Basketball action recognition method, electronic device and storage medium

Abstract

Embodiments of the present application provide a basketball action recognition method, an electronic device and a storage medium, and belong to the technical field of sports analysis. The method comprises: collecting target motion data flow of a target player through a motion tile screen pre-laid in a target space region; obtaining a motion data sequence based on the target motion data flow; wherein the motion data sequence comprises a plurality of basic action data, and a trigger time sequence of the plurality of basic action data is represented as time correlation information; performing action feature detection based on the basic action data of the motion data sequence to obtain a basic action label; performing action mode matching with a plurality of action semantic models based on the plurality of basic action labels and the time correlation information, and outputting advanced action recognition information based on a matching result. The embodiments of the present application can improve the accuracy of basketball complex action recognition.

Description

Technical Field

[0001] This application relates to the field of motion analysis technology, and in particular to a basketball motion recognition method, electronic device, and storage medium. Background Technology

[0002] In sports training, competitive events, and motion analysis, multimodal sensors can be used to collect athletes' motion data, and this data can be analyzed to identify athletes' movements. Related technologies use this motion data to train deep neural network models, which then output predefined motion labels based on the raw input data, thus completing motion classification and recognition. However, neural network models largely rely on training samples. Once athletes' movements are distorted due to defensive confrontations, variations in personal style, or the appearance of unseen non-standard movements, the model's recognition accuracy drops significantly, and it cannot provide a basis for judgment. Furthermore, because the internal decision-making mechanism of neural network models is opaque, it cannot provide a physical basis for judgment, resulting in serious shortcomings in critical application scenarios requiring high reliability, such as automated referee assistance and high-level competitive review.

[0003] Therefore, how to establish a mapping between physical motion data and basketball tactical behavior, and more accurately identify and analyze a wide range of basketball movements, has become an urgent technical problem to be solved. Summary of the Invention

[0004] The main objective of this application is to propose a basketball motion recognition method, electronic device, and storage medium, which aims to improve the accuracy of complex basketball motion recognition.

[0005] To achieve the above objectives, a first aspect of this application proposes a basketball action recognition method, the method comprising: The target athlete's motion data stream is collected by pre-laying sports floor tiles in the target space area; A motion data sequence is obtained by parsing the target motion data stream; wherein, the motion data sequence contains multiple basic motion data, and the triggering sequence of the multiple basic motion data is represented as time correlation information; Based on the basic motion data of the motion data sequence, motion feature detection is performed to obtain basic motion labels; Based on the multiple basic action tags and the time association information, action pattern matching is performed with multiple pre-configured action semantic models, and advanced action recognition information is output based on the matching results.

[0006] In some embodiments, before performing action pattern matching with multiple pre-configured action semantic models based on multiple basic action tags and the time association information, and outputting advanced action recognition information based on the matching results, the method further includes pre-configuring the action semantic models, specifically including: Obtain an advanced motion dataset; wherein, the advanced motion dataset is configured with multiple motion execution data, and each motion execution data corresponds to a basketball motion type; Physical action description is performed on each of the action execution data of the same basketball action type to obtain corresponding physical description information; Based on the physical description information, the action is decomposed to obtain multiple consecutive action sub-events; wherein, the multiple consecutive action sub-events are used to characterize the action execution process; Action label mapping is performed based on multiple consecutive action sub-events to obtain action feature labels corresponding to each action sub-event; wherein, each action feature label corresponds to a specific physical implementation condition; Based on the physical implementation conditions corresponding to multiple action feature labels and multiple consecutive action sub-events, parameter association processing is performed to construct the action semantic model corresponding to the basketball action type.

[0007] In some embodiments, the action semantic model is configured with timing constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds; The step of constructing an action semantic model corresponding to the basketball action type by performing parameter association processing based on the physical implementation conditions corresponding to multiple action feature labels and multiple consecutive action sub-events includes: Configure the timing constraint parameters based on the sequential occurrence order of multiple consecutive action sub-events and the maximum allowed time window; Based on the spatial motion trajectories and vector angles of multiple consecutive action sub-events during the action execution process, configure the spatial geometric constraint parameters; Based on the mechanical critical value, acceleration critical value, and displacement critical value contained in the physical implementation conditions corresponding to each of the aforementioned action feature tags, the physical quantity threshold is configured; The temporal constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds are respectively associated and mapped with the corresponding action feature labels and action sub-events to construct the action semantic model corresponding to the basketball action type.

[0008] In some embodiments, each of the action semantic models is configured with timing constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds; The process involves matching action patterns based on multiple basic action tags and time-related information with multiple pre-configured action semantic models, and outputting advanced action recognition information based on the matching results, including: Based on the time correlation information and multiple time series constraint parameters, time series matching is performed to obtain the time series matching result; Spatial trajectory analysis is performed based on the basic action data corresponding to multiple basic action tags to obtain spatial feature data. Spatial range matching is then performed based on the spatial feature data and multiple spatial geometric constraint parameters to obtain spatial matching results. Mechanical features are extracted based on the basic motion data corresponding to multiple basic motion tags to obtain motion physical quantity data. Then, the motion physical quantity data is compared with multiple physical quantity thresholds to obtain physical quantity matching results. Based on the temporal matching result, the spatial matching result, and the physical quantity matching result, a target action model is obtained by matching from multiple action semantic models, and the advanced action recognition information is output based on the target action model.

