Video moving object classification and identification method and system based on contour constraint

Abstract

The invention discloses a contour constraint-based video moving target classification and identification method and system, and relates to the technical field of target classification and identification. According to the video moving target classification and identification method based on the contour constraint, video data are acquired and input into a pre-trained machine learning model for contour extraction, a plurality of contour areas in each frame of image are obtained, each contour area is comprehensively analyzed, a moving contour area is identified, and analysis is performed based on a motion classification index of the area, so that the video moving target classification and identification efficiency is improved. Finally, the motion classification index is compared with a preset interval, each motion contour area is judged and classified, and category information of the motion contour area is obtained.By introducing combined application of contour constraint and a deep learning model, precise classification of the motion target in each frame of image is achieved, and compared with a traditional pixel-level classification method and a simple feature extraction method, the method has the advantages that the classification efficiency is improved, and the classification efficiency is improved. Based on contour information and multi-dimensional motion features, details and dynamic changes of the target can be captured more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of target classification and recognition, and in particular to a method and system for classifying and recognizing video moving targets based on contour constraints. Background Art

[0002] With the rapid development of video surveillance technology and computer vision, video moving target recognition has become one of the important technologies in the fields of intelligent security, traffic monitoring, smart home, etc. The task of moving target recognition is to automatically detect and classify moving objects from video data, which usually includes two main steps: target detection and target classification. However, due to the dynamic nature of video data, complex background, and changes in the target's motion state, the accurate recognition and classification of moving targets still faces many challenges.

[0003] In recent years, with the rise of deep learning technology, especially the widespread application of convolutional neural networks (CNN) in image classification and target detection, the accuracy and efficiency of target recognition have been significantly improved. By automatically learning image features, deep learning methods can effectively overcome the limitations of manual feature extraction in traditional methods and achieve remarkable results. However, although deep learning methods have achieved good performance in target recognition in static images, they still face certain challenges when facing complex background changes in dynamic video scenes, irregular target motion trajectories, and target morphological changes.

[0004] Based on the above, it is found that the limitations of the existing technology include at least the following problems. First, the existing methods are usually limited to a single feature extraction or template matching. When most technologies are applied in dynamic video scenes, they fail to fully consider the multi-dimensional changes of the target. These methods can extract certain static features of the target, but when faced with dynamic target behavior, they often lack in-depth analysis of the target's motion characteristics, resulting in low accuracy in the classification of moving targets, especially when faced with occlusion, deformation or rapid movement of the target, the recognition ability of the model is limited. Secondly, in the existing technology, some methods based on traditional deep learning models or image segmentation technology have not been able to effectively extract the contour information of the target, and most of the existing target classification methods are based on static analysis, which makes it difficult to comprehensively evaluate the motion characteristics of the target, which makes the target classification method perform poorly in complex dynamic scenes, especially when the behavior patterns and motion trajectories of different targets vary greatly, classification errors are prone to occur. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method and system for classifying and identifying moving targets in videos based on contour constraints, which solves the problems of low moving target recognition accuracy, poor adaptability to dynamic scenes and target classification errors existing in the prior art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for classifying and identifying video motion targets based on contour constraints, comprising the following steps: obtaining video data to be identified, wherein the video data includes several frames of image data, and each frame of image data includes several pixels and the pixel value and two-dimensional coordinates of each pixel; inputting the video data to be identified into a pre-trained machine learning model for contour extraction to obtain several contour areas of each frame of image, and each contour area includes several pixels; performing comprehensive analysis on the several contour areas of each frame of image to obtain several motion contour areas; for each motion contour area, analyzing the motion classification index respectively; performing judgment analysis on the motion classification index of each motion contour area and the preset motion classification interval respectively, and obtaining the category information of each motion contour area according to the judgment analysis results.

[0007] Furthermore, the machine learning model is specifically a convolutional neural network model, which includes an input layer for receiving each frame of image data, several convolution layers, activation layers and pooling layers, deconvolution layers, and an output layer for outputting several contour areas of each frame of image.

[0008] Furthermore, the specific steps for obtaining several contour areas of each frame image are as follows: in the input layer of the convolutional neural network model, receiving several frames of image data and performing preprocessing; in the convolutional layer of the convolutional neural network model, performing feature extraction processing on each frame of image data after preprocessing to obtain a feature map of each frame image; in the activation layer of the convolutional neural network model, performing nonlinear transformation processing on the feature map of each frame image; in the pooling layer of the convolutional neural network model, performing dimensionality reduction processing on the feature map of each frame image after nonlinear transformation processing to obtain a reduced dimensionality feature map of each frame image; in the deconvolution layer of the convolutional neural network model, performing spatial resolution restoration processing on the reduced dimensionality feature map of each frame image; in the output layer of the convolutional neural network model, performing contour recognition analysis on the reduced dimensionality feature map of each frame image after the spatial resolution restoration processing to obtain several contour areas of each frame image.

[0009] Furthermore, the specific steps of obtaining several motion contour areas are as follows: analyzing the edge gradient value of each contour area of ​​each frame image at each pixel point, and performing a comprehensive analysis in combination with the two-dimensional coordinates of each pixel point in each contour area of ​​each frame image to obtain the characteristic center coordinates of each contour area of ​​each frame image; counting the number of pixels in each contour area of ​​each frame image, and recording them as the area area value of each contour area of ​​each frame image, specifically determining the area area value of each contour area of ​​each frame image; analyzing several contour areas to be identified based on the area area value of each contour area of ​​each frame image, and the contour areas to be identified have corresponding characteristic center coordinates in each frame image; analyzing the displacement change index of each contour area to be identified in each frame image; performing judgment and analysis on the displacement change index of each contour area to be identified and the preset displacement change judgment interval, and marking the contour area to be identified whose displacement change index is higher than the preset displacement change judgment interval as a motion contour area; wherein the specific formula for calculating the characteristic center coordinates of each contour area of ​​each frame image is as follows: ;in, For the The frame image The feature center coordinates of the contour area, For the The frame image The contour area of ​​the The edge gradient value at the pixel point is is the intensity adjustment coefficient stored in the database, For the The frame image In the contour area The two-dimensional coordinates of the pixels, =1, 2, 3, ..., , is the number of image frames, =1, 2, 3, ..., , is the number of contour areas, =1, 2, 3, ..., , is the number of pixels in the contour area.

