Marathon race AI eagle eye recognition system based on image processing

Abstract

The invention discloses a marathon race AI eagle eye recognition system based on image processing, which relates to the technical field of image recognition and comprises a video extraction unit, an image processing unit, a number cloth detection unit, a number recognition unit, a re-recognition unit and a data management unit. According to the invention, intelligent screening is carried out on the video data through an automatic frame extraction technology, only clear frames with variable contents are reserved, redundant data interference is avoided, the monitoring efficiency is remarkably improved, and the risk of manual misjudgment is reduced; meanwhile, environment influence index calculation is combined, environment interference is dynamically evaluated through a light intensity and visibility sensor, image quality is automatically enhanced when illumination is too strong or visibility is too low, and the problem that the number cloth OCR recognition failure rate is high in complex scenes such as backlight and rain and fog is effectively solved; moreover, the feature vector extraction and re-identification technology constructs a feature similarity index by capturing multi-dimensional features of clothing, gaits, accessories and the like of players, and can still accurately track the identity of the players when the number cloth is shielded or damaged.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to an AI eagle-eye recognition system for marathon tracks based on image processing. Background Technology

[0002] With the booming development of mass participation sports events such as marathons, rampant cheating and irregularities such as runners using substitutes, carrying multiple chips, and taking shortcuts have emerged. Therefore, regulating the conduct and discipline of large-scale mass participation events has become a pressing issue. The Vesay team hopes that, with the assistance of artificial intelligence, AI vision and RFID components deployed at timing points along the racecourse will be used to automatically track and extract image information of each runner as they pass through the timing points. This will provide verifiable evidence for processing race results, ensuring fairness and impartiality, and contributing to the construction of a trustworthy participation system.

[0003] Traditional marathon races are often monitored manually. However, manual monitoring faces the dual challenges of low efficiency and subjective misjudgment. Referees need to review massive amounts of video frame by frame, and long-term high-intensity work can easily lead to a loss of focus. Especially in complex weather conditions such as backlighting, rain, and snow, when bib numbers are blurred or runners' postures overlap, the accuracy of manual recognition drops sharply, and it is impossible to respond to abnormal behavior in real time. On the other hand, traditional bib number OCR recognition systems have difficulty extracting effective features in strong backlighting scenes due to uneven lighting causing overexposure or underexposure of images, or when runners' bib numbers are obscured by objects they are holding. This results in a high failure rate and seriously restricts the reliability and fairness of intelligent management of marathon events.

[0004] To address the aforementioned technical deficiencies, a solution is proposed. Summary of the Invention

[0005] The purpose of this invention is to address the dual challenges of low efficiency and subjective misjudgment faced by manual monitoring. Referees need to review massive amounts of video frame by frame, and the long-term high-intensity work can easily lead to a loss of focus. Especially in complex weather conditions such as backlighting, rain, and snow, when bib numbers are blurred or runners' postures overlap, the accuracy of manual recognition drops sharply, and it is impossible to respond to abnormal behavior in real time. Traditional bib number OCR recognition systems also suffer from high failure rates in strong backlighting scenes due to uneven lighting causing overexposure or underexposure of images, or when runners' bib numbers are obscured by objects. These issues severely restrict the reliability and fairness of intelligent management of marathon events.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an AI eagle-eye recognition system for marathon tracks based on image processing, comprising: a video extraction unit, an image processing unit, a bib number detection unit, a bib number recognition unit, and a data management unit;

[0007] The video extraction unit is used to automatically extract video data from cameras at specified timing points on the marathon course;

[0008] The image processing unit is used to intelligently and automatically extract frames from video data on the marathon track, obtain image quality data from the video library, and calculate an image selection index through analysis to retain clear frames with varying content.

[0009] The bib number detection unit is used to identify the location of bibs in the image using a target detection algorithm;

[0010] The bib number recognition unit is used to collect environmental data on the marathon course using a sensor array, analyze and calculate the environmental impact index, and use OCR technology to recognize the numbers and letters on the bib number.

[0011] The re-identification unit is used to identify and record the feature data of the athletes, and calculate the feature similarity index by analyzing the feature data. When the bib number identification is affected, the athlete's features are identified.

