action video action recognition and training evaluation results.

[0114] In an embodiment of the present invention, the temporal relationship of skeletal points is analyzed by introducing a staggered time function to determine actions and continuous actions based on the time series differences of different skeletal points and their categories. A feature adaptive extraction algorithm based on a flexible trigger mechanism of a water tank ensures the temporal sequence and integrity of feature extraction. A multidimensional time series evaluation model based on a spiking neural network is constructed and trained. The cross-entropy loss function is used for training in the design of the multidimensional time series evaluation model to optimize the performance of the spiking neural network and improve the accuracy and robustness of motion behavior evaluation. This provides a scientific and effective evaluation tool for motion training and evaluation, greatly improving the practicality and scientific nature of training evaluation. The lightweight design of the spiking neural network enables the system to perform real-time data processing and motion recognition at a high speed and accuracy even under conditions of limited hardware resources. It also has a strong temporal evaluation capability and can monitor and analyze motion dynamics in real time, especially in training environments with high dynamic changes. It overcomes the problem that existing methods are unable to capture subtle dynamic changes of joints in a timely manner when processing complex motion sequences, and are unable to provide more fine-grained motion analysis. It also solves the problems of existing methods such as inefficiency, lack of robustness, adaptability, accuracy and stability, and the high-dimensional characteristics of human bone point data, which makes it difficult for existing technologies to balance data dimensionality and computational efficiency when extracting features.

[0115] The embodiment of the present invention provides a method for extracting key skeleton points from a target action video. Figure 2 The following is a schematic diagram of a method for extracting key skeleton points from a target action video. The method for extracting key skeleton points from a target action video specifically includes:

[0116] S101, loading at least one set of target action videos, and converting the target action videos frame by frame into at least one set of action images;

[0117] It should be noted that the input target action video data is a collected continuous video stream. In order to analyze the posture and movement of the human body in combat sports, it is necessary to convert the target action video frame by frame into at least one set of action images, and then extract the skeleton points in the video frames.

[0118] S102, using the human pose estimation algorithm BlazePose to perform key skeleton point detection on each frame of the action image, detecting the key skeleton points of the human body and numbering the key skeleton points. Each detected skeleton point is assigned a fixed number i;

[0119] Among them, the human pose estimation algorithm BlazePose performs key bone point detection on each frame of action image, analyzes the human body contour and joint points in the image, and estimates the coordinates of each bone point in three-dimensional space. Assume that at time t, the coordinates of bone point number i are:

[0120] p i (t)=(x i (t),y i (t),z i (t))

[0121] Among them, (x i (t),y i (t),z i (t)) represents the coordinates of the skeleton point i in three-dimensional space;

[0122] S103, obtaining key skeleton points detected by the human body, and filtering and denoising the key skeleton points detected by the human body using a low-pass filter;

[0123] Filtered coordinates Calculated by the following formula:

[0124]

[0125] When the corresponding encoded bone points are clustered based on the anatomical features of the region, the clustering result is expressed as:

[0126]

[0127] Among them, each cluster C k Represents a group of related bone points, reflecting their common features during movement.

[0128] In this embodiment, in order to reduce noise and improve data reliability, the extracted skeleton point coordinates need to be filtered. In this application, a low-pass filter is used to eliminate high-frequency components and retain the low-frequency trend information in the skeleton point coordinates. Filtering can effectively remove the jitter caused by posture estimation errors, and is also conducive to reflecting the basic form and change trend of human body movements, which is of great significance for accurately describing human body movements.

[0129] The embodiment of the present invention provides a method for converting serialized motion data into a discrete pulse signal set through a pulse function. Figure 3 The following is a schematic diagram of a method for converting serialized motion data into a discrete pulse signal set using a pulse function. The method for converting serialized motion data into a discrete pulse signal set using a pulse function specifically includes:

[0130] S201, load cluster center point sequence c k (t), calculate the speed and pulse the cluster center sequence to extract the temporal characteristics of the action;

[0131] Let v k (t) is the cluster center c k (t) The velocity at time t, defined as:

[0132]

[0133] Wherein, Δt is the time interval between adjacent sampling moments;

[0134] S202, preset a speed threshold function θ k , the speed threshold function is used to generate the pulse signal;