[0010] Furthermore, the specific steps of analyzing the displacement change index of each contour area to be identified are as follows: based on the area value of each contour area to be identified in each frame image, analyzing the adjacent frame area change values ​​of several groups of adjacent frame images of each contour area to be identified; based on the feature center coordinates of each contour area to be identified in each frame image, analyzing the adjacent frame position change values ​​and adjacent frame moving speed change values ​​of several groups of adjacent frame images of each contour area to be identified; comprehensively analyzing the adjacent frame area change values, adjacent frame position change values ​​and adjacent frame moving speed change values ​​of several groups of adjacent frame images of each contour area to be identified, to obtain the displacement change index of each contour area to be identified.

[0011] Furthermore, the specific formulas for calculating the adjacent frame area change values, adjacent frame position change values, adjacent frame moving speed change values, and displacement change index of several groups of adjacent frame images of each contour area to be identified are as follows: ;in, For the The contour area to be identified The area change value of the adjacent frame of the adjacent frame image. For the The contour area to be identified The area value of the region in the first frame of the group of adjacent frame images, For the The contour area to be identified The area value of the region in the second frame image of the group of adjacent frame images, For the The contour area to be identified The adjacent frame position change value of the adjacent frame image, For the The contour area to be identified The feature center coordinates in the first frame of the group of adjacent frame images, For the The contour area to be identified The feature center coordinates in the second frame of the group of adjacent frame images, For the The contour area to be identified The change value of the adjacent frame moving speed of the adjacent frame image. For the The time interval between adjacent frame images. For the The displacement change index of the contour area to be identified, is the area change weight coefficient stored in the database, is the position change weight coefficient stored in the database, is the moving speed change weight coefficient stored in the database, , =1, 2, 3, ..., , is the number of contour areas to be identified, =1, 2, 3, ..., , is the number of adjacent frame image groups.

[0012] Furthermore, for each motion contour area, the specific steps of analyzing the motion classification index are as follows: based on the characteristic center coordinates of each motion contour area in each frame image, analyzing the adjacent frame motion inertia index of each motion contour area; performing a comprehensive analysis on the adjacent frame motion inertia index, adjacent frame area change value, adjacent frame position change value and adjacent frame moving speed change value of several groups of adjacent frame images of each motion contour area to obtain the motion classification index of each motion contour area.

[0013] Furthermore, the specific formula for calculating the motion classification index of each motion contour area is as follows: ;in, For the The motion classification index of the motion contour area, is a natural constant, For the The inertia index of the adjacent frame motion of the motion contour area, is the motion inertia influence coefficient stored in the database, is the area change adjustment coefficient stored in the database, For the The first The area change value of the adjacent frame of the adjacent frame image. is the area change influence coefficient stored in the database, is the area change adjustment coefficient stored in the database, For the The first The adjacent frame position change value of the adjacent frame image, is the position change influence coefficient stored in the database, is the position change adjustment coefficient stored in the database, For the The first The change value of the adjacent frame moving speed of the adjacent frame image. is the moving speed change influence coefficient stored in the database, is the moving speed change adjustment coefficient stored in the database, , =1, 2, 3, ..., , is the number of motion contour areas, =1, 2, 3, ..., , is the number of adjacent frame image groups.

[0014] Furthermore, the specific formula for calculating the adjacent frame motion inertia index of each motion contour area is as follows: ;in, For the The inertia index of the adjacent frame motion of the motion contour area, For the The first The feature center coordinates in the first frame of the group of adjacent frame images, For the The first The feature center coordinates in the second frame of the group of adjacent frame images, For the The first The feature center coordinates in the first frame of the group of adjacent frame images, For the The first The feature center coordinates in the second frame of the group of adjacent frame images, is the position adjustment factor stored in the database, =1, 2, 3, ..., , is the number of motion contour areas, =1, 2, 3, ..., , is the number of adjacent frame image groups.

[0015] A video motion target classification and recognition system based on contour constraints includes: a video data acquisition module, a contour extraction module, a comprehensive analysis module, a motion analysis module, and a judgment analysis module; the video data acquisition module is used to acquire video data to be identified, the video data includes a plurality of frames of image data, and each frame of image data includes a plurality of pixels and a pixel value and a two-dimensional coordinate of each pixel; the contour extraction module is used to input the video data to be identified into a pre-trained machine learning model for contour extraction, and obtain a plurality of contour areas of each frame of image, and each contour area includes a plurality of pixels; the comprehensive analysis module is used to perform comprehensive analysis on a plurality of contour areas of each frame of image, and obtain a plurality of motion contour areas; the motion analysis module is used to analyze a motion classification index for each motion contour area; the judgment analysis module is used to perform judgment analysis on the motion classification index of each motion contour area and a preset motion classification interval, and obtain the category information of each motion contour area according to the judgment analysis result.

[0016] The present invention has the following beneficial effects:

[0017] (1) This method for classifying and identifying moving targets in videos based on contour constraints achieves accurate classification of moving targets in each frame of an image by introducing the joint application of contour constraints and deep learning models. Compared with traditional pixel-level classification methods and simple feature extraction methods, methods based on contour information and multi-dimensional motion features, including displacement, area, speed, etc., can more accurately capture the details and dynamic changes of the target. For example, by analyzing the target center coordinates, area, and motion trajectory, it can avoid misclassification problems caused by target occlusion, deformation, or rapid movement in complex scenes. In particular, when facing overlapping moving targets or complex shapes, it can effectively improve the accuracy of classification through accurate contour features and dynamic behavior analysis, thereby reducing misidentification caused by fuzzy features of static targets.

[0018] (2) This method for classifying and identifying moving targets in videos based on contour constraints not only considers the static features of the target at a certain moment through a comprehensive analysis of each group of adjacent frame images, but also combines the target's motion trajectory and dynamic behavior characteristics, such as displacement changes, speed changes, and area changes, so that the target classification can capture the changing trend of the moving target in real time. This method is particularly suitable for target classification in the presence of multi-target interference, high-speed motion, or complex backgrounds, effectively improving the adaptability to dynamic scenes and ensuring the robustness and accuracy of the classification process.

[0019] (3) This method for classifying and identifying moving targets in videos based on contour constraints uses a pre-trained convolutional neural network model for automatic feature extraction, avoiding the tedious process and inefficiency of manually designing features. The convolutional neural network can efficiently process image data while automatically learning the deep features of the target to improve the classification accuracy of moving targets. Especially under multi-target tracking and complex background interference, the convolutional neural network can quickly extract key information from the video stream, which not only improves the real-time performance but also greatly reduces the amount of calculation. By quickly analyzing each group of adjacent frame images, this method can maintain efficient target recognition and classification capabilities in application scenarios that require real-time response, such as video surveillance, traffic management, and security. In addition, the computational efficiency of this method is also suitable for use in real-time video stream analysis, ensuring efficient deployment and application in large-scale monitoring systems.

