Sports video key scene time node positioning method

Abstract

The invention discloses a sports video key scene time node positioning method, which relates to the field of video processing, and comprises the following steps: obtaining sports data, and setting a sports management space; obtaining sports video data; dividing the historical sports video data in the sports management space according to the sports data, and setting a key sports data set; constructing a three-layer branch processing model based on the key motion data set, and performing importance evaluation processing from three dimensions; performing resource allocation processing according to the importance evaluation result to obtain a resource allocation result; performing three-dimensional division matching on a corresponding image area in the sports video data according to the resource allocation result to obtain a hierarchical division result of three dimensions; carrying out collaborative integration according to a hierarchical division result and a three-layer branch processing model, and obtaining key scene time node positioning information of the corresponding sports video data; the sports video processing efficiency is improved to a certain extent.

Description

Technical Field

[0001] This invention relates to the field of video processing technology, and in particular to a method for locating key time nodes in sports videos. Background Technology

[0002] With the widespread use of the internet and mobile devices, the broadcasting channels for sports events have become increasingly diversified. In addition to traditional television broadcasts, online video platforms have become an important way for viewers to watch events. The emergence of a large amount of sports video content has led to an increasing demand from users for quick location of exciting moments. Especially in the absence of a global perspective or manual annotation, automatically identifying and accurately locating key scene time points has become a key technical challenge to improve the viewing experience. Existing methods mostly rely on manual annotation or simple feature extraction, which makes it difficult to achieve efficient and accurate scene segmentation and location. A search revealed Chinese patent CN117253174A, which discloses a method, system, and device for motion localization in sports videos based on deep learning, relating to the field of video analysis. The method includes: acquiring sports video data and generating an image dataset; inputting the image dataset into a pre-trained deep learning network for video understanding to learn video motion features at each moment, and integrating the video motion features from multiple consecutive moments into complete video motion features; inputting the complete video motion features into a multi-scale refinement network based on a feature pyramid network to obtain multi-scale offset information based on time points; combining the multi-scale offset information with the motion localization result using an offset focusing strategy to obtain a refined motion localization result; and performing localization error analysis based on the refined motion localization result. This invention enables high-quality motion localization in sports videos.

[0003] Compared with existing technologies, the invention patent with Chinese patent number CN117253174A can extract sports video action features through a pre-trained video understanding network. Based on the complete video feature sequence, it applies a multi-scale refinement network based on a feature pyramid network to predict offset information. According to the offset focusing strategy, the offset information is combined with the positioning results to achieve high-quality positioning of sports video actions.

[0004] However, in actual use, the above methods cannot adaptively extract and adjust the motion features of sports videos from different perspectives during the process of extracting motion features through pre-trained video understanding networks, which to some extent affects the efficiency of the motion localization process in sports videos. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of insufficient efficiency in existing technologies by proposing a method for locating key time nodes in sports videos.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for locating key time nodes in sports videos includes the following steps: Step S1: Set up a sports video management platform, obtain relevant sports data through the sports video management platform, and set up corresponding sports management spaces according to the type of sports data; Step S2: Obtain sports video data, determine the corresponding type based on the sports video data, and input the sports video data into the sports management space according to the corresponding type; Step S3: Divide the sports video data in the corresponding sports management space into key sports areas, key personnel, and key sports skills, and set up corresponding key sports datasets according to the division results; Step S4: Set up a three-layer branch processing model for the obtained key motion dataset according to the division criteria; obtain sports training data, divide the obtained sports training data into multiple time-series frame images according to the time order, and input the time-series frame images corresponding to the sports training data into the three-layer branch processing model in sequence to perform localization prediction analysis from three dimensions, and obtain the importance evaluation data of the corresponding image region in the predicted time-series frame image. Step S5: Perform resource allocation processing on the importance assessment data of the corresponding image region in the predicted time-series frame image corresponding to the sports video data in the motion management space, and obtain the resource allocation result corresponding to each pixel in the time-series frame image; Step S6: Based on the resource allocation results at the corresponding pixels, the time-series frame image corresponding to the sports video data is divided and matched from three dimensions to obtain the hierarchical division results of the corresponding image region in three dimensions; Step S7: The obtained hierarchical division results are sequentially integrated according to the three-layer branch processing model to obtain the key scene time node positioning information corresponding to the sports video data.