[0135] S203, the speed of the cluster center exceeds the preset speed threshold function θ k When it is considered to be in an active state at that moment, the corresponding pulse signal s is generated. k (t):

[0136]

[0137] Among them, s k (t) is the pulse signal of the kth cluster at time t;

[0138] S204, integrating at least one group of pulse signals s k (t) obtaining a discrete pulse signal set;

[0139] S205, load the pulse signal set, and serialize the pulse signals of each cluster center point to obtain the pulse signals in the form of time series:

[0140] S k ={s k (1),s k (2),…,s k (T)}

[0141] Where T is the time length of the entire sequence;

[0142] Define the weighted fusion pulse signal S f (t):

[0143]

[0144] Among them, w kRepresents the weight of the kth cluster, which is used to balance the contribution of different clusters to highlight the cluster that occupies a dominant position in the action characteristics. In this process, a unified pulse sequence S representing the overall action is obtained by weighted fusion of the pulse signals of different clusters. f (t), thus providing a reliable basis for subsequent feature extraction and action recognition.

[0145] The embodiment of the present invention provides a method for adjusting the weight of a pulse signal set based on a peak-shifting time function, and placing the pulse signal set after the weight adjustment into a flexible trigger mechanism for processing. Figure 4 A schematic diagram of a process flow for implementing a method for adjusting the weights of a pulse signal set based on a staggered time function and placing the weighted pulse signal set into a flexible triggering mechanism for processing is shown. The method for adjusting the weights of a pulse signal set based on a staggered time function and placing the weighted pulse signal set into a flexible triggering mechanism for processing specifically includes:

[0146] S301, loading a pulse signal set, and identifying a movement sequence between multiple skeleton points in the pulse signal set based on a staggered time function and movement time differences between the skeleton points;

[0147] S302, capturing the dynamic features of the target action and identifying the occurrence of coherent actions by analyzing the time series features between multiple skeleton point categories;

[0148] Among them, the peak time function τ ij (t) is defined as:

[0149] τ ij (t) = exp(-α|(t j -t i )-Δt ij |)

[0150] Among them, t i and t j Cluster C i and C j At the time stamp of moment t, α is the parameter for controlling the attenuation of the function. By exponentially attenuating the deviation of the time series of the cluster center point, when the time difference is greater than the preset delay Δt ij The closer the deviation is, the larger the function value is, indicating that the action timing is in line with expectations; on the contrary, if the deviation is large, the function value tends to zero, indicating that the action timing is abnormal;

[0151] The function τ ij (t) The output is used as a dynamic weight and acts on the pulse signal S f (t):

[0152]

[0153] in is the adjusted pulse signal, which is calculated by the peak time weight τ ij (t) Dynamically enhances the time difference feature.

[0154] It's important to note that the staggered timing function aims to identify the occurrence of specific actions or action sequences by analyzing the temporal differences between different skeletal points and their categories during the action. This function identifies the movement sequence between multiple skeletal points based on the temporal differences between them, accurately capturing the dynamic characteristics of the action and making appropriate judgments. Furthermore, by analyzing the temporal sequence characteristics between multiple skeletal point categories, it can effectively identify the occurrence of coherent actions.

[0155] S303, loading the pulse signal set processed by the peak-shifting time function, combining the flexible triggering mechanism based on the water tank to adaptively extract the features of the pulse signal set, and outputting the pulse feature set.

[0156] In this embodiment, if Figure 5 , which shows a logic diagram of a flexible triggering mechanism based on a water tank, wherein the flexible triggering mechanism based on a water tank adopts a nested water tank model to process a pulse signal set;

[0157] When the nested water tank model processes the pulse signal set, the small water tank state function Q is defined k (t), the tank state function Q k (t) represents the cumulative value of the small tank at time t, and the tank state function Q k The accumulation rules for (t) are as follows:

[0158]

[0159] in, Represents the adjusted cluster C after adjusting the peak time function k Pulse signal input, λ t is the linear time decay coefficient of the water tank, which is used to control the automatic emptying speed of the water tank;

[0160] To ensure that the data in the small tank represents the complete action, a cumulative threshold η is defined k , when Q k (t)≥η k When , the action is considered to be completed and the features are output:

[0161]

[0162] After the feature is exported, the small sink will be reset:

[0163] Q k (t) = 0

[0164] If there is not enough pulse signal input within a period of time, the accumulated value in the small tank will gradually decrease over time to avoid erroneous extraction of incomplete motion features.