[0020] (4) The video motion target classification and recognition system based on contour constraints has significantly improved the flexibility and scalability of the system by modularizing multiple functions such as video data acquisition, contour extraction, comprehensive analysis, motion analysis and judgment analysis. Each module is independent in function and works closely together, so that the system can be customized according to different application scenarios or needs. For example, if it is necessary to optimize the processing process of a certain stage, such as the accuracy of contour extraction, it can be achieved by adjusting the contour extraction module without redesigning the entire system. This modular structure not only facilitates the independent optimization of functions, but also facilitates the expansion of the system. For example, new algorithms or new modules can be easily integrated in the future, such as more sophisticated target recognition or more efficient classification algorithms. This design concept enables the system to quickly adapt and maintain efficiency in application scenarios of different scales and complexities to meet ever-changing needs.

[0021] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The present invention is a flow chart of a method for classifying and identifying moving targets in videos based on contour constraints.

[0023] Figure 2 The present invention is a flowchart of the specific steps of obtaining a plurality of contour areas of each frame image in a method for classifying and identifying video moving targets based on contour constraints.

[0024] Figure 3 The present invention is a block diagram of a video moving target classification and recognition system based on contour constraints. DETAILED DESCRIPTION

[0025] The overall idea of ​​the problem in the embodiment of this application is as follows:

[0026] First, video data is obtained. The video data includes several frames of images. The data of each frame of image contains several pixels and their pixel values ​​and two-dimensional coordinates. Through preprocessing, the image data is normalized to prepare for the subsequent deep learning model input. The preprocessed image data is input into the pre-trained convolutional neural network model for contour extraction. Through this model, several contour areas of each frame of image are obtained, and the key features such as the characteristic center coordinates and area of ​​each contour area are further analyzed. The motion classification index is calculated for each contour area. By calculating the displacement change, area change, position change, speed change and other features, and combining with the adjacent frame images for comprehensive analysis, finally, the target is marked as a moving target or a static target based on the motion classification index and the preset motion classification interval.

[0027] See also Figure 1The embodiment of the present invention provides a technical solution: a method for classifying and identifying video motion targets based on contour constraints, comprising the following steps: obtaining video data to be identified, the video data including a plurality of frames of image data, and each frame of image data including a plurality of pixels and a pixel value and a two-dimensional coordinate of each pixel; inputting the video data to be identified into a pre-trained machine learning model for contour extraction to obtain a plurality of contour areas of each frame of image, and each contour area including a plurality of pixels; performing comprehensive analysis on the plurality of contour areas of each frame of image to obtain a plurality of motion contour areas; analyzing the motion classification index of each motion contour area respectively; performing judgment analysis on the motion classification index of each motion contour area and a preset motion classification interval respectively, and obtaining the category information of each motion contour area according to the judgment analysis result.

[0028] The machine learning model is specifically a convolutional neural network model, which includes an input layer for receiving image data of each frame, several convolutional layers, activation layers and pooling layers, deconvolution layers, and an output layer for outputting several contour areas of each frame of the image, specifically including the following contents:

[0029] Input layer: Receives each frame of image data as the input of the model. The image size is H×W×C, where H is the height of the image, W is the width of the image, and C is the number of channels of the image (usually C=3 for RGB images). The input image is normalized (for example, the pixel values ​​are scaled to [0,1]) as the model input.

[0030] The first convolutional layer: extracts low-level features (e.g., edges, textures) of the input image through convolution operations. Specifically, a convolutional layer with a convolution kernel W1 of size 3×3 is used to perform a convolution operation on the input image to extract local features of the image.

[0031] The first activation layer (ReLU): increases the nonlinear ability of the network, enabling the network to learn more complex features. Specifically, the ReLU activation function is used to perform a nonlinear transformation on the output of the first convolutional layer. ReLU changes negative values ​​to 0 and keeps positive values ​​unchanged, which can speed up the training process and perform better on the gradient vanishing problem.

[0032] The first pooling layer: The pooling operation reduces the spatial dimension of the feature map, reduces the amount of calculation, and retains important spatial feature information. Specifically, the maximum pooling operation of size 2×2 is used, and the maximum value in each 2×2 area is used as the result of pooling. This pooling layer can reduce the resolution of the image by half and retain the most important features of the image.

[0033] The second convolutional layer: further extracts higher-level features, such as the texture or boundary details of the target. Specifically, it performs a convolution operation on the pooled feature map. The convolution kernel W2 is still 3×3, and a new feature map is output.

[0034] Second activation layer (ReLU): Performs nonlinear transformation on the output of the second convolutional layer to increase the expressive power of the model. Specifically, the ReLU activation function is used to process the output of the second convolutional layer.

[0035] In the second pooling layer, the pooling operation is used again to reduce the dimension of the feature map and retain the key features in the image. Specifically, a 2×2 maximum pooling layer is applied again to further reduce the amount of calculation.

[0036] Deconvolution layer: restores the spatial resolution of the image, gradually restores the detailed information of the target, generates a feature map of the same size as the original image through deconvolution operation, and finally generates the outline of the target. Specifically, the deconvolution (upsampling) layer is used to upsample the pooled feature map to the original image size. Through deconvolution, we can gradually restore the boundary information of the target, so that the final output image has a higher resolution.

[0037] Output layer: Outputs whether each pixel belongs to the contour area. The output is a H×W×1 binary image. A value of 1 indicates that the pixel is a contour, and a value of 0 indicates that the pixel is a background. Specifically, the contour label of each pixel is generated through the last layer of convolution or deconvolution operation.

[0038] Among them, the pre-training steps of the convolutional neural network model are as follows:

[0039] Obtain several frames of image data: collect several frames of image data sets, the image data includes the target contour (i.e. the moving target contour area in the image). These image data can be single-frame images extracted from video frames. Each image includes the target area and its corresponding contour annotation (i.e. which pixels belong to the target contour and which belong to the background). Each group of images in the data set includes: input image (containing the target image to extract the contour) and target contour label (annotation for each pixel, a value of 1 indicates that the pixel belongs to the target contour, and a value of 0 indicates that the pixel is the background).

[0040] Preprocess each set of image data:

[0041] Image normalization: Normalize the pixel values ​​of the image to the range of [0, 1] or [-1, 1] to improve the efficiency and stability of training.

[0042] Image cropping and scaling: Scale all images to a uniform size H×W×C, where H and W are the height and width of the image, and C is the number of channels in the image (usually 3, indicating RGB images).

[0043] Data enhancement: Increase the diversity of the data set by rotating, flipping, cropping, etc. to help improve the generalization ability of the model.

[0044] Divide the data set: Divide all data sets into training set and validation set, usually in a ratio of 80%:20%.