[0007] The above technical solution further includes: the process of acquiring relevant sports data through the sports video management platform includes: A sports video management platform is set up, which includes a data input terminal, a data processing terminal, and a data acquisition terminal. The data entry terminal is used to enter sports data from the sports video management platform. The sports data includes sports type data, sports venue data, and sports rule data. The data processing terminal is used to classify and process the obtained sports data, obtain the sports type corresponding to the sports data, and set up corresponding sports management space in the sports video management platform according to the sports type.

[0008] Furthermore, the process of acquiring sports video data includes: The data acquisition terminal is used to acquire sports video data and input the acquired sports video data into the data processing terminal; The data processing terminal inputs the obtained sports video data into the corresponding sports management space.

[0009] Furthermore, the process of dividing sports video data within the corresponding sports management space into sports areas, key personnel, and sports skills includes: Acquire historical sports video data within the sports management space, and then divide the acquired historical sports video data into multiple dimensions according to the corresponding sports data; Based on sports field data within sports data, historical sports video data is divided into sports regions to obtain key score-related regions, key rule restriction regions, and transition regions within the corresponding sports regions in the sports management space. Based on the sports rules data within the sports data, key personnel and key sports skills are segmented from historical sports video data. This allows for the acquisition of sports responsibility information and historical sports information for the corresponding key personnel within the sports management space; and the acquisition of skill function information and skill characteristic information for the corresponding sports skills within the sports management space for each sports type. Based on the multi-dimensional segmentation results and the key scenes in historical sports video data, corresponding key sports datasets were set up.

[0010] Furthermore, the process of setting up a three-layer branch processing model based on the obtained motion dataset according to the partitioning criteria includes: Obtain the key motion datasets corresponding to each type of motion within the motion management space. Analyze and process the obtained key motion datasets using deep learning algorithms to construct a three-layer branch processing model, which includes a character branch processing model, an action branch processing model, and a region branch processing model. The corresponding sports video data within the motion management space is input into a three-layer branch processing model. Based on the three dimensions of people, actions, and regions in the time-series frame images, the model performs localization prediction analysis to obtain the predicted location information of the corresponding image regions in the predicted time-series frame images where key scenes may exist, and obtains importance assessment data for the three dimensions.

[0011] Furthermore, the process of obtaining the resource allocation results corresponding to each pixel point within the time-series frame image includes: Obtain the prediction location information of the three-layer branch processing model and the importance evaluation data of the three dimensions of each image region in the prediction time frame image corresponding to the sports video data; The importance assessment data of each predicted position information in each key motion dataset in the three-layer branch processing model are fused and analyzed to obtain the correlation data between different characters and actions, characters and regions, and actions and regions. Based on the obtained correlation data, a correlation attention model based on the three-layer branch processing model is constructed. The importance assessment data of the three-layer branch processing model are input into the corresponding relevant attention model for analysis and processing to obtain the weight coefficients of the corresponding importance assessment data. The comprehensive evaluation data for the corresponding image region is obtained by weighting and integrating the importance assessment data of the three dimensions of each image region and the corresponding weight coefficients. The comprehensive evaluation data obtained from each image region is sequentially mapped into the predicted time frame image. The mapping results are then visualized, and the visualization results are averaged. Based on the averaged results, the resource management data is integrated and divided to obtain unit resource data. The obtained unit resource data is divided into pre-emptive elastic resource data and real-time elastic resource data. Based on the visualization processing results of the predicted time-series frame image, the pre-emptive elastic resource data is processed for resource allocation to obtain the resource allocation results corresponding to each image region of the predicted time-series frame image.

[0012] Furthermore, the process of segmenting and matching the time-series frame images corresponding to the sports video data from three dimensions based on the resource allocation results at the corresponding image regions includes: Obtain the resource allocation results corresponding to the sports video data. Based on the obtained resource allocation results, perform resource allocation processing on the preliminary missile resource data. Compare and analyze the image regions corresponding to the preliminary missile resource data with the predicted time-series frame images to obtain similarity data. Based on similarity data, real-time elastic resource data is allocated in real time to obtain the resource allocation results of real-time time-series frame images corresponding to sports video data; Based on the resource allocation results at the corresponding image regions, edge processing is performed on the time-series frame images corresponding to the sports video data to obtain the adjusted image regions. Each image region within the time-series frame image is divided and matched from three dimensions to obtain the hierarchical division results of the corresponding image region in the three dimensions.