[0012] The data management unit is used to store the recognition results into the database, and supports multi-dimensional querying of marathon athletes' race information by number, time, and location.

[0013] Furthermore, the image quality data includes the RGB values ​​of the image and the image gradient, the environmental data includes the ambient light intensity and visibility, and the feature data includes clothing features, gait features, and accessory features.

[0014] Furthermore, the calculation process for the image selection index of marathon track images is as follows:

[0015] S11. Obtain the RGB values ​​and gradient data of the image and perform analysis and calculation;

[0016] S12. Calculate the image selection index (CSIR) for marathon track images using the following formula:

[0017]

[0018] Where RGB represents the RGB values ​​of the marathon track image. e The RGB values ​​are the preset standard image values, M is the length of the marathon track image, N is the width of the marathon track image, and G is the length of the marathon track image. x (m,n) represents the gradient in the x-direction of the marathon track image, G y (m,n) represents the gradient in the y-direction of the marathon track image;

[0019] S13. Obtain the preset image filtering threshold CSIR. thA comparative analysis was conducted with the image screening index CSIR. When CSIR > CSIR th If CSIR ≤ CSIR, it indicates that the image of the marathon track in that frame is clear but has undergone significant changes and needs to be retained. th If the image of the marathon track is unclear or its content has not changed significantly, it needs to be deleted.

[0020] Furthermore, the calculation process for the environmental impact index of a marathon course is as follows:

[0021] S21. Acquire light intensity and visibility data and perform analysis and calculation;

[0022] S22. Calculate the environmental impact index E of the marathon course using the following formula:

[0023]

[0024] Where L represents the light intensity of the marathon track environment, L max L is the upper limit of the preset standard light intensity. min V represents the lower limit of the preset standard light intensity, and V is the visibility of the marathon track environment. min V is the preset lower limit of standard visibility. max , where a is the preset upper limit of standard visibility, a is the preset weighting coefficient of light intensity, b is the preset weighting coefficient of visibility, and a+b=1. The environmental impact index is used to reflect the degree of influence of the environment on the quality of the captured image.

[0025] S23. Obtain the preset environmental impact threshold E th Comparative analysis with the environmental impact index E, when E>E th When E > E, it indicates that the marathon course environment has a high impact on the quality of the captured image, and the bib number image will be automatically enhanced using image enhancement technology. th If the image quality is good, it means that the marathon track environment has a low impact on the quality of the captured image, and no image enhancement is needed.

[0026] Furthermore, the calculation process for the feature similarity index of marathon runners is as follows:

[0027] S31. Obtain initial feature data of marathon runners and match it with runner numbers. At the same time, obtain clothing feature, gait feature and accessory feature data and extract feature vectors, and then perform analysis and calculation.

[0028] S32. Calculate the characteristic similarity index S of marathon runners according to the following formula:

[0029]

[0030] Where I represents the total number of clothing features worn by a marathon runner, J represents the total number of gait features worn by a marathon runner, K represents the total number of accessory features worn by a marathon runner, and P represents the total number of clothing features worn by a marathon runner. i Let Q be the feature vector value of the i-th clothing feature of a runner on the marathon course. j Let R be the feature vector value of the j-th gait feature of a runner on a marathon course. k Let P be the feature vector value of the k-th accessory feature of a runner on the marathon course. e Q is the preset initial clothing feature vector value for marathon runners. e R represents the preset initial gait feature vector values ​​of a marathon runner. e The preset initial accessory feature vector values ​​for marathon runners are: α is the preset clothing weight coefficient, β is the preset gait weight coefficient, γ is the preset accessory weight coefficient, and α+β+γ=1;

[0031] S33. Obtain the preset feature similarity threshold S th Comparative analysis with the feature similarity index S, when S>S th If S ≤ S, it indicates that the runners on the marathon course have a high degree of similarity in initial characteristics to the marathon runners, and it is determined that they are the same person. The runners on the marathon course are then matched with corresponding runner numbers. th If the initial characteristics of the runners on the marathon course are not similar to those of the marathon runners, it indicates that the two are not the same person and the runner numbers do not match.