[0045] Initialize the convolutional neural network (CNN):

[0046] Initialize the weights and biases in the network to ensure that the network can effectively learn features. Initialization methods include Xavier initialization and He initialization.

[0047] Convolutional layer weight initialization: Use He initialization (suitable for ReLU activation function).

[0048] Bias initialization: Initialize the bias term to zero.

[0049] Activation function: Use ReLU activation function to increase the nonlinearity of the network.

[0050] Training based on the image training set:

[0051] Number of training cycles: Set the number of training cycles (for example, 50 or 100 training cycles). In each training cycle, we update the weights of the network through the backpropagation algorithm.

[0052] The processing steps of each training cycle are:

[0053] Input training images: Randomly select a batch of images from the training set for training. Each training input is an image and its corresponding contour label.

[0054] Forward propagation: Each batch of images is input into the CNN network, and features are gradually extracted and contour maps are generated through convolutional layers, activation layers, pooling layers, deconvolution layers, etc.

[0055] Calculate loss: Calculate the error between the predicted image and the true label, and use the cross entropy loss function (suitable for pixel-level classification tasks) to measure the gap between the generated image and the real image.

[0056] Backpropagation: The gradient in the network is calculated according to the loss function and the model parameters are updated through an optimizer (such as the Adam optimizer).

[0057] After each training cycle, an evaluation analysis is performed based on the image validation set:

[0058] Use the images in the validation set for forward propagation and calculate the loss value between the predicted contour image and the true contour image.

[0059] Evaluation metrics: Calculate the loss function value and accuracy of the model on the validation set to evaluate the performance of the model.

[0060] Parameter adjustment: Based on the evaluation results, if the loss on the validation set does not decrease or the accuracy does not improve, you may need to adjust the network's hyperparameters (such as learning rate, network structure, etc.).

[0061] Adjust model parameters based on evaluation analysis:

[0062] If the performance on the validation set is poor, you can adjust hyperparameters such as learning rate, batch size, and the number of convolutional layers to improve the training effect.

[0063] Use the Early Stopping technique to stop training if the loss on the validation set does not improve significantly within several training cycles to prevent overfitting.

[0064] Adjust the network structure: Based on the training and verification results, the model's layers, convolution kernel size and other structures can be adjusted to further optimize the contour extraction effect.

[0065] When the training is completed and the loss and accuracy on the validation set reach the expected standards, the training process ends and a trained network model is obtained.

[0066] Final model: This model is able to accurately extract the outline of the target from the image and can be used for subsequent target recognition, target tracking and other tasks.

[0067] Specifically, Figure 2As shown, the specific steps of obtaining several contour areas of each frame of image are as follows: in the input layer of the convolutional neural network model, several frames of image data are received and preprocessed, specifically including: inputting each frame of image data into the model and preprocessing the image, the preprocessing includes data normalization (scaling the pixel value to the range of [0, 1] or [-1, 1]) and image scaling (uniformly adjusting the image size to adapt to the network input); in the convolution layer of the convolutional neural network model, feature extraction processing is performed on each frame of image data after preprocessing to obtain a feature map of each frame of image, specifically including: locally scanning the input image through multiple convolution kernels to extract low-level features of the image (such as edges, corners, textures, etc.), the size of the convolution kernel is usually 3×3 or 5×5, and the convolution operation will generate feature maps of the image, which represent the key information of the local area in the image; in the activation layer of the convolutional neural network model, nonlinear transformation processing is performed on the feature map of each frame of image, specifically including: using the ReLU activation function to perform nonlinear transformation on the feature map output by the convolution layer. The ReLU activation function changes all negative values ​​to zero while keeping positive values ​​unchanged. In this way, the activation layer enhances the learning ability of the network, enabling it to learn more complex features from the image, especially helping to capture high-level features such as the texture and shape of the target; in the pooling layer of the convolutional neural network model, the feature map of each frame of the image after nonlinear transformation is reduced in dimension to obtain a reduced-dimensional feature map of each frame of the image, specifically including: using maximum pooling or average pooling to reduce the dimension of the feature map to reduce the spatial resolution of the image while retaining the key information in the image. The pooling operation helps to reduce the amount of calculation while maintaining important features in the image, such as the edge or contour information of the target, thereby making subsequent convolutional and deconvolutional layers more efficient; in the deconvolution layer of the convolutional neural network model, the reduced-dimensional feature map of each frame of the image is restored to its original spatial resolution. The convolution neural network performs spatial resolution processing, including: restoring the spatial resolution of the feature map after the pooling layer through upsampling (deconvolution operation). The deconvolution operation helps the network generate a higher resolution contour map by gradually restoring the size and details of the image, so as to more clearly identify the boundary information of the target; in the output layer of the convolution neural network model, the reduced-dimensional feature map of each frame of the image after the spatial resolution is restored is subjected to contour recognition analysis to obtain several contour areas of each frame of the image, including: classifying each pixel to determine whether it belongs to the contour area of ​​the target, using the Sigmoid activation function to map the predicted value of each pixel to the range of [0,1], and outputting a binary image by setting the threshold. The value of 1 indicates that the pixel belongs to the contour of the target, and the value of 0 indicates that the pixel is the background. These contour areas are the target areas finally extracted.

[0068] In this embodiment, by using the convolution layer, the method can automatically extract low-level features (such as edges, corners, textures, etc.) from the image. This feature extraction process can ensure the comprehensive capture of image information, especially in target contour recognition, it can effectively extract the boundary features of the target, and provide an accurate basis for subsequent classification and tracking. Compared with the traditional method of manually designing features, the use of convolutional neural networks can automatically optimize feature selection, reduce human bias, and improve extraction efficiency. In the activation layer, the output of the convolution layer is nonlinearly transformed by the ReLU activation function, which further enhances the network's ability to learn complex features. The nonlinear transformation of ReLU enables the model to better learn complex patterns in the image, especially in capturing the morphological changes, textures, details, etc. of the target. Compared with the linear activation function, the use of ReLU can speed up the training process and is better in dealing with the gradient vanishing problem, thereby helping the network to accurately classify and identify targets in more complex scenarios. Through the pooling layer, especially the maximum pooling or average pooling operation, the spatial resolution of the image is effectively reduced, but the key information is retained. This dimensionality reduction process reduces the burden of subsequent calculations while maintaining the features in the image that are crucial to target contour recognition, such as edges and morphology. The pooling operation not only improves the computational efficiency, but also prevents overfitting, allowing the model to run efficiently when processing large-scale data. Through the deconvolution layer, that is, the upsampling operation, the low-resolution feature map after pooling is restored to the original image size, and the image details and target boundaries are clearly restored. This process ensures the accurate recognition of the target contour, especially in high-resolution images, and can accurately extract the boundary information of the target, thereby providing high-quality input for the classification and tracking of moving targets. Through the Sigmoid activation function of the output layer, the predicted value of each pixel is mapped to the range of [0,1], and the threshold setting is used to determine whether each pixel belongs to the target contour area. This process accurately delineates the target area and background through the binarization operation, and finally extracts the target contour area in each frame of the image. Through this pixel-level precise classification, not only can the accurate recognition of the target area be guaranteed, but also the target separation in a complex background can be effectively dealt with, and the accuracy of subsequent analysis (such as motion analysis) is enhanced.