[0013] Furthermore, the process of obtaining the key scene time node positioning information corresponding to the relevant sports video data includes: Obtain the hierarchical division results of each image region within a time-series frame image in three dimensions; Obtain the key motion dataset corresponding to the three-layer branch processing model, and set up a key scene matching and sorting library for the key motion dataset according to the combination results of the three-layer branch processing model. Based on the hierarchical division results corresponding to each image region, the corresponding key scene matching and sorting library is obtained. The scene information in the image region is matched and analyzed with the key scene matching and sorting library in turn to determine whether there is a key scene. If there is, the key scene type is obtained. Based on the key scene type, set the sliding window parameter information, and perform retrieval and analysis on the corresponding image regions in each time frame image of the sports video data according to the sliding window parameter information to obtain the key scene time node positioning information corresponding to the sports video data.

[0014] The present invention has the following beneficial effects: 0. In this invention, sports data is collected, classified, and processed. A corresponding sports management space is set up, and sports video data is input into the sports management space for analysis and processing. Based on the sports data, historical sports video data in the corresponding sports management space is analyzed and processed to obtain three-dimensional division results. Based on the corresponding division results, a three-layer branch processing model is set up to perform positioning prediction analysis from the three dimensions of person, action, and region. Based on the positioning prediction analysis results, importance assessment data of the corresponding image region is obtained. Based on the importance assessment data, resource allocation processing is performed on the corresponding image region, which can improve the efficiency of sports video processing to a certain extent.

[0015] 1. In this invention, the corresponding image regions within the corresponding time-series frames of sports video data are divided and matched from three dimensions. Based on the hierarchical division results, the key scenes within the corresponding image regions are set in a corresponding order for matching analysis. The sliding window parameter information is adjusted in real time based on the matching results. The key scene time nodes within the corresponding image regions are located and retrieved using the sliding window parameter information. The corresponding positioning steps can be adaptively adjusted according to the sports video data, thereby improving the efficiency of sports video processing to a certain extent. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for locating key time nodes in sports videos proposed in this invention. Detailed Implementation

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

[0018] Example 1 like Figure 1 As shown, the present invention proposes a method for locating key time nodes in sports videos, which includes the following steps: Step S1: Set up a sports video management platform, obtain relevant sports data through the sports video management platform, and set up corresponding sports management spaces according to the type of sports data; Step S2: Obtain sports video data, determine the corresponding type based on the sports video data, and input the sports video data into the sports management space according to the corresponding type; Step S3: Divide the sports video data in the corresponding sports management space into key sports areas, key personnel, and key sports skills, and set up corresponding key sports datasets according to the division results; Step S4: Set up a three-layer branch processing model for the obtained key motion dataset according to the division criteria; obtain sports training data, divide the obtained sports training data into multiple time-series frame images according to the time order, and input the time-series frame images corresponding to the sports training data into the three-layer branch processing model in sequence to perform localization prediction analysis from three dimensions, and obtain the importance evaluation data of the corresponding image region in the predicted time-series frame image. Step S5: Perform resource allocation processing on the importance assessment data of the corresponding image region in the predicted time-series frame image corresponding to the sports video data in the motion management space, and obtain the resource allocation result corresponding to each pixel in the time-series frame image; Step S6: Based on the resource allocation results at the corresponding pixels, the time-series frame image corresponding to the sports video data is divided and matched from three dimensions to obtain the hierarchical division results of the corresponding image region in three dimensions; Step S7: The obtained hierarchical division results are sequentially integrated according to the three-layer branch processing model to obtain the key scene time node positioning information corresponding to the sports video data. As shown in steps S1 to S7 above, the present invention performs multi-step adaptive adjustment processing on the obtained sports video data by sequentially processing the data according to the corresponding sports data, three-dimensional angle, resource allocation adjustment, hierarchical division processing and collaborative integration. This improves the efficiency of sports video data processing to a certain extent and can maximize the efficiency of key scene time node positioning.