[0032] Furthermore, the sensor group includes a light intensity sensor and a visibility sensor.

[0033] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0034] This AI-powered marathon track recognition system, based on image processing, intelligently filters video data using automatic frame extraction technology, retaining only clear frames with varying content to avoid redundant data interference. This significantly improves monitoring efficiency and reduces the risk of human error. Simultaneously, it incorporates environmental impact index calculations, dynamically assessing environmental interference using light intensity and visibility sensors. When lighting is too strong or visibility is too low, it automatically enhances image quality, effectively addressing the high failure rate of bib number OCR recognition in complex scenarios such as backlighting, rain, and fog. Furthermore, feature vector extraction and re-identification technology captures multi-dimensional features such as runners' clothing, gait, and accessories to construct a feature similarity index. This allows for accurate tracking of runners even when bib numbers are obscured or damaged, ensuring fair competition. Attached Figure Description

[0035] Figure 1 A schematic diagram of the system flow of the present invention is shown. Detailed Implementation

[0036] 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.

[0037] Example:

[0038] like Figure 1 As shown, the marathon track AI Eagle Eye Recognition System based on image processing firstly extracts video data from cameras at designated timing points on the marathon track through the video extraction unit. Then, the image processing unit intelligently and automatically extracts frames from the video data on the marathon track and obtains image quality data from the video library. The image quality data includes the RGB values ​​and image gradients of the image. The system also calculates an image filtering index to retain clear frames with varying content, thus avoiding video data redundancy.

[0039] The calculation process for the image selection index of marathon track images is as follows:

[0040] S11. Obtain the RGB values ​​and gradient data of the image and perform analysis and calculation;

[0041] S12. Calculate the image selection index (CSIR) for marathon track images using the following formula:

[0042]

[0043] Where RGB represents the RGB values ​​of the marathon track image. e The RGB values ​​are the preset standard image values, M is the length of the marathon track image, N is the width of the marathon track image, and G is the length of the marathon track image. x (m,n) represents the gradient in the x-direction of the marathon track image, G y (m,n) represents the gradient in the y-direction of the marathon track image;

[0044] S13. Obtain the preset image filtering threshold CSIR. th A comparative analysis was conducted with the image screening index CSIR. When CSIR > CSIR th If CSIR ≤ CSIR, it indicates that the image of the marathon track in that frame is clear but has undergone significant changes and needs to be retained. th If the image of the marathon track is unclear or its content has not changed significantly, it needs to be deleted.

[0045] Then, the bib detection unit uses a target detection algorithm (YOLO target detection algorithm) to identify the position of the bibs in the image. The number recognition unit uses a sensor array to collect environmental data on the marathon course. The environmental data includes ambient light intensity and visibility. The sensor array includes a light intensity sensor (a photoresistor sensor that detects light intensity; its principle is that the resistance value changes with the light intensity, the resistance decreases when the light intensity increases, and the resistance increases when the light intensity decreases) and a visibility sensor (a light transmission sensor that detects visibility; its principle is that the visibility is estimated by measuring the attenuation of the light beam as it passes through the air). The environmental impact index is calculated through analysis, and OCR technology is used to identify the numbers and letters on the bib.

[0046] The calculation process for the environmental impact index of a marathon course is as follows:

[0047] S21. Acquire light intensity and visibility data and perform analysis and calculation;

[0048] S22. Calculate the environmental impact index E of the marathon course using the following formula:

[0049]

[0050] Where L represents the light intensity of the marathon track environment, L max L is the upper limit of the preset standard light intensity. min V represents the lower limit of the preset standard light intensity, and V is the visibility of the marathon track environment. min V is the preset lower limit of standard visibility. max is the preset upper limit of standard visibility, 'a' is the preset weighting coefficient of light intensity, and 'b' is the preset weighting coefficient of visibility, and a+b=1. The environmental impact index is used to reflect the degree of influence of the environment on the quality of the captured image.

[0051] S23. Obtain the preset environmental impact threshold E th Comparative analysis with the environmental impact index E, when E>E th When E > E, it indicates that the marathon course environment has a high impact on the quality of the captured image, and the bib number image will be automatically enhanced using image enhancement technology. th If the image quality is good, it means that the marathon track environment has a low impact on the quality of the captured image, and no image enhancement is needed.