[0069] Specifically, the specific steps for obtaining several motion contour areas are as follows: the edge gradient value of each contour area of ​​each frame image at each pixel point is analyzed by the Sobel operator or other gradient operations, and the characteristic center coordinates of each contour area of ​​each frame image are obtained by comprehensive analysis in combination with the two-dimensional coordinates of each pixel point in each contour area of ​​each frame image; the number of pixels in each contour area of ​​each frame image is counted, and they are recorded as the area area value of each contour area of ​​each frame image, specifically, the area area value of each contour area of ​​each frame image is determined; based on the area area value of each contour area of ​​each frame image, several contour areas to be identified are analyzed, and the contour areas to be identified have corresponding characteristic center coordinates in each frame image; based on the characteristic center coordinates of each contour area to be identified in each frame image, the displacement change index of each contour area to be identified is analyzed; the displacement change index of each contour area to be identified is respectively compared with the preset displacement change judgment interval, and the contour area to be identified whose displacement change index is higher than the preset displacement change judgment interval is marked as a motion contour area.

[0070] The specific formula for calculating the feature center coordinates of each contour area of ​​each frame image is as follows: ;in, For the The frame image The feature center coordinates of the contour area, For the The frame image The contour area of ​​the The edge gradient value at the pixel point is is the intensity adjustment coefficient stored in the database, For the The frame image In the contour area The two-dimensional coordinates of the pixels, =1, 2, 3, ..., , is the number of image frames, =1, 2, 3, ..., , is the number of contour areas, =1, 2, 3, ..., , is the number of pixels in the contour area.

[0071] In this embodiment, by using the Sobel operator or other gradient operations for edge detection, and combining the two-dimensional coordinates of each pixel, this method can accurately identify the contour area in each frame of the image. This method not only focuses on the outer boundary of the contour area, but also improves the accuracy of contour recognition by calculating the edge gradient value of each pixel point. It can effectively process areas with complex or blurred details in the image, avoiding the problem of easily ignoring contour details in traditional methods. By calculating the area value of each contour area, this method can effectively analyze the morphological changes of each target, thereby obtaining more detailed target contour features. The calculation of the area of ​​the region is crucial for multi-target classification and the evaluation of target size changes, especially in the case of target motion, deformation or occlusion. The area value can help the system better identify the dynamic changes of the target, making the target classification more accurate. By calculating the displacement change index of each contour area, this method not only focuses on the static contour of the target, but also combines the displacement change of the target between different frames to effectively quantify the motion characteristics of the target. This feature makes this method It can accurately distinguish dynamic moving targets from static targets in complex dynamic scenes, especially in the classification and tracking of moving targets, and avoid the influence of static targets or background interference on the system recognition effect. Through the comprehensive analysis of the feature center coordinates and the displacement change index, this method can evaluate the target's motion pattern in multiple dimensions, so that the classification of moving targets not only depends on position changes, but also considers the different dynamic characteristics of contour changes and target areas. This comprehensive feature analysis improves the robustness of the model, especially in the face of target morphological changes, rapid motion or partial occlusion, it can effectively improve the accuracy of classification and recognition. The displacement change index and the preset motion classification interval are judged and analyzed, making the classification decision of each contour area more accurate and faster. This process simplifies the processing of complex dynamic scenes through threshold judgment and classification interval setting, and avoids overfitting of the target motion pattern. This efficient judgment mechanism not only improves the classification speed, but also can process large-scale video data in real time, providing reliable support for intelligent monitoring systems and real-time target tracking.

[0072] Specifically, the specific steps of analyzing the displacement change index of each contour area to be identified are as follows: based on the area value of each contour area to be identified in each frame image, analyzing the adjacent frame area change values ​​of several groups of adjacent frame images of each contour area to be identified; based on the feature center coordinates of each contour area to be identified in each frame image, analyzing the adjacent frame position change values ​​and adjacent frame moving speed change values ​​of several groups of adjacent frame images of each contour area to be identified; comprehensively analyzing the adjacent frame area change values, adjacent frame position change values ​​and adjacent frame moving speed change values ​​of several groups of adjacent frame images of each contour area to be identified, to obtain the displacement change index of each contour area to be identified.

[0073] The specific formula for calculating the adjacent frame area change value, adjacent frame position change value, adjacent frame moving speed change value, and displacement change index of several groups of adjacent frame images of each contour area to be identified is as follows: ;in, For the The contour area to be identified The area change value of the adjacent frame of the adjacent frame image. For the The contour area to be identified The area value of the region in the first frame of the group of adjacent frame images, For the The contour area to be identified The area value of the region in the second frame image of the group of adjacent frame images, For the The contour area to be identified The adjacent frame position change value of the adjacent frame image, For the The contour area to be identified The feature center coordinates in the first frame of the group of adjacent frame images, For the The contour area to be identified The feature center coordinates in the second frame of the group of adjacent frame images, For the The contour area to be identified The change value of the adjacent frame moving speed of the adjacent frame image. For the The time interval between adjacent frame images. For the The displacement change index of the contour area to be identified, is the area change weight coefficient stored in the database, is the position change weight coefficient stored in the database, is the moving speed change weight coefficient stored in the database, , =1, 2, 3, ..., , is the number of contour areas to be identified, =1, 2, 3, ..., , is the number of adjacent frame image groups, and = -1.

[0074] It needs to be explained that , , The specific steps for obtaining are: determine through experimental design or actual verification methods, and continuously adjust through feedback from experimental data to ensure that the contribution of each feature in target classification reaches the best balance. Specifically, these coefficients reflect the importance of different features in image analysis. For example, changes in regional area may be more important for the recognition of static targets, while changes in position and speed play a decisive role in the recognition of dynamic targets. In order to optimize the performance of the model, these weight coefficients can be further optimized through tuning algorithms (such as grid search, Bayesian optimization) or cross-validation methods to ensure that the model can provide the best target recognition performance in different types of videos and scenes. In this way, the weight of each feature can be dynamically adjusted, so that the target recognition model can cope with changing environmental conditions and improve the accuracy and robustness of recognition.

