Basketball player action recognition method and system using visual detection technology

Abstract

The application relates to the technical field of image processing, in particular to a basketball player action recognition method applying visual detection technology, which comprises the following steps: extracting a two-dimensional skeleton of a basketball player from multiple continuous images of a basketball area, and recognizing an action performed by the basketball player based on the change of the two-dimensional skeleton in each image; when the jumping action of the basketball player is recognized, determining the landing risk of the basketball player after performing the jumping action according to the change of the lower limbs of the basketball player in the process of performing the jumping action; determining the influence degree of other basketball players on the current basketball player in the process of performing the jumping action; determining the actual risk of the current basketball player in the process of performing the jumping action, and determining the action type of the current basketball player in the process of performing the jumping action based on the actual risk. Thus, the application realizes the evaluation of the action risk of the basketball player.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and particularly relates to a basketball player action recognition method and system applying visual detection technology. BACKGROUND

[0002] In the context of rapid development of modern sports science and technology, accurate monitoring and analysis of athlete performance is increasingly important. As a highly competitive and fast-paced sport, basketball players' actions on the court are complex and varied. Artificial intelligence and machine learning technology has gradually penetrated into various aspects of basketball, especially in the identification of athlete actions, showing great potential. Through continuous training of deep learning models, athlete actions can be accurately predicted and analyzed, which is of great significance for understanding the competitive state of athletes and preventing sports injuries.

[0003] In some scenarios, there are still significant technical bottlenecks in identifying basketball player action risks based on visual detection. Existing image action recognition technology can only capture the current actions of athletes, but it is difficult to accurately infer the actual action risks from these actions. In basketball, the generation of athlete action risks is influenced by a variety of factors, including the athlete's own movement style, the correctness of the posture, and the situation of being hit by other athletes in intense competition. These complex factors interact, greatly increasing the difficulty of accurately assessing athlete action risks, such as risk assessment of basketball players' jumping actions. Therefore, how to assess the action risk of basketball players is a technical problem that needs to be solved by those skilled in the art. SUMMARY

[0004] To solve the technical problem of assessing the action risk of basketball players, the purpose of the present application is to provide a basketball player action recognition method and system applying visual detection technology, and the technical solution adopted is as follows:

[0005] The embodiment of the application provides a basketball player action recognition method applying visual detection technology, comprising: extracting a two-dimensional skeleton of a basketball player from a plurality of continuous images of a basketball area, and recognizing an action performed by the basketball player based on changes of the two-dimensional skeleton in each image; when it is recognized that the basketball player performs a jumping action, determining a landing risk of the basketball player after performing the jumping action according to changes of lower limbs of the basketball player in a process of performing the jumping action; determining an influence degree of the basketball player on the current basketball player in the process of performing the jumping action according to information of key points of the two-dimensional skeleton of the current basketball player and two-dimensional skeletons of other basketball players; determining an actual risk of the current basketball player in the process of performing the jumping action according to an included angle between a line connecting a head key point and a foot key point of the current basketball player when landing and the ground, the landing risk and the influence degree, and determining an action type of the current basketball player in the process of performing the jumping action based on the actual risk.

[0006] Optionally, the landing risk of the basketball player after performing the jumping action is determined according to changes of lower limbs of the basketball player in the process of performing the jumping action, comprising: determining a probability that the lower limbs of the basketball player exist a take-off action according to changes of a first knee angle and a thigh angle in the two-dimensional skeleton before taking off when the basketball player performs the jumping action; determining that the basketball player performs the jumping action in a case where the probability is greater than a preset threshold; determining a jumping distance, a jumping height and a movement speed before taking off of the basketball player in the process of performing the jumping in the two-dimensional skeleton of the basketball player in each image; determining an inertia of the basketball player after landing according to the jumping distance, the jumping height and the movement speed; determining a time difference of the lower limbs of the basketball player after landing and a first distance between landing positions of the lower limbs of the basketball player according to a time of landing of the two-dimensional skeleton of the basketball player and the landing positions in each image; determining a synchronism of the lower limbs of the basketball player after landing based on the time difference and the first distance; and determining the landing risk of the basketball player after landing according to the inertia, the synchronism and changes of a second knee angle and a hip joint angle in the two-dimensional skeleton of the basketball player after landing.

[0007] Optionally, the probability that the lower limbs of the basketball player exist the take-off action is determined according to changes of the first knee angle and the thigh angle in the two-dimensional skeleton before taking off when the basketball player performs the jumping action, comprising: fitting the first knee angle in the two-dimensional skeleton before taking off when the basketball player performs the jumping action with time to obtain a first fitting curve, and fitting the thigh angle in the two-dimensional skeleton before taking off when the basketball player performs the jumping action with time to obtain a second fitting curve; taking a mean value of a second derivative function of the first fitting curve as a knee take-off degree, and obtaining a first average slope of the second fitting curve; calculating a first ratio between the knee take-off degree and the first average slope, and performing normalization processing on the first ratio to obtain the probability.

[0008] Optionally, determining the inertia suffered by the basketball player after landing according to the jumping distance, the jumping height and the movement speed comprises: calculating a second ratio between the jumping distance and the jumping height, and calculating an absolute value of a first difference between the second ratio and a predetermined value; calculating a first product between the absolute value of the first difference and the movement speed; and performing normalization processing on the first product to obtain the inertia.