[0019] In the specific implementation process, the step S1 of setting up a sports video management platform, obtaining relevant sports data through the sports video management platform, and setting up corresponding sports management spaces according to the type of sports data includes: S11: Set up a sports video management platform, which includes a data input terminal, a data processing terminal, and a data acquisition terminal; S12: The data entry terminal is used to enter sports data from the sports video management platform. The sports data includes sports type data, sports venue data, and sports rule data. S13: The data processing terminal is used to classify and process the obtained sports data, obtain the sports type corresponding to the sports data, and set up the corresponding sports management space in the sports video management platform according to the sports type; As shown in steps S11 to S13 above, by setting up corresponding sports management spaces for different types of sports data, the sports video data corresponding to the corresponding sports data types can be input into the sports management space for unified analysis and processing. This can, to a certain extent, integrate the utilization rate of corresponding resources within the sports video management platform. In addition, sports type data includes various types such as corresponding track and field events and ball games. Sports field data includes the field size and function data involved in the corresponding type of sports. Sports rule data includes the scoring rules, penalty rules, and other competition rule data involved in the corresponding type of sports, which facilitates the subsequent classification and processing of sports video data at different levels.

[0020] In the specific implementation process, step S2, which involves acquiring sports video data, determining the corresponding type based on the sports video data, and inputting the sports video data into the sports management space according to the corresponding type, includes: S21: The data acquisition terminal is used to acquire sports video data and input the acquired sports video data into the data processing terminal; S22: The data processing terminal compares and analyzes the obtained sports video data with the sports data to obtain the sports type of the sports video data, and then connects the sports video data to the sports management space according to the corresponding sports type; As shown in steps S21 to S22 above, within the sports video management platform, the obtained sports video data is pre-classified according to the type of the corresponding sports management space, and sports video data of the same type is input into the corresponding sports management space for unified processing, thereby improving the efficiency of different types of sports video data in the key scene time node positioning process within the sports video management platform.

[0021] In the specific implementation process, the step S3 of dividing the sports video data within the corresponding sports management space into key sports regions, key personnel, and key sports skills, and setting up corresponding key sports datasets based on the division results, includes: S31: Obtain historical sports video data within the sports management space, and divide the obtained historical sports video data into multiple dimensions according to the corresponding sports data; S311: Based on the sports field data in the sports data, divide the historical sports video data into sports regions and obtain the key score-related regions, key rule restriction regions and transition regions of the corresponding sports regions in the sports management space; S312: Based on the sports rule data within the sports data, key personnel are categorized in historical sports video data. The sports responsibility information and historical sports information of the corresponding key personnel within the sports management space are obtained. Among them, the sports responsibility information includes the corresponding athlete type, such as the position and role of different athletes in ball sports, the referee on the field, etc. The historical sports information is obtained based on the identity information of the corresponding key personnel in the corresponding historical sports video data, and the relevant video segments in the historical sports video data of the corresponding athletes are obtained based on the corresponding identity information. S313: Based on the sports rule data within the sports data, the historical sports video data is divided into key sports skills, and the skill function information and skill characteristic information of the corresponding sports skills in the sports management space corresponding to the sports type are obtained. Among them, the skill function information and physical characteristic information include the common movements and common characteristics of the corresponding sports type, such as passing skills, shooting skills, and receiving skills in ball sports. S32 matches the multi-dimensional segmentation results with key scenes in historical sports video data and sets up corresponding key motion datasets. As shown in steps S31 to S32 above, the historical sports video data within the corresponding sports management space are divided into three dimensions: region, personnel, and motor skills. Target monitoring is performed based on the Faster R-CNN model to obtain the associated key scenes within the historical sports video data corresponding to the division results. The obtained associated key scenes are then processed in accordance with the corresponding differentiation results. The key scenes involved include historical sports video data corresponding to the preceding and following time periods, and the corresponding key sports dataset is obtained, providing data support for the subsequent construction of a three-layer branch processing model.