[0075] In this embodiment, by analyzing the area change values, position change values ​​and moving speed change values ​​of adjacent frames, the method can comprehensively evaluate the dynamic characteristics of the target between consecutive image frames. The introduction of this multi-dimensional feature helps the system to better capture the motion pattern of the target, especially in the context of changes in target size, displacement and speed. By comprehensively considering these factors, the recognition accuracy of moving targets is significantly improved. For example, the displacement change and area change of the target combined with the speed change can accurately reflect the movement trend of the target, thereby reducing the misjudgment caused by changes in target shape, acceleration or deceleration. Through the displacement change index and the comprehensive analysis of adjacent frame images, it is possible to more accurately determine whether each target is a moving target. The displacement change index combined with information such as area change and speed change can quantify the dynamic changes of the target between consecutive frames, helping the system to effectively identify moving targets and static targets in complex backgrounds, especially for those targets in rapidly changing scenes or complex backgrounds. A high classification accuracy can be maintained. In this way, the system can avoid misidentification of stationary targets and missed identification of moving targets. , thereby improving the overall recognition effect. By calculating the change values ​​of adjacent frame images, especially the area change, position change and moving speed change, not only the accuracy of target classification is improved, but also the adaptability of the system in different environments is improved. For example, when facing fast-moving targets, partial occlusion or scenes with drastic changes in target shape, traditional methods are easily affected. This method can effectively reduce background interference and occlusion effects through comprehensive analysis of multiple features, and enhance the robustness of moving target classification. In addition, the model can flexibly adjust parameters according to the change coefficient to ensure that efficient classification results can be provided in different application scenarios. This method adjusts the influence of each feature on target classification through weighting coefficients, making the calculation process efficient and adjustable. For example, the area change weight coefficient, position change weight coefficient and moving speed change weight coefficient ensure the reasonable combination of different features and avoid excessive interference from some irrelevant features. By optimizing the setting of feature weights, the system can dynamically adjust the classification standard, further improving the calculation efficiency and classification accuracy of the system, and adapting to the efficient demand for target tracking and recognition in real-time video streams.

[0076] Specifically, for each motion contour area, the specific steps of analyzing the motion classification index are as follows: based on the characteristic center coordinates of each motion contour area in each frame image, analyze the adjacent frame motion inertia index of each motion contour area; conduct a comprehensive analysis of the adjacent frame motion inertia index, adjacent frame area change value, adjacent frame position change value and adjacent frame moving speed change value of several groups of adjacent frame images of each motion contour area to obtain the motion classification index of each motion contour area.

[0077] The specific formula for calculating the motion classification index of each motion contour area is as follows: ;in, For the The motion classification index of the motion contour area, is a natural constant, and in this embodiment, its value is 2.718. For the The inertia index of the adjacent frame motion of the motion contour area, is the motion inertia influence coefficient stored in the database, is the area change adjustment coefficient stored in the database, For the The first The area change value of the adjacent frame of the adjacent frame image. is the area change influence coefficient stored in the database, is the area change adjustment coefficient stored in the database, For the The first The adjacent frame position change value of the adjacent frame image, is the position change influence coefficient stored in the database, is the position change adjustment coefficient stored in the database, For the The first The change value of the adjacent frame moving speed of the adjacent frame image. is the moving speed change influence coefficient stored in the database, is the moving speed change adjustment coefficient stored in the database, , =1, 2, 3, ..., , is the number of motion contour areas, =1, 2, 3, ..., , is the number of adjacent frame image groups, and = -1.

[0078] It needs to be explained that , , , The specific steps for obtaining are as follows: These coefficients represent the degree of influence of each feature on the classification result among different features of the target image (such as area change, position change, speed change, etc.). Their specific values ​​are usually determined by experimental data or optimization algorithms. Experimental verification or cross-validation methods are usually used to optimize the values ​​of these coefficients so that the weights of different features in the target classification model are balanced, thereby improving the accuracy and stability of the classification. For example, the area change of the target may have a greater impact on some scenes, while the position change and speed change are particularly important for the identification of dynamic targets.

[0079] , , , The specific acquisition steps are as follows: These coefficients represent the adjustment coefficients of specific features, which are used to further adjust the contribution of each feature in calculating the target classification index. They are usually optimized through optimization algorithms (such as gradient descent or genetic algorithm) to dynamically adjust the importance of each feature in the target classification. The acquisition of specific values ​​is also based on actual data and feature analysis, so that the system can adaptively adjust the weights in different scenarios and target types to obtain the best classification effect.

[0080] The specific formula for calculating the adjacent frame motion inertia index of each motion contour area is as follows: ;in, For the The inertia index of the adjacent frame motion of the motion contour area, For the The first The feature center coordinates in the first frame of the group of adjacent frame images, For the The first The feature center coordinates in the second frame of the group of adjacent frame images, For the The first The feature center coordinates in the first frame of the group of adjacent frame images, For the The first The feature center coordinates in the second frame of the group of adjacent frame images, It is the position adjustment factor stored in the database, which is used to prevent the denominator from being 0. =1, 2, 3, ..., , is the number of motion contour areas, =1, 2, 3, ..., , is the number of adjacent frame image groups, and = -1.

[0081] In this embodiment, by analyzing the area change values, position change values ​​and moving speed change values ​​of adjacent frames, the method can comprehensively evaluate the dynamic characteristics of the target between consecutive image frames. The introduction of this multi-dimensional feature helps the system to better capture the motion pattern of the target, especially in the context of changes in target size, displacement and speed. By comprehensively considering these factors, the recognition accuracy of moving targets is significantly improved. For example, the displacement change and area change of the target combined with the speed change can accurately reflect the movement trend of the target, thereby reducing the misjudgment caused by changes in target shape, acceleration or deceleration. Through the displacement change index and the comprehensive analysis of adjacent frame images, it is possible to more accurately determine whether each target is a moving target. The displacement change index combined with information such as area change and speed change can quantify the dynamic changes of the target between consecutive frames, helping the system to effectively identify moving targets and static targets in complex backgrounds, especially for those targets in rapidly changing scenes or complex backgrounds. A high classification accuracy can be maintained. In this way, the system can avoid misidentification of stationary targets and missed identification of moving targets. , thereby improving the overall recognition effect. By calculating the change values ​​of adjacent frame images, especially the area change, position change and moving speed change, not only the accuracy of target classification is improved, but also the adaptability of the system in different environments is improved. For example, when facing fast-moving targets, partial occlusion or scenes with drastic changes in target shape, traditional methods are easily affected. This method can effectively reduce background interference and occlusion effects through comprehensive analysis of multiple features, and enhance the robustness of moving target classification. In addition, the model can flexibly adjust parameters according to the change coefficient to ensure that efficient classification results can be provided in different application scenarios. This method adjusts the influence of each feature on target classification through weighting coefficients, making the calculation process efficient and adjustable. For example, the area change weight coefficient, position change weight coefficient and moving speed change weight coefficient ensure the reasonable combination of different features and avoid excessive interference from some irrelevant features. By optimizing the setting of feature weights, the system can dynamically adjust the classification standard, further improving the calculation efficiency and classification accuracy of the system, and adapting to the efficient demand for target tracking and recognition in real-time video streams.