[0009] Optionally, determining the synchronism of the lower limbs of the basketball player after landing based on the time difference and the first distance comprises: calculating a second product between the time difference and the first distance, and performing inverse proportional normalization processing on the second product to obtain the synchronism.

[0010] Optionally, determining the landing risk of the basketball player after landing according to the inertia, the synchronism, and the change of the second knee angle and the hip joint angle in the two-dimensional skeleton after the basketball player lands after performing the jumping action comprises: obtaining a third fitting curve of the second knee angle in the two-dimensional skeleton after the basketball player lands after performing the jumping action changing with time, and a fourth fitting curve of the hip joint angle in the two-dimensional skeleton after the basketball player lands after performing the jumping action changing with time; calculating a third product between a first average angle of the third fitting curve and a second average angle of the fourth fitting curve, and calculating a third ratio between the third product and the synchronism; calculating a fourth product between the third ratio and the inertia, and performing normalization processing on the fourth product to obtain the landing risk.

[0011] Optionally, determining the degree of influence of other basketball players on the current basketball player during the jumping action according to the information of the key points of the two-dimensional skeleton of the current basketball player and the two-dimensional skeletons of the other basketball players comprises: taking the nearest key points of the current basketball player and the adjacent other basketball players as the affected key points of the current basketball player, and obtaining the nearest distances between the affected key points and the adjacent other basketball players; obtaining a second distance between the affected key points of the current basketball player and the trunk skeleton line thereof; fitting the change of the nearest distances with time to obtain a fifth fitting curve, and fitting the change of the second distances between the affected key points of the current basketball player and the trunk skeleton line thereof with time to obtain a sixth fitting curve; calculating an average value between a second average slope of the fifth fitting curve and a third average slope of the sixth fitting curve, and a mean square error between the fifth fitting curve and the sixth fitting curve; calculating a fifth product between the mean square error and the average value, and performing inverse proportional normalization processing on the fifth product to obtain the degree of influence.

[0012] Optionally, determining the actual risk of the current basketball player when performing a jumping action based on the angle between the line between the key points of the head and the key points of the feet and the ground when the current basketball player lands, the landing risk and the impact level includes: fitting the changes over time in the angle between the line between the key points of the head and the key points of the feet and the ground when the current basketball player lands to obtain a sixth fitting curve; obtaining a fourth average slope of the sixth fitting curve; calculating a fourth ratio between the impact level and the fourth average slope, and calculating a sixth product between the fourth ratio and the landing risk; and normalizing the sixth product to obtain the actual risk.

[0013] Optionally, determining the action type of the current basketball player's jumping action based on the actual risk includes: when the actual risk is less than a first threshold, determining that the action type of the current basketball player's jumping action is a low-risk jumping type; when the actual risk is greater than or equal to the first threshold and less than a second threshold, determining that the action type of the current basketball player's jumping action is a medium-risk jumping type; when the actual risk is greater than or equal to the second threshold, determining that the action type of the current basketball player's jumping action is a high-risk jumping type.

[0014] In a second aspect, an embodiment of the present invention provides a basketball player action recognition system using visual detection technology, comprising: a processor and a memory; wherein the memory is used to store computer programs that can be run on the processor; and the processor is used to execute the program stored in the memory to implement the steps of the basketball player action recognition method using visual detection technology as in the first aspect.

[0015] The present invention has the following beneficial effects: first, a two-dimensional skeleton of a basketball player is extracted from multiple continuous images of a basketball area, and the actions of the basketball player are identified based on changes in the two-dimensional skeleton in each image; then, when it is identified that the basketball player is performing a jumping action, the landing risk of the basketball player after the jumping action is determined according to changes in the lower limbs of the basketball player during the jumping action; secondly, based on information on key points of the two-dimensional skeleton of the current basketball player and the two-dimensional skeletons of other basketball players, the degree to which the current basketball player is affected by other basketball players during the jumping action is determined; finally, based on the angle between the line between the head key point and the foot key point of the current basketball player when landing and the ground, the landing risk and the degree of influence, the actual risk of the current basketball player in the jumping action is determined, and the action type of the current basketball player's jumping action is determined based on the actual risk.

[0016] In this way, the present invention can analyze a basketball player's jumping movements during exercise based on the player's two-dimensional skeleton, obtain the player's landing risk based on the changes in the player's lower limbs during the jump, and then determine the degree of influence of other basketball players on the current player based on the influence of other basketball players on the current player. Based on this influence and the landing risk, the actual risk of the basketball player when performing the jumping movement is determined, thereby making the risk assessment of the basketball player's jumping movement more accurate. Therefore, the above-mentioned embodiments of the present invention achieve the purpose of assessing the movement risk of a basketball player. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A flowchart of a method for identifying basketball player movements using visual detection technology is provided in an embodiment of the present invention.

[0019] Figure 2 A schematic diagram of a two-dimensional skeleton of a basketball player provided by an embodiment of the present invention.

[0020] Figure 3 A schematic diagram of a first knee angle and thigh angle provided in an embodiment of the present invention.

[0021] Figure 4 A schematic diagram of the hip joint angle provided in an embodiment of the present invention.

[0022] Figure 5 A schematic diagram of the angle between the line connecting the key points of the head and the key points of the feet of a current basketball player when landing is provided in an embodiment of the present invention.