[0165] The nesting mechanism of the nested sink model is expressed as:

[0166] Multiple small tanks are nested in a large tank, and the state function Q of the large tank is D (t) The accumulation rules are:

[0167]

[0168] Among them, I(Q k (t)≥η k ) is the small sink output labeling function. When the small sink outputs a complete action, the value is 1, otherwise it is 0; γ t is the linear attenuation coefficient of the large water tank, and the cumulative threshold of the large water tank is η D , when Q D (t)≥η D When , the whole set of actions is considered completed, the whole set of action features are output, and the large sink is reset:

[0169] Q D (t) = 0

[0170] When the accumulated value of the sink reaches the threshold, the output feature can be expressed as the overall feature F(t) of the skeleton point serialization pulse:

[0171] F(t)={S output (t)|Q D (t)≥η D}

[0172] Among them, F(t) represents the feature output at time t, S output (t) is the serialized pulse signal processed by the staggered time function and the water tank mechanism. This mechanism ensures the timing and integrity of feature extraction through the adjustment of the staggered time function and the design of the nested water tank model, effectively improving the accuracy and robustness of action recognition.

[0173] It should be noted that the embodiments of the present invention utilize a staggered time function and a sink-based flexible triggering mechanism to achieve adaptive feature extraction and recognition. The staggered time function analyzes the temporal features of the skeleton point sequence to separate the features of different pulse signal categories, thereby determining the occurrence of an action based on the temporal relationships between the skeleton points. Furthermore, the temporal sequence features between multiple skeleton point categories are analyzed to identify the occurrence of a coherent action. The sink-based flexible triggering mechanism manages the accumulation and judgment of pulse signals to achieve precise action recognition. For example, during a punch, there are time differences between the peak motions of various skeleton points. The staggered time function can coordinate these time differences, enabling more accurate feature extraction. The sink-based flexible triggering mechanism utilizes a sink model to manage the accumulation and judgment of pulse signals. The small sink represents the judgment of a single action. When the accumulation of a set of sequenced pulse signals reaches the small sink threshold, the action is considered complete, and its accumulated value is added to the large sink. The large sink represents the entire action set. When the small sink data accumulates to a certain threshold, the output is a complete set of action features. The sink also has a time decay mechanism. If there is not enough pulse signal input within a certain period of time, the data in the sink will gradually be cleared to avoid outputting incomplete actions.

[0174] For example, in a police martial arts combat training scenario, at the beginning of the training, the trainee trains according to the sequence of police martial arts combat movements. Through the process of extracting skeleton points, the system algorithm identifies each key skeleton point and numbers it, clusters it, and records the coordinates after filtering. Next, the raw data is pulsed and serialized based on the serialized pulse function to convert the motion data into a series of pulse signals. These pulse signals are then input into the feature extraction sink of the system. The sink mechanism adopts a nested design of small sinks and large sinks to gradually accumulate the trainee's movement features:

[0175] (1) Small tank: First, judge the completion of a single fighting action (such as basic boxing, kicking, throwing, grappling and other technical actions).

[0176] (2) Large tank: The actions of multiple small tanks are accumulated into the large tank and calibrated into a complete combination technique (for example, a series of defense and control combination techniques).

[0177] In this embodiment, if Figure 6 As shown, a recognition flow chart of a multidimensional time series evaluation model is shown. The multidimensional time series evaluation model is a pulse neural network, which consists of an input layer, multiple hidden layers, a pooling layer and an output layer. The input layer of the network receives the overall feature F(t) of the pulse feature set;

[0178] The input layer accepts the overall feature F(t) as input, which is used to capture the motion pattern of the skeleton points in the time series;

[0179] Each hidden layer corresponds to a specific skeleton point category. The hidden layer processes information through a pulse neuron model. Using the pulse neuron model, the membrane potential u of each neuron is i (t) According to the input feature accumulation, the update rule is:

[0180] u i (t) = u i (t-1)+w i F(t)-τu i (t-1)

[0181] Among them, w i is the connection weight, and τ is the time constant that controls the decay of membrane potential.