[0082] See also Figure 3The embodiment of the present invention provides a technical solution: a video motion target classification and recognition system based on contour constraints, comprising: a video data acquisition module, a contour extraction module, a comprehensive analysis module, a motion analysis module, and a judgment analysis module; the video data acquisition module is used to acquire video data to be identified, the video data includes a plurality of frames of image data, and each frame of image data includes a plurality of pixels and a pixel value and a two-dimensional coordinate of each pixel; the contour extraction module is used to input the video data to be identified into a pre-trained machine learning model for contour extraction, and obtain a plurality of contour areas of each frame of image, and each contour area includes a plurality of pixels; the comprehensive analysis module is used to perform comprehensive analysis on a plurality of contour areas of each frame of image, and obtain a plurality of motion contour areas; the motion analysis module is used to analyze a motion classification index for each motion contour area; the judgment analysis module is used to perform judgment analysis on the motion classification index of each motion contour area and a preset motion classification interval, and obtain the category information of each motion contour area according to the judgment analysis result.

[0083] In summary, this application has at least the following effects:

[0084] By introducing the joint application of contour constraints and deep learning models, accurate classification of moving targets in each frame of the image is achieved. Compared with traditional pixel-level classification methods and simple feature extraction methods, methods based on contour information and multi-dimensional motion features, including displacement, area, speed, etc., can more accurately capture the details and dynamic changes of the target. For example, by analyzing the target center coordinates, area and motion trajectory, it can avoid misclassification problems caused by occlusion, deformation or rapid movement of the target in complex scenes. Especially when facing overlapping or complex moving targets, it can effectively improve the accuracy of classification through precise contour features and dynamic behavior analysis, thereby reducing misidentification caused by fuzzy features of static targets.

[0085] Through comprehensive analysis of each group of adjacent frame images, not only the static characteristics of the target at a certain moment are considered, but also the motion trajectory and dynamic behavior characteristics of the target, such as displacement changes, speed changes and area changes, so that the target classification can capture the changing trend of the moving target in real time. This method is particularly suitable for target classification in multi-target interference, high-speed motion or complex backgrounds, effectively improving the adaptability to dynamic scenes and ensuring the robustness and accuracy of the classification process.

[0086] A pre-trained convolutional neural network model is used for automatic feature extraction, avoiding the tedious process and inefficiency of manually designing features. The convolutional neural network can efficiently process image data while improving the classification accuracy of moving targets by automatically learning the deep features of the target. Especially under multi-target tracking and complex background interference, the convolutional neural network can quickly extract key information from the video stream, which not only improves the real-time performance, but also greatly reduces the amount of calculation. By quickly analyzing each group of adjacent frame images, this method can maintain efficient target recognition and classification capabilities in application scenarios that require real-time response, such as video surveillance, traffic management, and security. In addition, the computational efficiency of this method is also suitable for use in real-time video stream analysis, ensuring efficient deployment and application in large-scale monitoring systems.

[0087] The flexibility and scalability of the system are significantly improved by modularizing multiple functions such as video data acquisition, contour extraction, comprehensive analysis, motion analysis and judgment analysis. Each module is functionally independent and works closely together, so that the system can be customized according to different application scenarios or needs. For example, if it is necessary to optimize the processing process at a certain stage, such as the accuracy of contour extraction, it can be achieved by adjusting the contour extraction module without redesigning the entire system. This modular structure not only facilitates the independent optimization of functions, but also facilitates the expansion of the system. For example, new algorithms or new modules can be easily integrated in the future, such as more sophisticated target recognition or more efficient classification algorithms. This design idea enables the system to quickly adapt and maintain efficiency in application scenarios of different scales and complexities to meet changing needs.

[0088] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0089] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for classifying and identifying moving objects in a video based on contour constraints, characterized in that: The following steps are involved: Acquire video data to be identified, wherein the video data includes a plurality of frames of image data, and each frame of image data includes a plurality of pixels and a pixel value and a two-dimensional coordinate of each pixel; The video data to be identified is input into the pre-trained machine learning model for contour extraction to obtain several contour areas of each frame image, and each contour area includes several pixel points; Perform comprehensive analysis on several contour areas of each frame image to obtain several motion contour areas; For each motion contour region, the motion classification index was analyzed separately; The motion classification index of each motion contour area is respectively compared with the preset motion classification interval, and the category information of each motion contour area is obtained according to the judgment and analysis results.

2. The method for classifying and identifying moving objects in video based on contour constraints according to claim 1 is characterized in that: The machine learning model is specifically a convolutional neural network model, which includes an input layer for receiving image data of each frame, several convolution layers, activation layers and pooling layers, a deconvolution layer, and an output layer for outputting several contour areas of each frame of image.

3. The method for classifying and identifying moving objects in video based on contour constraints according to claim 2 is characterized in that: The specific steps to obtain several contour areas of each frame image are as follows: In the input layer of the convolutional neural network model, several frames of image data are received and preprocessed; In the convolution layer of the convolutional neural network model, feature extraction is performed on each frame of image data after preprocessing to obtain a feature map of each frame of image; In the activation layer of the convolutional neural network model, the feature map of each frame image is subjected to nonlinear transformation processing; In the pooling layer of the convolutional neural network model, the feature map of each frame image after nonlinear transformation is subjected to dimensionality reduction processing to obtain a dimensionality reduction feature map of each frame image; In the deconvolution layer of the convolutional neural network model, the reduced-dimensional feature map of each frame is processed to restore the spatial resolution; In the output layer of the convolutional neural network model, contour recognition analysis is performed on the reduced-dimensional feature map of each frame of image after the spatial resolution is restored to obtain several contour areas of each frame of image.

