阅读说明：本技术 基于卷积神经网络的橄榄球辅助裁判系统 (Football auxiliary referee system based on convolutional neural network ) 是由 刘红彬 于 2021-06-30 设计创作，主要内容包括：本发明涉及一种基于卷积神经网络的橄榄球辅助裁判系统,包括图像收集模块、红外检测模块和数据集中处理模块,所述图像收集模块与红外检测模块控制连接,红外检测模块通过发出的红外激光被遮挡而发出电信号给图像收集模块,图像收集模块中的摄影机拍摄视频上传至数据集中处理模块,通过数据集中处理模块中设置的智能综合分析终端进行分析判断,得到辅助裁判结果。通过卷积神经网络的多个集合分类进行计算,保证裁判计分结果的准确性,使得裁判更科学,减少争议。(The invention relates to a football auxiliary judgment system based on a convolutional neural network, which comprises an image collection module, an infrared detection module and a data centralized processing module, wherein the image collection module is in control connection with the infrared detection module, the infrared detection module sends an electric signal to the image collection module through shielding of sent infrared laser, a camera in the image collection module shoots a video and uploads the video to the data centralized processing module, and an intelligent comprehensive analysis terminal arranged in the data centralized processing module is used for analyzing and judging to obtain an auxiliary judgment result. The accuracy of the scoring result of the referee is ensured by calculating through a plurality of sets of the convolutional neural network in a classified manner, so that the referee is more scientific, and disputes are reduced.)

1. A football auxiliary referee system based on a convolutional neural network is characterized in that: including image collection module, infrared detection module and data centralized processing module, image collection module and infrared detection module control connection, infrared detection module is sheltered from and sends the signal of telecommunication for image collection module through the infrared laser that sends, and the video is shot to the data centralized processing module on the camera among the image collection module, carries out analysis and judgement through the intelligent comprehensive analysis terminal that sets up in the data centralized processing module, obtains supplementary judge result.

2. The convolutional neural network-based football-aided referee system of claim 1, wherein: the infrared detection module is provided with infrared rays at the boundary of each different area, the two sides of the court in the width direction are provided with an infrared laser and a corresponding receiving detector, the infrared laser is used for emitting infrared rays on the boundary of each different area, when a team member exceeds the boundary, the infrared rays are shielded, the corresponding receiving detector receives information that the infrared ray reflection is incomplete, and then feedback information of people passing the lines is sent to the image collecting module.

3. A football auxiliary referee method based on a convolutional neural network is characterized in that: the data centralized processing module processes data based on a convolutional neural network, and the specific method comprises the following steps: the method comprises the steps of practicing various over-boundary conditions, collecting image data of each condition through an image collection module, inputting an image data set of the image collection module into a first convolution layer of a convolution neural network in a data set centralized processing module, and processing data in the first convolution layer through mapping; the processed image data is then output mapped to a second convolutional layer of the convolutional neural network where the data is processed by a second portion of the weighting factors.

4. The football assisted referee method based on convolutional neural network as defined in claim 3, wherein: the first convolution layer is an initial processing system, the initial processing system receives input image data and corresponding time data, and a storage weight factor set is attached to the initial processing system; the second convolutional layer is a compute hub system for processing the data set of the first convolutional layer.

5. The football assisted referee method based on convolutional neural network as defined in claim 4, wherein: and cutting the data set into a plurality of data sets, and classifying each data set according to the situation of the competition field.

6. The football assisted referee method based on convolutional neural network as defined in claim 5, wherein: and (4) defining a judge division point through the weight factor set, and obtaining final judge score result data when more than 90% of calculation workload is included.

Technical Field

The invention relates to a football auxiliary referee system based on a convolutional neural network, and belongs to the technical field of computer image analysis.

Background

The football is a ball with more complex rules, and the confrontation is usually controversial because the boundary on the field is more, the distance between adjacent areas is small, and the movement process is instant. The slight difference is difficult to judge, common video equipment can only judge by sending to a judge, subjective judgment components may exist, uncontrollable conditions exist if different conditions are judged manually, and correct standard reference is not provided.

Disclosure of Invention

The purpose of the invention is as follows: in view of the above existing problems and disadvantages, the present invention provides a football-assisted referee system based on a convolutional neural network, which analyzes and judges a referee result through the convolutional neural network, so that the final result is more accurate.

The technical scheme is as follows: in order to realize the purpose, the invention adopts the following technical scheme: the utility model provides a football auxiliary judge system based on convolutional neural network, includes image collection module, infrared detection module and data centralized processing module, image collection module and infrared detection module control connection, infrared detection module is sheltered from and send the signal of telecommunication for image collection module through the infrared laser that sends, and video shooting video among the image collection module is uploaded to data centralized processing module, carries out analysis and judgment through the intelligent comprehensive analysis terminal that sets up among the data centralized processing module, obtains auxiliary judge result.

Furthermore, the infrared detection module is provided with infrared rays at the boundary of each different area, two sides of the court in the width direction are provided with an infrared laser and a corresponding receiving detector, the infrared laser is used for emitting infrared rays on the boundary of each different area, when a team member exceeds the boundary, the infrared rays are shielded, and the corresponding receiving detector receives information that the infrared reflection is incomplete, and then sends feedback information of people passing the lines to the image collection module.

Further, the data centralized processing module processes data based on a convolutional neural network, and the specific method comprises the following steps: the method comprises the steps of practicing various over-boundary conditions, collecting image data of each condition through an image collection module, inputting an image data set of the image collection module into a first convolution layer of a convolution neural network in a data set centralized processing module, and processing data in the first convolution layer through mapping; the processed image data is then output mapped to a second convolutional layer of the convolutional neural network where the data is processed by a second portion of the weighting factors.