[0052] Then, the re-identification unit identifies and records the athlete's feature data, including clothing features, gait features, and accessory features. The feature similarity index is calculated by analyzing the feature data, and the athlete's features are identified when bib number recognition is affected.

[0053] The calculation process for the feature similarity index of marathon runners is as follows:

[0054] S31. Obtain initial feature data of marathon runners and match it with runner numbers. At the same time, obtain clothing feature data (perform refined feature extraction of runners' clothing, including the color, texture, and brand logo of tops, pants, and shoes), gait feature data (perform biometric analysis of runners' dynamic running posture, including cadence, stride length, arm swing posture, and body tilt angle), and accessory feature data (record the combination of hats, sunglasses, sports watches, waist bags, and headphone auxiliary equipment worn by runners) and extract feature vectors, and then perform analysis and calculation.

[0055] S32. Calculate the characteristic similarity index S of marathon runners according to the following formula:

[0056]

[0057] Where I represents the total number of clothing features worn by a marathon runner, J represents the total number of gait features worn by a marathon runner, K represents the total number of accessory features worn by a marathon runner, and P represents the total number of clothing features worn by a marathon runner. i Let Q be the feature vector value of the i-th clothing feature of a runner on the marathon course. j Let R be the feature vector value of the j-th gait feature of a runner on a marathon course. k Let P be the feature vector value of the k-th accessory feature of a runner on the marathon course. e Q is the preset initial clothing feature vector value for marathon runners. e R represents the preset initial gait feature vector values ​​of a marathon runner. e The preset initial accessory feature vector values ​​for marathon runners are: α is the preset clothing weight coefficient, β is the preset gait weight coefficient, γ is the preset accessory weight coefficient, and α+β+γ=1;

[0058] S33. Obtain the preset feature similarity threshold S th Comparative analysis with the feature similarity index S, when S>S th If S ≤ S, it indicates that the runners on the marathon course have a high degree of similarity in initial characteristics to the marathon runners, and it is determined that they are the same person. The runners on the marathon course are then matched with corresponding runner numbers. th If the initial characteristics of the runners on the marathon course are not similar to those of the marathon runners, it indicates that the two are not the same person and the runner numbers do not match.

[0059] This invention employs automatic frame extraction technology to intelligently filter video data, retaining only clear frames with varying content to avoid redundant data interference, significantly improving monitoring efficiency and reducing the risk of human error. Simultaneously, it combines environmental impact index calculation with dynamic assessment of environmental interference using light intensity and visibility sensors. When lighting is too strong or visibility is too low, it automatically enhances image quality, effectively solving the problem of high failure rates in bib number OCR recognition under complex scenarios such as backlighting, rain, and fog. Furthermore, feature vector extraction and re-identification technology captures multi-dimensional features such as contestant clothing, gait, and accessories to construct a feature similarity index, enabling accurate tracking of contestant identities even when bib numbers are obscured or damaged, ensuring fairness in the competition.

[0060] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0061] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0062] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A marathon track AI eagle-eye recognition system based on image processing, characterized in that, include: The system includes a video extraction unit, an image processing unit, a bib number detection unit, a number recognition unit, a re-recognition unit, and a data management unit. The video extraction unit is used to automatically extract video data from cameras at specified timing points on the marathon course; The image processing unit is used to intelligently and automatically extract frames from video data on the marathon track, obtain image quality data from the video library, and calculate an image selection index through analysis to retain clear frames with varying content. The bib number detection unit is used to identify the location of bibs in the image using a target detection algorithm; The bib number recognition unit is used to collect environmental data on the marathon course using a sensor array, analyze and calculate the environmental impact index, and use OCR technology to recognize the numbers and letters on the bib number. The re-identification unit is used to identify and record the feature data of the athletes, and calculate the feature similarity index by analyzing the feature data. When the bib number identification is affected, the athlete's features are identified. The data management unit is used to store the recognition results into the database, and supports multi-dimensional querying of marathon athletes' race information by number, time, and location.