4. The method for classifying and identifying moving objects in video based on contour constraints according to claim 3 is characterized in that: The specific steps to obtain several motion contour areas are as follows: Analyze the edge gradient value at each pixel point of each contour area of ​​each frame image, and perform comprehensive analysis based on the two-dimensional coordinates of each pixel point in each contour area of ​​each frame image to obtain the feature center coordinates of each contour area of ​​each frame image; Counting the number of pixels in each contour area of ​​each frame image, and recording them as the area value of each contour area of ​​each frame image, specifically determining the area value of each contour area of ​​each frame image; Based on the area value of each contour region of each frame image, a number of contour regions to be identified are analyzed, and the contour regions to be identified have corresponding feature center coordinates in each frame image; Based on the feature center coordinates of each contour area to be identified in each frame of the image, the displacement change index of each contour area to be identified is analyzed; The displacement change index of each contour area to be identified is respectively compared with the preset displacement change judgment interval, and the contour area to be identified whose displacement change index is higher than the preset displacement change judgment interval is marked as a motion contour area; The specific formula for calculating the feature center coordinates of each contour area of ​​each frame image is as follows: ； in, For the The frame image The feature center coordinates of the contour area, For the The frame image The contour area of ​​the The edge gradient value at the pixel point is is the intensity adjustment coefficient stored in the database, For the The frame image In the contour area The two-dimensional coordinates of the pixels, =1, 2, 3, ..., , is the number of image frames, =1, 2, 3, ..., , is the number of contour areas, =1, 2, 3, ..., , is the number of pixels in the contour area.

5. The method for classifying and identifying moving objects in video based on contour constraints according to claim 4 is characterized in that: The specific steps for analyzing the displacement change index of each contour area to be identified are as follows: Based on the area value of each contour region to be identified in each frame of image, analyzing the adjacent frame area change values ​​of several groups of adjacent frame images of each contour region to be identified; Based on the feature center coordinates of each contour region to be identified in each frame image, analyzing adjacent frame position change values ​​and adjacent frame movement speed change values ​​of several groups of adjacent frame images of each contour region to be identified; Comprehensive analysis is performed on adjacent frame area change values, adjacent frame position change values ​​and adjacent frame moving speed change values ​​of several groups of adjacent frame images of each contour area to be identified, so as to obtain a displacement change index of each contour area to be identified.

6. The method for classifying and identifying moving objects in video based on contour constraints according to claim 5 is characterized in that: The specific formula for calculating the adjacent frame area change value, adjacent frame position change value, adjacent frame moving speed change value, and displacement change index of several groups of adjacent frame images of each contour area to be identified is as follows: ； in, For the The contour area to be identified The area change value of the adjacent frame of the adjacent frame image. For the The contour area to be identified The area value of the region in the first frame of the group of adjacent frame images, For the The contour area to be identified The area value of the region in the second frame image of the group of adjacent frame images, For the The contour area to be identified The adjacent frame position change value of the adjacent frame image, For the The contour area to be identified The feature center coordinates in the first frame of the group of adjacent frame images, For the The contour area to be identified The feature center coordinates in the second frame of the group of adjacent frame images, For the The contour area to be identified The change value of the adjacent frame moving speed of the adjacent frame image. For the The time interval between adjacent frame images. For the The displacement change index of the contour area to be identified, is the area change weight coefficient stored in the database, is the position change weight coefficient stored in the database, is the moving speed change weight coefficient stored in the database, , =1, 2, 3, ..., , is the number of contour areas to be identified, =1, 2, 3, ..., , is the number of adjacent frame image groups.

7. The method for classifying and identifying moving objects in video based on contour constraints according to claim 5, characterized in that: For each motion contour region, the specific steps of analyzing the motion classification index are as follows: Based on the feature center coordinates of each motion contour area in each frame image, analyzing the motion inertia index of each motion contour area in the adjacent frames; A comprehensive analysis is performed on adjacent frame motion inertia indexes, adjacent frame area change values, adjacent frame position change values ​​and adjacent frame moving speed change values ​​of several groups of adjacent frame images of each motion contour area to obtain a motion classification index for each motion contour area.

8. The method for classifying and identifying moving objects in video based on contour constraints according to claim 7 is characterized in that: The specific formula for calculating the motion classification index of each motion contour area is as follows: ； in, For the The motion classification index of the motion contour area, is a natural constant, For the The inertia index of the adjacent frame motion of the motion contour area, is the motion inertia influence coefficient stored in the database, is the area change adjustment coefficient stored in the database, For the The first The area change value of the adjacent frame of the adjacent frame image. is the area change influence coefficient stored in the database, is the area change adjustment coefficient stored in the database, For the The first The adjacent frame position change value of the adjacent frame image, is the position change influence coefficient stored in the database, is the position change adjustment coefficient stored in the database, For the The first The change value of the adjacent frame moving speed of the adjacent frame image. is the influence coefficient of the moving speed change stored in the database, is the moving speed change adjustment coefficient stored in the database, , =1, 2, 3, ..., , is the number of motion contour areas, =1, 2, 3, ..., , is the number of adjacent frame image groups.

9. The method for classifying and identifying moving objects in video based on contour constraints according to claim 7, characterized in that: The specific formula for calculating the adjacent frame motion inertia index of each motion contour area is as follows: ； in, For the The inertia index of the adjacent frame motion of the motion contour area, For the The first The feature center coordinates in the first frame of the group of adjacent frame images, For the The first The feature center coordinates in the second frame of the group of adjacent frame images, For the The first The feature center coordinates in the first frame of the group of adjacent frame images, For the The first The feature center coordinates in the second frame of the group of adjacent frame images, is the position adjustment factor stored in the database, =1, 2, 3, ..., , is the number of motion contour areas, =1, 2, 3, ..., , is the number of adjacent frame image groups.

10. A video moving target classification and recognition system based on contour constraints, using the video moving target classification and recognition method based on contour constraints according to any one of claims 1 to 9, characterized in that: include: Video data acquisition module, contour extraction module, comprehensive analysis module, motion analysis module, judgment analysis module; The video data acquisition module is used to acquire the video data to be identified, wherein the video data includes a plurality of frames of image data, and each frame of image data includes a plurality of pixels and a pixel value and a two-dimensional coordinate of each pixel; The contour extraction module is used to input the video data to be identified into a pre-trained machine learning model to perform contour extraction, and obtain a number of contour areas of each frame image, and each contour area includes a number of pixel points; The comprehensive analysis module is used to perform comprehensive analysis on a plurality of contour areas of each frame image to obtain a plurality of motion contour areas; The motion analysis module is used to analyze the motion classification index for each motion contour area; The judgment and analysis module is used to perform judgment and analysis on the motion classification index of each motion contour area and the preset motion classification interval, and obtain the category information of each motion contour area according to the judgment and analysis result.

Images