[0182] The neuron pulse emission mechanism of the multidimensional time series evaluation model is:

[0183] When the membrane potential u i (t) exceeds the threshold θ i When , the neuron will output a pulse signal, and the output signal is defined as:

[0184]

[0185] Here, the output signal o(t) represents the pulse emission state of the neuron at time t, which is used to reflect the occurrence of motor behavior;

[0186] A pooling layer is set between the hidden layer and the output layer. The pooling layer is used to integrate the output features from multiple hidden neurons and reduce the dimensionality of the data. The pooling layer uses max pooling to select the maximum value from the output of multiple spiking neurons to ensure that the most significant features are focused on in action recognition.

[0187] The pooling operation of the pooling layer is defined as:

[0188] P(t)=max(o1(t),o2(t),...,o n (t))

[0189] Among them, P(t) is the output of the pooling layer, o n (t) is the pulse output of the nth neuron in the hidden layer;

[0190] The output layer is responsible for converting the pulse signal of the hidden layer into the final action recognition result and outputting the feature The calculation method is:

[0191]

[0192] Among them, w oiis the connection weight between the output layer neurons and the hidden layer neurons, and N is the number of hidden layer neurons.

[0193] The cross entropy loss function is used to optimize the performance of the spiking neural network during training of the multidimensional time series evaluation model. The loss function of the multidimensional time series evaluation model is defined as:

[0194]

[0195] Adjust the connection weights w in the network through the back-propagation algorithm oi , to minimize losses.

[0196] On the other hand, the embodiment of the present invention also provides an action evaluation system based on staggered time pulses. Figure 7 The schematic diagram of the structure of the action evaluation system based on the peak time pulse is shown. The action evaluation system based on the peak time pulse specifically includes:

[0197] The data processing module 100 is used to obtain a target action video, extract key skeleton points from the target action video, number the key skeleton points, and cluster the corresponding coded skeleton points based on the anatomical features of the region;

[0198] The pulse processing module 200 converts the serialized motion data into a discrete pulse signal set through a pulse function, and performs serialization processing on the pulse signal set;

[0199] The feature extraction module 300 is used to load the pulse signal set, adjust the weight of the pulse signal set based on the peak time function, and put the weighted pulse signal set into the flexible trigger mechanism for processing to obtain the pulse feature set;

[0200] The action evaluation module 400 is used to construct and train a multidimensional time series evaluation model. It takes the pulse feature set as input and executes the multidimensional time series evaluation model. The multidimensional time series evaluation model performs police martial arts fighting action recognition and training evaluation on the target action video based on the pulse feature set, and outputs the action recognition and training evaluation results of the target action video.

[0201] In this embodiment, the data processing module 100, the pulse processing module 200, the feature extraction module 300, and the action evaluation module 400 can be connected by 5G, DTU or local area network communication.

[0202] To ensure the reliability of output features, the action evaluation system, based on staggered pulses, incorporates timeout detection and automatic clearing mechanisms to prevent inaccurate labeling of incomplete or interfering action data. The system evaluates the validity of accumulated pulse signals within a preset time window and automatically clears the relevant data if the action threshold is not met, thereby filtering out noise and interference and improving the stability and accuracy of action labeling.

[0203] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium storing computer program instructions, which can be executed by a processor. When the computer program instructions are executed, the method of any of the above embodiments is implemented.

[0204] Memory, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs, and modules, such as the program instructions / modules corresponding to the police martial arts combat action evaluation method based on staggered time pulses in the embodiments of this application. The memory can include a program storage area and a data storage area. The program storage area can store an operating system and at least one application required for a function; the data storage area can store data generated by the use of the police martial arts combat action evaluation method based on staggered time pulses. In some embodiments, the memory can optionally include a memory remote from the processor, which can be connected to the local module via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0205] As another aspect of the present invention, a computer device is provided. The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the method of any one of the above embodiments is implemented.

[0206] The various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure herein may be implemented or executed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination of these components. A general-purpose processor may be a microprocessor, but alternatively, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP, and / or any other such configuration.