[0022] In the specific implementation process, the step S4, which involves setting up a three-layer branch processing model based on the division criteria for the obtained key sports dataset; acquiring sports training data; dividing the acquired sports training data into multiple time-series frame images according to time sequence; and sequentially inputting the time-series frame images corresponding to the sports training data into the three-layer branch processing model for localization prediction analysis from three dimensions to obtain importance assessment data of the corresponding image regions within the predicted time-series frame images, includes: S41: Obtain the key motion dataset corresponding to each type of motion within the motion management space, analyze and process the obtained key motion dataset based on deep learning algorithms, and construct a three-layer branch processing model, which includes a character branch processing model, an action branch processing model, and a region branch processing model. S411: In the process of constructing the character branching model, action feature recognition and event feature recognition are performed on the corresponding key motion datasets, and the corresponding feature recognition results are compared. The key motion datasets for comparison are divided into training set, validation set and test set. The training set is used for model parameter learning, the validation set is used to adjust the model's hyperparameters and monitor the training process to prevent overfitting, and the test set is used to evaluate the final performance of the model. The obtained training set is trained and adjusted based on deep learning algorithms to construct the character branching model. The constructed character branching model is validated and tested based on the validation set and test set, and the corresponding character branching model is output. S412: In the process of constructing the action branch processing model, action feature recognition and region feature recognition are performed on the corresponding key motion datasets, and the corresponding feature recognition results are compared. The key motion datasets for comparison are divided into training set, validation set and test set. The training set is used for model parameter learning, the validation set is used to adjust the model's hyperparameters and monitor the training process to prevent overfitting, and the test set is used to evaluate the final performance of the model. The obtained training set is trained and adjusted based on the deep learning algorithm to construct the action branch processing model. The constructed action branch processing model is validated and tested based on the validation set and test set, and the corresponding action branch processing model is output. S413: In the process of constructing the region branch processing model, region feature recognition and event feature recognition are performed on the corresponding key motion datasets, and the corresponding feature recognition results are compared. The key motion datasets for comparison are divided into training set, validation set and test set. The training set is used for model parameter learning, the validation set is used to adjust the model's hyperparameters and monitor the training process to prevent overfitting, and the test set is used to evaluate the final performance of the model. The obtained training set is trained and adjusted based on deep learning algorithms to construct the region branch processing model. The constructed region branch processing model is validated and tested based on the validation set and test set, and the corresponding region branch processing model is output. S42: Input the corresponding sports video data within the motion management space into the three-layer branch processing model, set the corresponding frame rate for the sports video data, divide the sports video data into multiple time-series frame images according to the set frame rate, and perform localization prediction analysis based on the corresponding three-dimensional dimensions of people, actions and regions within the time-series frame images; obtain the predicted position information of the corresponding image region within the predicted time-series frame image where key scenes may exist; S43: Assess the importance of key scenes at corresponding predicted locations within the corresponding image region. Set an importance membership matrix based on the correlation between the corresponding person, action, and region and the corresponding sports rule information within the sports data. Assess the importance of the predicted locations of key scenes obtained in the three-layer branch processing model based on the importance membership matrix. Obtain importance assessment data for each dimension.