[0023] Figure 6 A schematic structural diagram of a basketball player action recognition system using visual detection technology provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail a basketball player motion recognition method using visual detection technology, including its specific implementation, structure, features, and effectiveness. In the following description, references to different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0025] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0026] The following describes in detail a specific solution of a basketball player action recognition method using visual detection technology provided by the present invention with reference to the accompanying drawings.

[0027] Example 1:

[0028] See also Figure 1 , which shows a flow chart of a basketball player action recognition method using visual detection technology provided by one embodiment of the present invention, including:

[0029] S101 , extracting a two-dimensional skeleton of a basketball player from a plurality of consecutive images of a basketball area, and recognizing actions performed by the basketball player based on changes in the two-dimensional skeleton in each image.

[0030] Specifically, the basketball area in the embodiment of the present invention includes multiple basketball players. During the exercise of the basketball players, the team doctor needs to identify high-risk actions that may cause injury during the exercise process in order to analyze the changes in the body load of the athletes during the exercise and develop personalized rehabilitation plans for the athletes. Therefore, in order to perform action recognition on the basketball players, the embodiment of the present invention first needs to capture images of the basketball players. Since basketball is a fast sport, the movements of basketball players are frequent and rapid, and in order to capture every detail of the basketball players' movements, a high frame rate (for example, 60 frames per second or higher) and high-definition camera is required to ensure that every moment of the basketball players' rapid movements is captured. The camera needs to cover the entire sports area of ​​the basketball court to ensure that the movements of the basketball players can be captured at different positions during the game or training.

[0031] Furthermore, after capturing an image of the basketball area, in order to obtain the movement of the basketball players in the image, it is first necessary to extract the basketball players in the image. The embodiment of the present invention utilizes YOLO to identify basketball players in an image. YOLO divides the image into grids, uses a convolutional neural network to extract image features, and predicts the position and size of the bounding box through regression, while performing target classification. Finally, YOLO outputs the bounding box of each detected basketball player. After obtaining the bounding box containing the basketball players, OpenPose is used to extract the human key points of each basketball player in the image, and the human key points extracted from the image are connected to obtain a two-dimensional skeleton of the basketball player. For example, as Figure 2 As shown, Figure 2 A schematic diagram of a two-dimensional skeleton of a basketball player provided by an embodiment of the present invention. Figure 2 Contains the main torso of the human body.

[0032] Furthermore, as an optional embodiment of the present invention, identifying the actions of a basketball player based on changes in the two-dimensional skeleton in each image includes: determining the probability that the basketball player's lower limbs have a jumping action based on changes in the first knee angle and thigh angle in the two-dimensional skeleton before the basketball player leaves the ground when performing a jumping action; if the probability is greater than a preset threshold, determining that the basketball player has performed a jumping action.

[0033] Specifically, for the basketball player in the image captured by the above operation, when the basketball player jumps, the action is relatively intense, which may cause a high risk of injury to the basketball player after the action. Therefore, to determine whether the basketball player has performed a jumping action, the embodiment of the present invention first needs to detect whether the player's lower limbs have left the ground based on changes in the basketball player's background in the scene. When the basketball player is detected to have left the ground, two image data segments are captured, namely, the segment where the basketball player left the ground and the segment where the player touched the ground before and after leaving the ground.

[0034] Furthermore, after extracting the two-dimensional skeleton, the embodiment of the present invention obtains the two-dimensional skeleton of the lower limbs of the basketball player before leaving the ground. The two-dimensional skeleton of the lower limbs contains the two legs of the basketball player. When the basketball player performs a jumping action, the lower limbs will bend and fold to store power, and then exert force to jump up. Therefore, it is necessary to analyze whether the lower limbs of the basketball player have a process of storing power to jump. First, the embodiment of the present invention obtains the knee joints of the two legs in the two-dimensional skeleton of the basketball player's lower limbs, and obtains the first knee angle of the basketball player based on the angle between the thigh and calf skeleton lines on both sides of the knee joint, and obtains the thigh angle between the two thigh skeleton lines in the two-dimensional skeleton of the basketball player's lower limbs. For example, Figure 3 As shown, Figure 3A schematic diagram of a first knee angle and a thigh angle provided by an embodiment of the present invention. Based on the first knee angle and the thigh angle, it is possible to identify whether a basketball player has taken a jump.

[0035] S102, when recognizing that the basketball player is performing a jumping action, determining the landing risk of the basketball player after the jumping action according to changes in the lower limbs of the basketball player during the jumping action.

[0036] Specifically, when a basketball player has an incorrect landing posture or overloads the joints during a game or training, it may cause ligament damage or muscle strain. However, there are a lot of running, jumping, and confrontational actions in basketball, and it is necessary to analyze whether there are risks in the actions. In the embodiment of the present invention, for the basketball player in the image obtained by the above operation, when the basketball player jumps, the action is relatively violent, which may cause the basketball player to be at high risk of injury after making the action. Therefore, in order to determine whether the basketball player has performed a jumping action, it is first necessary to detect whether the player's lower limbs have left the ground based on the changes in the background of the basketball player in the scene to determine whether the basketball player has performed a jumping action and determine the landing risk of the jumping action.