[0207] In summary, the present invention provides a method and system for evaluating police martial arts combat actions based on staggered time pulses. In an embodiment of the present invention, staggered time functions are introduced to analyze the temporal relationship of skeletal points, so that actions and continuous actions can be judged based on the time series differences of different skeletal points and their categories. A feature adaptive extraction algorithm based on a flexible trigger mechanism of a water tank ensures the temporal and complete nature of feature extraction. By constructing and training a multidimensional time series evaluation model based on a spiking neural network, the cross-entropy loss function is used for training in the multidimensional time series evaluation model design, thereby optimizing the performance of the spiking neural network and improving the accuracy and robustness of motion behavior evaluation. This provides a scientific and effective evaluation tool for action training and evaluation, significantly enhancing the practicality and scientific nature of training evaluation. The lightweight design of the spiking neural network enables the system to perform real-time data processing and action recognition at a high speed and accuracy even under conditions of limited hardware resources. It also has strong temporal evaluation capabilities and can monitor and analyze action dynamics in real time, especially in training environments with high dynamic changes. It overcomes the problem that existing methods are unable to capture subtle dynamic changes of joints in a timely manner when processing complex motion sequences, and are unable to provide more fine-grained motion analysis. It also solves the problems of existing methods such as inefficiency, lack of robustness, adaptability, accuracy and stability, and the high-dimensional characteristics of human bone point data, which makes it difficult for existing technologies to balance data dimensionality and computational efficiency when extracting features.

[0208] The present invention captures the time delay relationship of skeleton point movement through the staggered time function, and combines it with serialized pulse processing to accurately identify the motion characteristics of different categories. It divides the related skeleton points into different clusters through clustering, thereby accurately reflecting their common characteristics during the movement process, thereby ensuring the integrity of feature extraction and laying the foundation for high-precision action labeling and analysis.

[0209] This invention supports hierarchical management of complex combat moves through cluster analysis of skeleton points. Each cluster center represents a motion feature. When the accumulated pulse signals from multiple cluster centers reach a preset threshold, the system identifies it as a complete set of moves. This hierarchical accumulation and output mechanism ensures the integrity and structure of the action sequence, providing strong support for subsequent complex behavior evaluation.

[0210] The present invention converts the original skeleton point data into a pulse signal sequence with significant discrimination through a flexible trigger mechanism based on a water tank, effectively improving the efficiency of feature extraction. The lightweight design of the multidimensional time series evaluation model enables the system to perform real-time data processing and action recognition at a high speed and accuracy even under conditions of limited hardware resources.

[0211] It should be noted that for the aforementioned embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0212] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also fall within the scope of protection of the present invention.