Further, the first convolution layer is an initial processing system, the initial processing system receives input image data and corresponding time data, and a storage weight factor set is attached to the initial processing system; the second convolutional layer is a compute hub system for processing the data set of the first convolutional layer.

Furthermore, the data set is cut into a plurality of data sets, and each data set is classified according to the situation of the match field.

Further, a judgment dividing point is defined through the weight factor set, and when the calculation workload is more than 90%, final judgment score result data is obtained.

Has the advantages that: compared with the prior art, the invention has the following advantages: the football auxiliary judgment system is a novel football auxiliary judgment system based on the convolutional neural network, and the final judgment result is more accurate by combining the case and the convolutional neural network.

Detailed Description

The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, as various equivalent modifications of the invention will occur to those skilled in the art upon reading the present disclosure and fall within the scope of the appended claims.

This football auxiliary judge system based on convolutional neural network includes image collection module, infrared detection module and data centralized processing module, because the dispute area of football is more, infrared detection module needs to be provided with the infrared ray at every different boundary line department, court width direction's both sides are equipped with infrared laser and corresponding receiving detector, infrared laser is used for transmitting the infrared ray on the boundary line of every different districts, when the team member surpassed the boundary line, the infrared ray is sheltered from, the incomplete information of infrared reflection is received to corresponding receiving detector, then send someone feedback information of crossing the line to image collection module. The image collection module is in control connection with the infrared detection module, the infrared detection module sends an electric signal to the image collection module by shielding the sent infrared laser, a camera in the image collection module shoots a video and uploads the video to the data centralized processing module, and the video is analyzed and judged through an intelligent comprehensive analysis terminal arranged in the data centralized processing module to obtain an auxiliary judgment result. The infrared laser is connected with the image collecting module, so that the image collecting module obtains corresponding conditions through the training of the convolutional neural network, and the analysis and judgment are carried out to judge how to score and judge.

The data centralized processing module processes data based on a convolutional neural network, the convolutional neural network can output data after one-dimensional, two-dimensional or three-dimensional processing, and a large amount of calculation is performed on the order of tens of thousands of times per second, so that a more accurate result is obtained. The method comprises the steps of practicing various over-boundary conditions, collecting image data of each condition through an image collection module, inputting an image data set of the image collection module into a first convolution layer of a convolution neural network in a data set centralized processing module, and processing data in the first convolution layer through mapping; the processed image data is then output mapped to a second convolutional layer of the convolutional neural network where the data is processed by a second portion of the weighting factors. Because the operation of the convolutional neural network calculation is many, the running cost of the equipment is more, and the cost is easy to be increased, so that the resource is more saved when the data is processed in a data set mode, and the cost is reduced.

The first convolution layer is an initial processing system that receives input image data and corresponding time data, while appending a set of storage weight factors. The second convolutional layer is a compute hub system for processing the data set of the first convolutional layer. And cutting the data set into a plurality of data sets, and classifying each data set according to the situation of the competition field. The match field situation can be more refined through classification, and therefore a more accurate referee scoring result is obtained. And (4) defining a judge division point through the weight factor set, and obtaining final judge score result data when more than 90% of calculation workload is included. Namely, when the calculation workload meets the accuracy requirement, the final accurate scoring judgment result can be output.

The positions of the modules are set, various competition fields are simulated firstly, and various accelerators are used for accelerating training to adapt to fields in various different areas. The CNN engine 30 may be used in the image acquisition module to connect to the convolutional neural network system bus of the data centralized processing module, and to write the collected data into the memory. The convolution engine 32 can make the convolution layer function by reading the address in the controller through the mapping relation of the image, and the image buffer 31 reads the designated address buffered by the controller 60 through the input port from the output port of the convolution engine 32 through the sub-sampler 34. The switch 35 allows image data to be provided by the controller 36 first. The information of the convolution engine 32 or the generated information of the sub-sampler 34 is then read back to the image buffer 31. The initial ROI image is first loaded into the image buffer 31, and the initial ROI image data during motion is usually extended from the address 0 × 00 of the image buffer with an offset of 0. The cache is addressed in a two-dimensional way, also with an offset of the address line, after the features are extracted in the first convolution layer and the sub-sampling layer, a plurality of mappings are generated, 5 at this time, the first layer mapping 0 …, the first layer mapping 4, are written into the address line. Convolutional layer 2 generates 10 maps from the 5 maps generated by convolutional layer 1, no sub-sampling is performed, the new map generated by convolutional layer 2 may overwrite the area of the image buffer, the second layer maps 1 … 3 are shifted continuously in the buffer with respect to each other and map 0; map 5 … 7 is also continuously shifted with respect to map 4. The second layer maps 8 and 9 are written into the address space of the first layer map. If the classification extraction is complete, any generated vectors may be written to controller 39.

In extracting features, the weights are convolution kernel values, while in feature classification of images, the weights are full connection layer neurons or connection weight values, there are 5 input mappings and the kernel size is 5 × 5, (75) weights just before the mapping. 2048 elements are currently set, the output vector has 15 elements, the weight cache size is 1024, corresponding weights are calculated and read, then next 1024 weights are read, calculation is performed again, and 1024 weights are taken out. The calculation is not stopped.

The first component 105 of the weight factors of 5%, 15%, 25%, 35%, 45%, 55%, 65%, 75%, 85% and 95% and the second component 110 of the remaining weight factors can be adopted, and the accuracy of calculation can also be improved by changing the division points, so that the final obtained referee scoring result is more in line with the requirement.