2. The marathon track AI eagle-eye recognition system based on image processing according to claim 1, characterized in that, The image quality data includes the RGB values ​​and image gradient of the image; the environmental data includes the ambient light intensity and visibility; and the feature data includes clothing features, gait features, and accessory features.

3. The marathon track AI eagle-eye recognition system based on image processing according to claim 1, characterized in that, The calculation process for the image selection index of marathon track images is as follows: S11. Obtain the RGB values ​​and gradient data of the image and perform analysis and calculation; S12. Calculate the image selection index (CSIR) for marathon track images using the following formula: Where RGB represents the RGB values ​​of the marathon track image. e The RGB values ​​are the preset standard image values, M is the length of the marathon track image, N is the width of the marathon track image, and G is the length of the marathon track image. x (m,n) represents the gradient in the x-direction of the marathon track image, G y (m,n) represents the gradient in the y-direction of the marathon track image; S13. Obtain the preset image filtering threshold CSIR. th A comparative analysis was conducted with the image screening index CSIR. When CSIR > CSIR th If CSIR ≤ CSIR, it indicates that the image of the marathon track in that frame is clear but has undergone significant changes and needs to be retained. th If the image of the marathon track is unclear or its content has not changed significantly, it needs to be deleted.

4. The marathon track AI eagle-eye recognition system based on image processing according to claim 1, characterized in that, The calculation process for the environmental impact index of a marathon course is as follows: S21. Acquire light intensity and visibility data and perform analysis and calculation; S22. Calculate the environmental impact index E of the marathon course using the following formula: Where L represents the light intensity of the marathon track environment, L max L is the upper limit of the preset standard light intensity. min V represents the lower limit of the preset standard light intensity, and V is the visibility of the marathon track environment. min V is the preset lower limit of standard visibility. max is the preset upper limit of standard visibility, 'a' is the preset weighting coefficient of light intensity, and 'b' is the preset weighting coefficient of visibility, and a+b=1. The environmental impact index is used to reflect the degree of influence of the environment on the quality of the captured image. S23. Obtain the preset environmental impact threshold E th Comparative analysis with the environmental impact index E, when E>E th When E > E, it indicates that the marathon course environment has a high impact on the quality of the captured image, and the bib number image will be automatically enhanced using image enhancement technology. th If the image quality is good, it means that the marathon track environment has a low impact on the quality of the captured image, and no image enhancement is needed.

5. The marathon track AI eagle-eye recognition system based on image processing according to claim 1, characterized in that, The calculation process for the feature similarity index of marathon runners is as follows: S31. Obtain initial feature data of marathon runners and match it with runner numbers. At the same time, obtain clothing feature, gait feature and accessory feature data and extract feature vectors, and then perform analysis and calculation. S32. Calculate the characteristic similarity index S of marathon runners according to the following formula: Where I represents the total number of clothing features worn by a marathon runner, J represents the total number of gait features worn by a marathon runner, K represents the total number of accessory features worn by a marathon runner, and P represents the total number of clothing features worn by a marathon runner. i Let Q be the feature vector value of the i-th clothing feature of a runner on the marathon course. j Let R be the feature vector value of the j-th gait feature of a runner on a marathon course. k Let P be the feature vector value of the k-th accessory feature of a runner on the marathon course. e Q is the preset initial clothing feature vector value for marathon runners. e R represents the preset initial gait feature vector values ​​of a marathon runner. e The preset initial accessory feature vector values ​​for marathon runners are: α is the preset clothing weight coefficient, β is the preset gait weight coefficient, γ is the preset accessory weight coefficient, and α+β+γ=1; S33. Obtain the preset feature similarity threshold S th Comparative analysis with the feature similarity index S, when S>S th If S ≤ S, it indicates that the runners on the marathon course have a high degree of similarity in initial characteristics to the marathon runners, and it is determined that they are the same person. The runners on the marathon course are then matched with corresponding runner numbers. th If the initial characteristics of the runners on the marathon course are not similar to those of the marathon runners, it indicates that the two are not the same person and the runner numbers do not match.

6. The marathon track AI eagle-eye recognition system based on image processing according to claim 1, characterized in that, The sensor group includes a light intensity sensor and a visibility sensor.

Images