[0037] Furthermore, as an optional embodiment of the present invention, determining the landing risk of a basketball player after performing a jumping action based on changes in the lower limbs of the basketball player during the jumping action includes: determining the probability that the basketball player's lower limbs have a take-off action based on changes in the first knee angle and thigh angle in the two-dimensional skeleton before the basketball player leaves the ground when performing the jumping action; when the probability is greater than a preset threshold, determining that the basketball player has performed a jumping action, and determining the jumping distance, jumping height and movement speed of the basketball player before taking off during the jumping process based on the two-dimensional skeleton of the basketball player in each image; determining the inertia of the basketball player after landing based on the jumping distance, jumping height and movement speed; determining the time difference of the basketball player's lower limbs landing and the first distance between the lower limb landing points based on the time and landing point of both feet in the two-dimensional skeleton of the basketball player in each image; determining the synchronization of the basketball player's lower limbs landing based on the time difference and the first distance; determining the landing risk of the basketball player after performing a jumping action based on inertia, synchronization, and changes in the second knee angle and hip joint angle in the two-dimensional skeleton after the basketball player lands after performing the jumping action.

[0038] Specifically, when a basketball player performs a power-accumulating jump, the knees will bend and extend, and the athlete's knee angle will decrease and then increase. Furthermore, before the jump, the athlete's two thighs will merge to accumulate power, so the thigh angle will decrease. Therefore, as an optional embodiment of the present invention, based on the changes in the first knee angle and thigh angle in the two-dimensional skeleton before the basketball player leaves the ground when performing a jump, determining the probability that the basketball player's lower limbs have a jump includes: fitting the changes in the first knee angle in the two-dimensional skeleton before the basketball player leaves the ground with time when performing a jump to obtain a first fitting curve; fitting the changes in the thigh angle in the two-dimensional skeleton before the basketball player leaves the ground with time when performing a jump to obtain a second fitting curve; recording the mean of the second-order derivative function of the first fitting curve as the knee jump degree, and obtaining the first average slope of the second fitting curve; calculating a first ratio between the knee jump degree and the first average slope, and normalizing the first ratio to obtain a probability.

[0039] Specifically, this embodiment of the present invention fits the basketball player's knee angle in the image before the player leaves the ground to a first fitting curve q1, and fits the thigh angle to a second fitting curve q2. The second derivative of the first fitting curve q1 is obtained. Because the knee angle decreases first and then increases during takeoff, the fitted curve for the knee angle before the jump is a concave function. Therefore, the mean value s of the second derivative is recorded as the knee takeoff angle. Furthermore, because the athlete's thighs converge and accumulate power before takeoff, the first average slope k of the second fitting curve q2 is obtained.

[0040] Furthermore, the embodiment of the present invention uses the following formula to calculate the probability that the basketball player's lower limbs are in a take-off action:

[0041]

[0042] In the above formula, is the probability of a take-off action in the lower limbs. It's the knee jump. is the first average slope of the second fitting curve q2. is the hyperbolic tangent function, the ratio between the basketball player's knee jump and the first average slope of the second fitting curve q2 The larger the value, the greater the probability that the basketball player's lower limbs have a take-off action.

[0043] Furthermore, the embodiment of the present invention compares the probability of the acquired jump action with the preset threshold value. Compare, if it is greater than the preset threshold It indicates that the basketball player has made a jump.

[0044] Furthermore, when a basketball player performs a jump, different jump trajectories result in different inertial effects upon landing. When a basketball player jumps from a stationary position, the joints, muscles, and ligaments in the body only need to cushion the impact upon landing. However, when a basketball player performs a long-distance jump during a sprint, the body not only needs to cushion the impact upon landing, but the joints and ligaments also need to withstand the shear forces caused by the inertia of the lateral velocity. Therefore, the embodiment of the present invention first obtains the basketball player's pre-takeoff velocity v based on the movement speed of the basketball player's head key point in the image. Then, the basketball player's liftoff phase is obtained, and the distance x between the liftoff point and the landing point in the image during the liftoff phase is obtained, which is recorded as the jump distance. Furthermore, the difference h between the height of the highest point of the basketball player's head key point during the liftoff phase and the height of the basketball player's head key point at the beginning of the liftoff phase is obtained, which is recorded as the basketball player's jump height. The basketball player's head key point at the beginning of the liftoff phase refers to the head key point at the moment when the basketball player's lower limbs leave the ground.

[0045] Furthermore, as an optional embodiment of the present invention, determining the inertia of a basketball player after landing based on the jumping distance, jumping height, and movement speed includes: calculating a second ratio between the jumping distance and the jumping height, and calculating the absolute value of a first difference between the second ratio and a predetermined value; calculating a first product between the absolute value of the first difference and the movement speed; and normalizing the first product to obtain inertia.

[0046] Specifically, the embodiment of the present invention uses the following formula to calculate inertia:

[0047]

[0048] In the above formula, It is the inertia a basketball player experiences after landing. is the jump distance. is the jump height. is the speed of movement before take-off. is the hyperbolic tangent function, which is used to Normalize the product of the absolute value of the difference between the ratio of the jump distance and the jump height of a basketball player and 1 and the speed before take-off. The larger the value, the greater the speed of the basketball player before taking off, and the greater the difference between the jump height and the jump distance, which means that the basketball player will be subject to greater inertia after taking off and landing.