[0023] In the specific implementation process, the step S5, which involves processing the importance assessment data of the corresponding image region within the predicted time-series frame image corresponding to the sports video data in the motion management space for resource allocation, and obtaining the resource allocation result corresponding to each pixel point within the time-series frame image, includes: S51: Obtain the importance assessment data of each image region in the prediction time frame image corresponding to the sports video data in three dimensions, and perform weighted integration based on the importance assessment data of each image region in three dimensions to obtain the comprehensive assessment data of the corresponding image region. S511: Obtain the importance assessment data of the key motion dataset and the corresponding key scene prediction location information in the three-layer branch processing model in step 4. S512: Based on the constructed three-layer branch processing model, analyze and process the predicted location information of key scenes that appear at the same time to obtain the correlation data between different characters and actions, characters and regions, and actions and regions within the three-layer branch processing model. S513: Construct a character action-related attention model based on the correlation data between different characters and actions, obtain the correlation data between different characters and different actions and the weight coefficients of the importance assessment data between different characters and actions, and determine the correlation data of the corresponding character when recognizing the corresponding action through the corresponding weight coefficients, thereby dynamically adjusting the corresponding importance assessment data. A person-region related attention model is constructed based on the correlation data between different people and regions. The correlation data between different people and different regions and the weight coefficients of the importance assessment data between different people and regions are obtained. The correlation data when the corresponding person appears in the corresponding region is judged by the corresponding weight coefficients, thereby dynamically adjusting the corresponding importance assessment data. A character-action related attention model is constructed based on the correlation data between different characters and actions. The correlation data between different characters and different actions, as well as the weight coefficients of the importance assessment data between different characters and actions, are obtained. The correlation data when the corresponding action appears in the corresponding area is judged by the corresponding weight coefficients, thereby dynamically adjusting the corresponding importance assessment data. S514: Input the importance evaluation data of the three-layer branch processing model into the corresponding relevant attention model for analysis and processing, and obtain the weight coefficients of the importance evaluation data corresponding to each prediction position information; S515: The corresponding image region within the corresponding predicted time frame image is weighted according to the weight coefficients of the three dimensions and the corresponding importance evaluation data to obtain the comprehensive evaluation data within the corresponding image region. S52: Map the comprehensive evaluation data obtained from each image region sequentially into the predicted time frame image, perform visualization processing based on the mapping results, perform mean processing on the visualization processing results, integrate and divide the resource management data based on the mean processing results, and obtain unit resource data; S521: Map the comprehensive evaluation data into the predicted time frame image, perform standardization processing based on each comprehensive evaluation data, and perform visualization processing based on the standardization processing results; S522: Perform visualization mean processing on the visualization processing results to obtain the mean data corresponding to the predicted time frame image; S523: Divide the corresponding elastic resource data in the resource management data into units based on the corresponding mean data, and obtain unit resource data so that the unit resource data conforms to the allocation and adjustment of the visualization processing results; S53: The obtained unit resource data is divided into preliminary linear resource data and real-time elastic resource data. The preliminary linear resource data consists of multiple unit resource data set for resource scheduling and allocation within the corresponding image region based on the mapping visualization processing results of each image region in the prediction time-series frame image. The real-time elastic resource data consists of multiple unit resource data set for resource scheduling and allocation within each image region based on the mapping visualization processing results of each image region in the prediction time-series frame image. The preliminary linear resource data is processed for resource allocation based on the visualization processing results of the prediction time-series frame image to obtain the resource allocation results corresponding to each image region in the prediction time-series frame image. As shown in steps S51 to S53 above, this invention sets up relevant attention models between different characters and actions, characters and regions, and actions and regions. These relevant attention models dynamically adjust the importance assessment data of different characters, actions, and regions according to corresponding weighting coefficients, thereby improving the accuracy of sports video data analysis. Resource adjustments are made based on the corresponding resource allocation results within the predicted time-series frame image, which improves efficiency in the positioning analysis process to some extent. Furthermore, by setting up pre-emptive elastic resource data to analyze and process the predicted time-series frame image, and adjusting the real-time elastic resource data based on the analysis results, resource scheduling waste caused by inaccurate predicted time-series frame images is avoided to some extent, and the analysis efficiency under prediction error conditions is also improved.

[0024] In the specific implementation process, the step S6 of dividing and matching the temporal frame image corresponding to the sports video data from three dimensions according to the resource allocation result at the corresponding pixel point, and obtaining the hierarchical division result of the three dimensions at the corresponding image region includes: S61: Obtain the resource allocation results corresponding to the sports video data, perform resource allocation processing on the preliminary missile resource data according to the obtained resource allocation results, and compare and analyze the image region corresponding to the preliminary missile resource data with the predicted time frame image to obtain similarity data. S611: Pre-allocate the missile-like resource data to the corresponding image region within the predicted time-series frame image, and obtain the real-time time-series frame image corresponding to the predicted time-series frame image; S612: By analyzing and processing the data in the corresponding image region within the real-time time-series frame image using the preliminary missile-based resource data, similarity data with the predicted key data is obtained; S62: Allocate real-time elastic resource data based on similarity data in real time, and obtain the resource allocation results of the real-time time sequence frame images corresponding to the sports video data; S621: Set the corresponding similarity data within the corresponding image area, set the similarity comparison interval, and if the similarity data belongs to the similarity comparison interval, then adjust the real-time elastic resource data to the image area corresponding to the first-time elastic resource data. S622: If the similarity data does not belong to the similarity comparison interval, the real-time elastic resource data will be distributed to other image areas, and the real-time resource allocation of the real-time sequence frame image will be adjusted in real time. S63: Perform edge processing on the time-series frame image corresponding to the sports video data based on the resource allocation results at the corresponding image region to obtain the adjusted image region; S64: Divide and match each image region in the time frame image from three dimensions to obtain the hierarchical division result of the corresponding image region in the three dimensions. The hierarchical division result is the division result of the corresponding dimension of each image region in the time frame image.