Claims

1. A police martial arts fighting action evaluation method based on staggered time pulses, characterized in that: The police martial arts fighting action evaluation method based on staggered time pulses includes: Obtain the target action video, extract key skeleton points from the target action video, number the key skeleton points, and cluster the corresponding coded skeleton points based on the anatomical features of the region; Obtain a serialized cluster center point sequence, convert the serialized motion data into a discrete pulse signal set through a pulse function, and perform serial processing on the pulse signal set; Loading the pulse signal set, adjusting the weight of the pulse signal set based on the staggered time function, and placing the weighted pulse signal set into the flexible trigger mechanism for processing to obtain the pulse feature set; Construct and train a multidimensional time series evaluation model, using the pulse feature set as input and executing the multidimensional time series evaluation model. The multidimensional time series evaluation model performs police martial arts combat action recognition and training evaluation on the target action video based on the pulse feature set, and outputs the action recognition and training evaluation results of the target action video; The method of adjusting the weight of the pulse signal set based on the peak-shifting time function and placing the weight-adjusted pulse signal set into the flexible trigger mechanism for processing includes: Load the pulse signal set, identify the movement sequence between multiple skeleton points in the pulse signal set based on the movement time difference between the skeleton points based on the staggered time function, capture the dynamic characteristics of the target action, and identify the occurrence of coherent actions by analyzing the time series characteristics between multiple skeleton point categories; Among them, the peak time function τ mj (t) is defined as: τ mj (t)=exp(-β|(t j -t m )-Δt mj |) Among them, t m and t j Cluster C m and C j At the time stamp of moment t, β is the parameter for controlling the attenuation of the function. By exponentially attenuating the deviation of the time series of the cluster center point, when the time difference is less than the preset delay Δt mj The closer the value, the larger the function value, indicating that the action sequence is in line with expectations; on the contrary, the function value tends to zero, indicating that the action sequence has abnormal timing; The function τ mj (t) The output is used as a dynamic weight and acts on the pulse signal S f (t) in is the adjusted pulse signal, which is calculated by the peak time weight τ mj (t) Dynamically enhance the time difference feature; Load the pulse signal set processed by the peak time function Combined with the flexible trigger mechanism based on the water tank, the pulse signal set features are adaptively extracted and the pulse feature set is output; The flexible trigger mechanism based on the water tank uses a nested water tank model to process the pulse signal set; When the nested water tank model processes the pulse signal set, the small water tank state function Q is defined k (t), the tank state function Q k (t) represents the cumulative value of the small tank at time t, and the tank state function Q k The accumulation rules for (t) are as follows: in, Represents the adjusted cluster C after adjusting the peak time function k Pulse signal input, λ t is the linear time decay coefficient of the water tank, which is used to control the automatic emptying speed of the water tank; To ensure that the data in the small tank represents the complete action, a cumulative threshold η is defined k , when Q k (t)≥η k When , the action is considered to be completed and the features are output: After the feature is exported, the small sink will be reset: Q k (t)=0 If there is not enough pulse signal input within a period of time, the accumulated value in the small water tank will gradually decrease over time.

2. The method for evaluating police martial arts combat movements based on staggered time pulses according to claim 1, wherein: The method for extracting key skeleton points from a target action video includes: Loading at least one set of target action videos, and converting the target action videos frame by frame into at least one set of action images; The human pose estimation algorithm BlazePose is used to detect key skeleton points in each frame of action images. The key skeleton points of the human body are detected and numbered. Each detected skeleton point is assigned a fixed number i. Among them, the human pose estimation algorithm BlazePose performs key bone point detection on each frame of action image, analyzes the human body contour and joint points in the image, and estimates the coordinates of each bone point in three-dimensional space. Assume that at time t, the coordinates of bone point number i are: p i (t)=(x i (t),y i (t),z i (t)) Among them, (x i (t),y i (t),z i (t)) represents the coordinates of the skeleton point i in three-dimensional space; Obtain the key skeleton points detected by the human body, and filter and reduce the noise of the key skeleton points detected by the human body based on a low-pass filter; Filtered coordinates Calculated by the following formula: When the corresponding encoded bone points are clustered based on the anatomical features of the region, the clustering result is expressed as: Among them, each cluster C k Represents a group of related skeletal points, reflecting their common features during movement.

3. The police martial arts combat action evaluation method based on staggered time pulses according to claim 2, characterized in that: The method of converting serialized motion data into a discrete pulse signal set by using a pulse function comprises: Load the cluster center point sequence c k (t), calculate the speed and pulse the cluster center sequence to extract the temporal characteristics of the action; Let v k (t) is the cluster center c k (t) The velocity at time t, defined as: Wherein, Δt is the time interval between adjacent sampling moments; Preset a speed threshold function θ k , used to generate a pulse signal; The speed of the cluster center exceeds the preset speed threshold function θ k When it is considered to be in an active state at that moment, the corresponding pulse signal s is generated. k (t): Among them, s k (t) is the pulse signal of the kth cluster at time t; Integrate at least one set of pulse signals s k (t) obtaining a discrete pulse signal set; Load the pulse signal set and serialize the pulse signals of each cluster center point to obtain the pulse signals in the form of time series: S k ={s k (1),s k (2),…,s k (T)} Where T is the time length of the entire sequence; Define the weighted fusion pulse signal S f (t): Among them, w k Represents the weight of the kth cluster, which is used to balance the contribution of different clusters to highlight the cluster that occupies a dominant position in the action characteristics. In this process, a unified pulse sequence S representing the overall action is obtained by weighted fusion of the pulse signals of different clusters. f (t).