[0049] Furthermore, when a basketball player is subjected to inertia, a better landing posture can absorb most of the impact, thereby reducing the risk. Therefore, when a basketball player is subjected to inertia, the risks caused by the basketball player's landing posture are analyzed.

[0050] Furthermore, after a basketball player jumps, the muscles and ligaments in the player's body absorb the impact of landing. If the player's landing posture is incorrect, the body may not be able to absorb the impact well when landing, which may lead to a higher risk to the joint ligaments. When the basketball player bends the knees and hips in time when landing, the muscles and ligaments in the body can better absorb the impact of jumping and landing, and the ability to absorb multiple shocks when both feet land at the same time is also better. Therefore, it is necessary to analyze whether the two legs of the basketball player bend the knees and hips in time when landing. The time difference t between the landing points of the two lower limbs of the basketball player is obtained according to the time points of the landing of the two lower limbs of the basketball player, and the first distance z between the landing points of the two lower limbs of the athlete after the two lower limbs of the athlete land is obtained. When the two lower limbs of the athlete land synchronously, the time difference between the landing points of the basketball player's lower limbs is smaller and the distance between the landing points of the lower limbs is also smaller.

[0051] Furthermore, as an optional embodiment of the present invention, determining the synchronization of the landing of the lower limbs of a basketball player based on the time difference and the first distance includes: calculating the second product between the time difference and the first distance, and performing inverse proportional normalization processing on the second product to obtain synchronization.

[0052] Specifically, the embodiment of the present invention uses the following formula to calculate synchronization:

[0053]

[0054] In the above formula, It is the synchronization of the landing of the lower limbs of basketball players. It is the time difference when the lower limbs hit the ground. It is the first distance the lower limbs touch the ground. It is an exponential function with the natural constant e as the base, used to Perform inverse proportional normalization. The product of the time difference between the lower limb landing of the basketball player and the first distance , the smaller the value, the higher the synchronization of lower limb landing.

[0055] Furthermore, in the embodiment of the present invention, a two-dimensional skeleton of a basketball player after landing is obtained, and the second knee angle is obtained therein, and the hip joint angle of the player is obtained according to the angle between the torso skeleton line and the thigh skeleton line in the two-dimensional skeleton of the player. For example, Figure 4 As shown, Figure 4 A schematic diagram of hip joint angles provided in accordance with an embodiment of the present invention.

[0056] Furthermore, after a basketball player lands from a jump, the body bends the knees and hips in time to unload the force, which reduces the damage to the joint ligaments caused by the impact of landing. At this time, the knee angle and hip angle of the basketball player will gradually decrease. The embodiment of the present invention obtains a third fitting curve of the second knee angle of the basketball player after landing, and obtains the first average angle of the third fitting curve, which is recorded as The first average angle is the average angle of the second knee angle change after the basketball player lands. And obtain the fourth fitting curve of the hip joint angle after the basketball player lands, and obtain the second average angle of the fourth fitting curve and record it as ,The second average angle is the average angle of the hip joint angle change of the basketball player after landing.

[0057] Furthermore, as an optional embodiment of the present invention, determining the landing risk of a basketball player after landing after a jumping action based on inertia, synchronization, and changes in the second knee angle and hip joint angle in the two-dimensional skeleton after the basketball player lands after a jump action includes: obtaining a third fitting curve of the second knee angle in the two-dimensional skeleton of the basketball player after the jump action changes with time, and a fourth fitting curve of the hip joint angle in the two-dimensional skeleton of the basketball player after the jump action lands with time; calculating a third product between the first average angle of the third fitting curve and the second average angle of the fourth fitting curve, and calculating a third ratio between the third product and synchronization; calculating a fourth product between the third ratio and inertia, and normalizing the fourth product to obtain the landing risk.

[0058] Specifically, the embodiment of the present invention uses the following formula to calculate the landing risk:

[0059]

[0060] In the above formula, It is the landing risk of basketball players when they jump. is the first mean angle of the third fitted curve of the second knee angle. is the second mean angle of the fourth fitting curve of the hip joint angle. is the hyperbolic tangent function, which is used to Perform normalization processing. It is the synchronization of the landing of the lower limbs of basketball players. It is the inertia of the basketball player after landing. The ratio between the average angle of the fitting curve of the second knee angle and hip joint angle of the basketball player and the synchronization of the lower limb landing The larger the value, the lower the synchronization of the basketball player when landing and the larger the angle between the knee and hip joints. Therefore, it means that the athlete did not land at the same time and did not flex the knees and hips in time to unload the force. Therefore, the risk of landing is greater under the inertia of the basketball player's current jumping movement.

[0061] S103 , determining the degree to which the current basketball player is affected by other basketball players during a jumping action based on information about key points of the current basketball player's two-dimensional skeleton and the two-dimensional skeletons of other basketball players.

[0062] Specifically, the landing risk of a basketball player's jumping action obtained in the above-mentioned embodiment of the present invention is due to the fact that basketball is a confrontational sport. Therefore, basketball players will be subject to risks caused by their own jumping actions during the jumping process, and may also be collided with other basketball players during the jumping process. Therefore, it is also necessary to analyze whether the basketball player will be affected by other basketball players in the current jumping landing risk situation, thereby increasing his risk.