[0025] In the specific implementation process, the step S7 of sequentially integrating the obtained hierarchical division results according to the three-layer branch processing model to obtain the key scene time node positioning information corresponding to the sports video data includes: S71: Obtain the hierarchical division results of each image region in the time frame image in three dimensions; S72: Obtain the key motion dataset corresponding to the three-layer branch processing model, and set up a key scene matching and sorting library for the key motion dataset according to the combination results of the three-layer branch processing model. The key scene matching and sorting library is obtained by sorting the key scene occurrence rate corresponding to the combination of the corresponding partition results in the three-layer branch processing model. S73: Obtain the corresponding key scene matching and sorting library based on the hierarchical division results of each image region, and perform matching analysis between the scene information in the image region and the key scene matching and sorting library in turn to determine whether there is a key scene. If there is, obtain the key scene type. S74: Set sliding window parameter information according to key scene type, and perform retrieval and analysis on the corresponding image area in each time frame image in the sports video data according to the sliding window parameter information to obtain the key scene time node positioning information corresponding to the sports video data; As shown in steps S71 to S74 above, in the process of setting the corresponding sliding window parameter information through the key scene type, the window size parameter and window step length parameter are set according to the average time span and frequency of occurrence of the corresponding key scene type, respectively. Based on the set sliding window parameter information, the corresponding image region in the corresponding time frame image in the sports video data is retrieved and analyzed, thereby improving the efficiency of the key scene time node positioning analysis process.

[0026] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for locating key time nodes in sports videos, characterized in that, Includes the following steps: Step S1: Set up a sports video management platform, obtain relevant sports data through the sports video management platform, and set up corresponding sports management spaces according to the type of sports data; Step S2: Obtain sports video data, determine the corresponding type based on the sports video data, and input the sports video data into the sports management space according to the corresponding type; Step S3: Divide the sports video data in the corresponding sports management space into key sports areas, key personnel, and key sports skills, and set up corresponding key sports datasets according to the division results; Step S4: Set up a three-layer branch processing model for the obtained key motion dataset according to the division criteria; Acquire sports training data, divide the acquired sports training data into multiple time-series frame images according to time sequence, and input the time-series frame images corresponding to the sports training data into a three-layer branch processing model for localization prediction analysis from three dimensions to obtain importance assessment data of the corresponding image regions in the predicted time-series frame images; Step S5: Perform resource allocation processing on the importance assessment data of the corresponding image region in the predicted time-series frame image corresponding to the sports video data in the motion management space, and obtain the resource allocation result corresponding to each pixel in the time-series frame image; Step S6: Based on the resource allocation results at the corresponding pixels, the time-series frame image corresponding to the sports video data is divided and matched from three dimensions to obtain the hierarchical division results of the corresponding image region in three dimensions; Step S7: The obtained hierarchical division results are sequentially integrated according to the three-layer branch processing model to obtain the key scene time node positioning information corresponding to the sports video data.

2. The method for locating key time nodes in sports videos according to claim 1, characterized in that, The process of obtaining relevant sports data through the sports video management platform includes: A sports video management platform is set up, which includes a data input terminal, a data processing terminal, and a data acquisition terminal. The data entry terminal is used to enter sports data from the sports video management platform. The sports data includes sports type data, sports venue data, and sports rule data. The data processing terminal is used to classify and process the obtained sports data, obtain the sports type corresponding to the sports data, and set up corresponding sports management space in the sports video management platform according to the sports type.

3. The method for locating key time nodes in sports videos according to claim 2, characterized in that, The process of acquiring sports video data includes: The data acquisition terminal is used to acquire sports video data and input the acquired sports video data into the data processing terminal; The data processing terminal inputs the obtained sports video data into the corresponding sports management space.

4. The method for locating key time nodes in sports videos according to claim 3, characterized in that, The process of dividing sports video data within the corresponding sports management space into sports regions, key personnel, and sports skills includes: Acquire historical sports video data within the sports management space, and divide the acquired historical sports video data according to the corresponding sports data; Based on sports field data within sports data, historical sports video data is divided into sports regions to obtain key score-related regions, key rule restriction regions, and transition regions within the corresponding sports regions in the sports management space. Based on the sports rules data within the sports data, key personnel and key sports skills are segmented from historical sports video data. This allows for the acquisition of sports responsibility information and historical sports information for the corresponding key personnel within the sports management space; and the acquisition of skill function information and skill characteristic information for the corresponding sports skills within the sports management space for each sports type. Based on the multi-dimensional segmentation results and the key scenes in historical sports video data, corresponding key sports datasets were set up.