4. The method for evaluating police martial arts combat movements based on staggered time pulses according to claim 3, wherein: The nesting mechanism of the nested sink model is expressed as: Multiple small tanks are nested in a large tank, and the state function Q of the large tank is D (t) The accumulation rules are: Among them, I(Q k (t)≥η k ) is the small sink output labeling function. When the small sink outputs a complete action, the value is 1, otherwise it is 0; γ t is the linear attenuation coefficient of the large water tank, and the cumulative threshold of the large water tank is η D , when Q D (t)≥η D When , the whole set of actions is considered completed, the whole set of action features are output, and the large sink is reset: When the accumulated value of the sink reaches the threshold, the output feature can be expressed as the overall feature F(t) of the skeleton point serialization pulse: F(t)={S output (t)∣Q D (t)≥η D } Among them, F(t) represents the feature output at time t, S output (t) is the serialized pulse signal after being processed by the peak time staggering function and the water tank mechanism.

5. The police martial arts combat action evaluation method based on staggered time pulses according to claim 2, characterized in that: The multidimensional time series evaluation model is a pulse neural network, which consists of an input layer, multiple hidden layers, a pooling layer and an output layer. The input layer of the network receives the overall feature F(t) of the pulse feature set; The input layer accepts the overall feature F(t) as input, which is used to capture the motion pattern of the skeleton points in the time series; Each hidden layer corresponds to a specific skeleton point category. The hidden layer processes information through a pulse neuron model. Using the pulse neuron model, the membrane potential u of each neuron is e (t) According to the input feature accumulation, the update rule is: u e (t)=u e (t-1)+w e F(t)-τu e (t-1) Among them, w e is the connection weight, and τ is the time constant that controls the decay of membrane potential.

6. The method for evaluating police martial arts combat movements based on staggered time pulses according to claim 5, characterized in that: The neuron pulse emission mechanism of the multidimensional time series evaluation model is: When the membrane potential u e (t) exceeds the threshold θ e When , the neuron will output a pulse signal, and the output signal is defined as: Here, the output signal o(t) represents the pulse emission state of the neuron at time t, which is used to reflect the occurrence of motor behavior; A pooling layer is set between the hidden layer and the output layer. The pooling layer is used to integrate the output features from multiple hidden neurons and reduce the dimensionality of the data. The pooling layer uses max pooling to select the maximum value from the output of multiple spiking neurons to ensure that the most significant features are focused on in action recognition. The pooling operation of the pooling layer is defined as: P(t)=max(o1(t),o2(t),...,o n (t)) Among them, P(t) is the output of the pooling layer, o n (t) is the pulse output of the nth neuron in the hidden layer; The output layer is responsible for converting the pulse signal of the hidden layer into the final action recognition result and outputting the feature The calculation method is: Among them, w oe is the connection weight between the output layer neurons and the hidden layer neurons, and N is the number of hidden layer neurons.

7. The method for evaluating police martial arts combat movements based on staggered time pulses according to claim 6, wherein: The cross entropy loss function is used to optimize the performance of the spiking neural network during training of the multidimensional time series evaluation model. The loss function of the multidimensional time series evaluation model is defined as: Adjust the connection weights w in the network through the back-propagation algorithm oi , to minimize losses.

8. A motion evaluation system based on staggered time pulses, for implementing the police martial arts combat motion evaluation method based on staggered time pulses as described in any one of claims 1 to 7, characterized in that: The action evaluation system based on staggered time pulses includes: A data processing module is used to obtain a target action video, extract key skeleton points from the target action video, number the key skeleton points, and cluster the corresponding coded skeleton points based on the anatomical features of the region; The pulse processing module converts the serialized motion data into a discrete pulse signal set through a pulse function and performs serial processing on the pulse signal set; A feature extraction module is used to load a pulse signal set, adjust the weight of the pulse signal set based on the staggered time function, and process the weighted pulse signal set in a flexible trigger mechanism to obtain a pulse feature set; The action evaluation module is used to build and train a multidimensional time series evaluation model. It takes the pulse feature set as input and executes the multidimensional time series evaluation model. The multidimensional time series evaluation model performs police martial arts fighting action recognition and training evaluation on the target action video based on the pulse feature set, and outputs the action recognition and training evaluation results of the target action video.

Images