[0063] Furthermore, as an optional embodiment of the present invention, determining the degree to which the current basketball player is influenced by other basketball players during a jumping action based on information about key points of the current basketball player's two-dimensional skeleton and other basketball players' two-dimensional skeletons includes: taking the closest key point between the current basketball player and other adjacent basketball players as the affected key point of the current basketball player, and obtaining the closest distance between the affected key point and other adjacent basketball players; obtaining the second distance between the affected key point of the current basketball player and his torso skeleton line; fitting the change over time of the closest distance to obtain a fifth fitting curve, and fitting the change over time of the second distance between the affected key point of the current basketball player and his torso skeleton line to obtain a sixth fitting curve; calculating the average value between the second average slope of the fifth fitting curve and the third average slope of the sixth fitting curve, as well as the mean square error between the fifth fitting curve and the sixth fitting curve; calculating the fifth product between the mean square error and the average value, and performing inverse proportional normalization on the fifth product to obtain the degree of influence.

[0064] Specifically, according to the left side of the bounding box of the basketball player in the image, the current basketball player is recorded as the adjacent player in the take-off stage with the nearest basketball player. The key point of the current basketball player and the adjacent player is recorded as the affected key point of the current basketball player, and the nearest distance d between the affected key point and the key point of the adjacent other basketball player and the affected key point is obtained. Then the second distance g between the affected key point of the basketball player and the trunk skeleton line is obtained. The closer the distance between the other basketball player and the basketball player, and the closer the contacted part to the trunk of the basketball player, the more likely it will affect the jumping basketball player.

[0065] Therefore, the embodiment of the present application obtains the normalized fifth fitting curve q3 of the nearest distance of the basketball player in the take-off stage and the adjacent player. And the normalized sixth fitting curve q4 of the second distance between the affected key point of the basketball player in the take-off stage and the trunk skeleton line. And the mean square error MSE between the fitting curves q3, q4, and the average value y between the second average slope and the third average slope of the fitting curves q3, q4 are obtained.

[0066] Further, the embodiment of the present application calculates the influence degree by the following formula:

[0067]

[0068] In the above formula, is the influence degree of the current basketball player in the process of jumping action by other basketball players. is the mean square error of the fitting curves q3, q4. is the average value of the second average slope and the third average slope of the fitting curves q3, q4. is the exponential function with the natural constant e as the base, which is used to inversely normalize . Wherein the product of the mean square error and the slope of the fitting curve The smaller the value, the closer the adjacent basketball players in the take-off stage and the closer to the trunk of the current basketball player, so it indicates that the basketball player in the take-off stage will be more affected by the surrounding basketball players.

[0069] S104, according to the angle between the line connecting the head key point and the foot key point of the current basketball player when landing and the ground, the landing risk and the influence degree, determine the actual risk of the current basketball player in the jumping action, and determine the action type of the current basketball player in the jumping action based on the actual risk.

[0070] Specifically, when the basketball player is jumping, the surrounding basketball players can be colliding with him or protecting the jumping basketball player, so the embodiment of the application needs to analyze the performance of the basketball player after landing to obtain the actual risk.

[0071] Further, as an optional embodiment of the application, determining the actual risk of the current basketball player in the jumping action according to the included angle between the line between the head key point and the foot key point of the current basketball player when landing and the ground, the landing risk and the influence degree includes: fitting the included angle between the line between the head key point and the foot key point of the current basketball player when landing and the ground with time to obtain a sixth fitting curve; obtaining a fourth average slope of the sixth fitting curve; calculating a fourth ratio between the influence degree and the fourth average slope, and calculating a sixth product between the fourth ratio and the landing risk; and normalizing the sixth product to obtain the actual risk.

[0072] Specifically, the embodiment of the application obtains the line between the head key point and the foot key point of the basketball player in the landing stage, and obtains the included angle between the line and the ground , which is the inclination of the player. As shown in Figure 5 , as shown in Figure 5 , the embodiment of the application provides a current basketball player's head key point and foot key point between the line and the ground when landing. Since the basketball player is protected, the body inclination after landing will be smaller and smaller, and when colliding with other basketball players, the body inclination after landing can be larger and larger due to imbalance, so the embodiment of the application obtains the fourth average slope c of the sixth fitting curve of the inclination of the basketball player in the landing stage. The foot key point refers to the key point at the ankle position of the first landing foot.

[0073] Further, the embodiment of the application calculates the actual risk by the following formula:

[0074]

[0075] In the above formula, is the actual risk of the basketball player in the jumping action. is the fourth average slope of the sixth fitting curve. is a hyperbolic tangent function, which is used to normalize . For the ratio between the influence of the basketball player on other basketball players and the fourth average slope of the sixth fitting curve The larger the value, the greater the influence of other basketball players on the current basketball player, and the more serious the tilt of the body after jumping and landing. Therefore, it is clear that in the risky situation of the current basketball player's jump, the greater the influence of other basketball players, and therefore the greater the actual risk.

[0076] Furthermore, the embodiment of the present invention classifies the actual risks of the basketball player's jumping action obtained above. As an optional embodiment of the present invention, determining the action type of the current basketball player's jumping action based on the actual risk includes: when the actual risk is less than a first threshold, determining that the action type of the current basketball player's jumping action is a low-risk jumping type; when the actual risk is greater than or equal to the first threshold and less than the second threshold, determining that the action type of the current basketball player's jumping action is a medium-risk jumping type; when the actual risk is greater than or equal to the second threshold, determining that the action type of the current basketball player's jumping action is a high-risk jumping type.