5. The method for locating key time nodes in sports videos according to claim 4, characterized in that, The process of setting up a three-layer branch processing model based on the obtained motion dataset and the division criteria includes: Obtain the key motion datasets corresponding to each type of motion within the motion management space. Analyze and process the obtained key motion datasets using deep learning algorithms to construct a three-layer branch processing model, which includes a character branch processing model, an action branch processing model, and a region branch processing model. The corresponding sports video data within the motion management space is input into a three-layer branch processing model. Based on the three dimensions of people, actions, and regions in the time-series frame images, the model performs localization prediction analysis to obtain the predicted location information of the corresponding image regions in the predicted time-series frame images where key scenes may exist, and obtains importance assessment data for the three dimensions.

6. The method for locating key time nodes in sports videos according to claim 5, characterized in that, The process of obtaining the resource allocation results corresponding to each pixel point in the time frame image includes: Obtain the prediction location information of the three-layer branch processing model and the importance evaluation data of the three dimensions of each image region in the prediction time frame image corresponding to the sports video data; The importance assessment data of each predicted position information in each key motion dataset in the three-layer branch processing model are fused and analyzed to obtain the correlation data between different characters and actions, characters and regions, and actions and regions. Based on the obtained correlation data, a correlation attention model based on the three-layer branch processing model is constructed. The importance assessment data of the three-layer branch processing model are input into the corresponding relevant attention model for analysis and processing to obtain the weight coefficients of the corresponding importance assessment data. The comprehensive evaluation data for the corresponding image region is obtained by weighting and integrating the importance assessment data of the three dimensions of each image region and the corresponding weight coefficients. The comprehensive evaluation data obtained from each image region is sequentially mapped into the predicted time frame image. The mapping results are then visualized, and the visualization results are averaged. Based on the averaged results, the resource management data is integrated and divided to obtain unit resource data. The obtained unit resource data is divided into pre-emptive elastic resource data and real-time elastic resource data. Based on the visualization processing results of the predicted time-series frame image, the pre-emptive elastic resource data is processed for resource allocation to obtain the resource allocation results corresponding to each image region of the predicted time-series frame image.

7. The method for locating key time nodes in sports videos according to claim 6, characterized in that, The process of dividing and matching the time-series frame images corresponding to the sports video data from three dimensions based on the resource allocation results at the corresponding image regions includes: Obtain the resource allocation results corresponding to the sports video data. Based on the obtained resource allocation results, perform resource allocation processing on the preliminary missile resource data. Compare and analyze the image regions corresponding to the preliminary missile resource data with the predicted time-series frame images to obtain similarity data. Based on similarity data, real-time elastic resource data is allocated in real time to obtain the resource allocation results of real-time time-series frame images corresponding to sports video data; Based on the resource allocation results at the corresponding image regions, edge processing is performed on the time-series frame images corresponding to the sports video data to obtain the adjusted image regions. Each image region within the time-series frame image is divided and matched from three dimensions to obtain the hierarchical division results of the corresponding image region in the three dimensions.

8. The method for locating key time nodes in sports videos according to claim 7, characterized in that, The process of obtaining the key scene time node positioning information corresponding to the relevant sports video data includes: Obtain the hierarchical division results of each image region within a time-series frame image in three dimensions; Obtain the key motion dataset corresponding to the three-layer branch processing model, and set up a key scene matching and sorting library for the key motion dataset according to the combination results of the three-layer branch processing model. Based on the hierarchical division results corresponding to each image region, the corresponding key scene matching and sorting library is obtained. The scene information in the image region is matched and analyzed with the key scene matching and sorting library in turn to determine whether there is a key scene. If there is, the key scene type is obtained. Based on the key scene type, set the sliding window parameter information, and perform retrieval and analysis on the corresponding image regions in each time frame image of the sports video data according to the sliding window parameter information to obtain the key scene time node positioning information corresponding to the sports video data.

Images