[0077] Specifically, in the embodiment of the present invention, the first threshold and the second threshold can be determined according to actual conditions. In the embodiment of the present invention, the first threshold is 0.3, and the second threshold is 0.7.

[0078] Furthermore, when It indicates a low risk jump when When it indicates a medium risk jump, It indicates a high-risk jump.

[0079] The embodiments of the present invention are designed to increase basic strength training, especially leg and core strength, for basketball players with low-risk jumps, helping them improve their overall strength and stability. For basketball players with medium-risk jumps, the jumping posture of the basketball players is evaluated to ensure that their movements are standardized and to reduce injuries caused by improper posture. The training intensity is strictly controlled, and high-intensity jump training and rest days are reasonably arranged to avoid excessive fatigue or inappropriate training intensity. For basketball players with high-risk jumps, jumping techniques and landing skills are strengthened to ensure a stable body posture during jumping, avoid inward-bowed knees or unnatural postures, and strengthen the core muscles (abdominal muscles, back muscles, etc.) to increase body stability and reduce the risk of injury.

[0080] The present invention can analyze a basketball player's jumping movements during exercise based on the player's two-dimensional skeleton, determine the player's landing risk based on the changes in the player's lower limbs during the jump, and then determine the degree of influence of other basketball players on the current player based on the influence of other basketball players on the current player. Based on this influence and the landing risk, the actual risk of the basketball player when performing the jumping movement is determined, thereby making the risk assessment of the basketball player's jumping movement more accurate. Therefore, the above-mentioned embodiments of the present invention achieve the purpose of assessing the movement risk of a basketball player.

[0081] Example 2:

[0082] Corresponding to the basketball player action recognition method using visual detection technology provided in the above embodiment, based on the same technical concept, an embodiment of the present invention also provides a basketball player action recognition system using visual detection technology. The basketball player action recognition system using visual detection technology is used to execute the basketball player action recognition method using visual detection technology. Figure 6 A schematic diagram of a basketball player action recognition system using visual detection technology is provided as another embodiment of the present invention. Figure 6 As shown. Basketball player action recognition systems using visual detection technology may have relatively large differences due to different configurations or performances, and may include one or more processors 601 and memory 602. The memory 602 is used to store computer programs that can be run on the processor 601. The processor 601 is used to execute the programs stored in the memory 602 to achieve the above Figure 1 The various steps in the method embodiment are described above. Memory 602 may be either a temporary or permanent storage device. The application stored in memory 602 may include one or more modules (not shown), each of which may include a series of computer-executable instructions for a basketball player motion recognition system using visual detection technology.

[0083] Furthermore, the processor 601 can be configured to communicate with the memory 602, and execute a series of computer-executable instructions in the memory 602 on the basketball player motion recognition system using visual detection technology. The basketball player motion recognition system using visual detection technology can also include one or more power supplies 603, one or more wired or wireless network interfaces 604, one or more input / output interfaces 605, and one or more keyboards 606.

[0084] Specifically in this embodiment, the basketball player action recognition system using visual detection technology includes a processor, a communication interface, a memory and a communication bus; wherein the processor, the communication interface and the memory communicate with each other through the bus; the memory is used to store computer programs; the processor is used to execute the programs stored in the memory to achieve the above Figure 1 The various steps in the method embodiment have the beneficial effects of the above method embodiments. To avoid repetition, the embodiments of the present invention will not be described again here.

[0085] It should be noted that the basketball player motion recognition system using visual detection technology provided in an embodiment of the present invention and the basketball player motion recognition method using visual detection technology provided in an embodiment of the present invention are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned basketball player motion recognition method using visual detection technology, and has the same or similar beneficial effects, and the repeated parts will not be repeated.

[0086] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0087] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A basketball player action recognition method using visual detection technology, characterized in that: include: extracting a two-dimensional skeleton of a basketball player from a plurality of consecutive images of a basketball area, and recognizing an action performed by the basketball player based on changes in the two-dimensional skeleton in each image; When recognizing that the basketball player is performing a jumping action, determining the landing risk of the basketball player after the jumping action according to changes in the lower limbs of the basketball player during the jumping action; determining, based on information about key points of the two-dimensional skeleton of the current basketball player and the two-dimensional skeletons of the other basketball players, the extent to which the current basketball player is influenced by the other basketball players during a jumping action; determining an actual risk of the current basketball player performing a jumping action based on an angle between a line connecting key points of the head and feet of the current basketball player and the ground when the player lands, the landing risk, and the impact level, and determining an action type of the current basketball player performing the jumping action based on the actual risk; Determining the landing risk of the basketball player after the jumping action according to the changes in the lower limbs of the basketball player during the jumping action includes: Determining a probability that the basketball player's lower limbs are performing a jumping action based on changes in the first knee angle and the thigh angle in the two-dimensional skeleton before the basketball player leaves the ground when performing a jumping action; and determining that the basketball player is performing a jumping action if the probability is greater than a preset threshold; determining the jumping distance, jumping height, and movement speed before taking off of the basketball player during the jumping process according to the two-dimensional skeleton of the basketball player in each of the images; determining the inertia of the basketball player after landing based on the jumping distance, the jumping height, and the movement speed; Determine the time difference between the landing of the lower limbs of the basketball player and the first distance between the landing points of the lower limbs according to the landing time and landing points of the two-dimensional skeleton of the basketball player in each of the images; determining synchronization of landing of the basketball player's lower limbs based on the time difference and the first distance; The landing risk of the basketball player after performing a jumping action is determined based on the inertia, the synchronization, and changes in the second knee angle and the hip joint angle in the two-dimensional skeleton after the basketball player lands after performing a jumping action.

2. The basketball player action recognition method using visual detection technology according to claim 1 is characterized in that: Determining the probability that the basketball player's lower limbs have taken off a jumping action based on changes in the first knee angle and the thigh angle in the two-dimensional skeleton before the basketball player leaves the ground when performing a jumping action includes: Fitting the change of the first knee angle of the two-dimensional skeleton before the basketball player leaves the ground during a jump to obtain a first fitting curve, and fitting the change of the thigh angle of the two-dimensional skeleton before the basketball player leaves the ground during a jump to obtain a second fitting curve; Recording the mean of the second-order derivative function of the first fitting curve as the knee jump degree, and obtaining a first average slope of the second fitting curve; A first ratio between the knee jump degree and the first average slope is calculated, and the first ratio is normalized to obtain the probability.

3. The basketball player action recognition method using visual detection technology according to claim 1 is characterized in that: The determining of the inertia of the basketball player after landing based on the jumping distance, the jumping height, and the movement speed includes: calculating a second ratio between the jump distance and the jump height, and calculating an absolute value of a first difference between the second ratio and a predetermined value; Calculating a first product between the absolute value of the first difference and the movement speed; Normalizing the first product to obtain the inertia.

4. The basketball player action recognition method using visual detection technology according to claim 1 is characterized in that: Determining the synchronization of the landing of the basketball player's lower limbs based on the time difference and the first distance includes: A second product between the time difference and the first distance is calculated, and an inverse proportional normalization process is performed on the second product to obtain the synchronization.

5. The basketball player action recognition method using visual detection technology according to claim 1 is characterized in that: Determining the landing risk of the basketball player after performing a jumping action based on the inertia, the synchronization, and changes in the second knee angle and the hip joint angle in the two-dimensional skeleton after the basketball player lands after performing a jumping action includes: Obtaining a third fitting curve of the second knee angle in the two-dimensional skeleton changing over time after the basketball player lands after performing a jump, and a fourth fitting curve of the hip joint angle in the two-dimensional skeleton changing over time after the basketball player lands after performing a jump; calculating a third product between the first average angle of the third fitting curve and the second average angle of the fourth fitting curve, and calculating a third ratio between the third product and the synchronicity; A fourth product between the third ratio and the inertia is calculated, and the fourth product is normalized to obtain the landing risk.

6. The basketball player action recognition method using visual detection technology according to any one of claims 1 to 5, characterized in that: Determining the degree to which the current basketball player is influenced by other basketball players during the jumping action based on the information of key points of the two-dimensional skeleton of the current basketball player and the two-dimensional skeletons of other basketball players includes: Taking the closest key point between the current basketball player and other adjacent basketball players as the affected key point of the current basketball player, and obtaining the closest distance between the affected key point and the other adjacent basketball players; Obtaining a second distance between the affected key point of the current basketball player and the torso skeleton line; Fitting the change of the closest distance over time to obtain a fifth fitting curve, and fitting the change of the second distance between the affected key point of the current basketball player and the torso skeleton line over time to obtain a sixth fitting curve; Calculating an average value between the second average slope of the fifth fitting curve and the third average slope of the sixth fitting curve, and a mean square error between the fifth fitting curve and the sixth fitting curve; A fifth product between the mean square error and the average value is calculated, and an inverse proportional normalization process is performed on the fifth product to obtain the influence degree.

7. The method for identifying basketball player movements using visual detection technology according to any one of claims 1 to 5, characterized in that: Determining the actual risk of the current basketball player in performing a jumping action based on the angle between the line connecting the key points of the head and the key points of the feet of the current basketball player and the ground when the player lands, the landing risk, and the impact degree includes: Fitting the change over time of the angle between the line connecting the key points of the head and the foot of the current basketball player and the ground when the player lands, to obtain a sixth fitting curve; Obtaining a fourth average slope of the sixth fitting curve; calculating a fourth ratio between the impact level and the fourth average slope, and calculating a sixth product between the fourth ratio and the landing risk; The sixth product is normalized to obtain the actual risk.

8. The basketball player action recognition method using visual detection technology according to claim 1 is characterized in that: Determining the action type of the current basketball player's jumping action based on the actual risk includes: When the actual risk is less than a first threshold, determining that the action type of the current jumping action performed by the basketball player is a low-risk jumping type; When the actual risk is greater than or equal to the first threshold and less than a second threshold, determining that the action type of the current jumping action performed by the basketball player is a medium-risk jumping type; When the actual risk is greater than or equal to the second threshold, it is determined that the action type of the current jumping action performed by the basketball player is a high-risk jumping type.

9. A basketball player action recognition system using visual detection technology, characterized in that: include: A processor and a memory; wherein the memory is used to store a computer program that can be run on the processor; The processor is configured to execute a program stored in the memory to implement the steps of the method for identifying basketball player movements using visual detection technology as described in any one of claims 1 to 8.

Images