[0009] In some embodiments, the step of performing physical motion description based on the execution data of each of the same basketball motion types to obtain corresponding physical description information includes: Based on the action execution data, the motion subject is analyzed to obtain the action execution subject; Based on the action execution data, action type analysis is performed to obtain action type information; Based on the action execution data, action intent is identified to determine action intent information; Based on the action execution subject, the action type information, and the action intent information, a text structure template is generated to obtain the physical description information corresponding to the action execution data.

[0010] In some embodiments, the step of performing action pattern matching with multiple pre-configured action semantic models based on multiple basic action tags and the time association information, and outputting advanced action recognition information based on the matching results, includes: In response to obtaining a target action model by matching multiple action semantic models, the identity information, action execution timestamp, and action location data of the target athlete corresponding to the target action model are extracted; The system retrieves action corpus information that matches the target action model from a preset basketball action corpus; wherein the basketball action corpus is configured with vocabulary information specifically for basketball. Based on the target athlete's identity information, the action execution timestamp, the action location data, and the action corpus information, structured text is generated, and the advanced action recognition information is output.

[0011] In some embodiments, the step of matching a target action model from multiple action semantic models based on the temporal matching result, the spatial matching result, and the physical quantity matching result, and outputting the advanced action recognition information based on the target action model, further includes: In response to the fact that the spatial matching result or the physical quantity matching result does not completely match any of the action semantic models, a parameter deviation value is obtained by calculating the difference between the spatial matching result and the physical quantity matching result and the spatial geometric constraint parameter or the physical quantity threshold. Based on the parameter deviation value meeting the preset tolerance range, and based on the spatial matching result and the physical quantity matching result, the action semantic model with the highest similarity is matched and determined as the target action model.

[0012] To achieve the above objectives, a second aspect of this application proposes a basketball motion recognition system, the system comprising: The motion data acquisition module is used to collect the target motion data stream of the target athlete by using a motion tile screen that is pre-laid in the target space area; The motion data parsing module is used to parse the target motion data stream to obtain a motion data sequence; wherein, the motion data sequence contains multiple basic motion data, and the triggering sequence of the multiple basic motion data is represented as time correlation information; An action label matching module is used to perform action feature detection based on the basic action data of the motion data sequence to obtain basic action labels; The basketball motion recognition module is used to perform motion pattern matching with multiple pre-configured motion semantic models based on multiple basic motion tags and time association information, and output advanced motion recognition information based on the matching results.

[0013] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the basketball motion recognition method described in the first aspect.

[0014] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the basketball motion recognition method described in the first aspect.

[0015] The basketball motion recognition method, electronic device, and storage medium proposed in this application collect target motion data streams of target athletes by pre-laying sports floor tiles in the target spatial area; parse the target motion data streams to obtain motion data sequences; wherein, the motion data sequences contain multiple basic motion data, and the triggering time sequence of the multiple basic motion data is represented as time correlation information; perform motion feature detection based on the basic motion data of the motion data sequences to obtain basic motion labels; perform motion pattern matching with multiple pre-configured motion semantic models based on multiple basic motion labels and time correlation information, and output advanced motion recognition information based on the matching results. Therefore, this application obtains a motion data sequence by parsing the target motion data stream, extracts continuous motion data on the time axis to obtain a motion data sequence, and sets corresponding basic action labels for the continuously triggered basic action data in the motion data sequence. Different athletes may have differences when performing the same complex basketball action, but the basic action type is the same. Therefore, obtaining basic action labels with definite meaning can eliminate the interference of individual differences among athletes and the differences caused by the deformation of the opponent to a certain extent. Since complex basketball actions are often composed of multiple short-term, highly dynamic basic behaviors, feature matching can be performed based on multiple consecutive basic action labels and time-related information, and action pattern matching can be performed with multiple pre-configured action semantic models. Each action semantic model corresponds to a complex basketball action. Through this temporal logic matching mechanism, accurate identification of complex basketball actions can be achieved. Attached Figure Description

[0016] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart illustrating a basketball motion recognition method according to an embodiment of this application; Figure 2 This is another flowchart illustrating the basketball motion recognition method according to an embodiment of this application; Figure 3 This is another flowchart illustrating the basketball motion recognition method according to an embodiment of this application; Figure 4 This is another flowchart illustrating the basketball motion recognition method according to an embodiment of this application; Figure 5 This is another flowchart illustrating the basketball motion recognition method according to an embodiment of this application; Figure 6 This is another flowchart illustrating the basketball motion recognition method according to an embodiment of this application; Figure 7 This is another flowchart illustrating the basketball motion recognition method according to an embodiment of this application; Figure 8This is another flowchart illustrating the basketball motion recognition method according to an embodiment of this application; Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0019] Unless otherwise defined, 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 application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] In the fields of basketball sports analysis and intelligent sports training, how to use the collected sports data to perform timely and standardized automated analysis of on-court actions and tactical events has become a research hotspot in the fields of sports technology and data mining.

[0021] Currently, the relevant technologies for motion analysis and event recognition in basketball mainly rely on the following two types of technical means: The first category is purely basic data collection and recording, which only involves recording the athlete's three-dimensional coordinates, movement speed, acceleration, and plantar pressure distribution on the field. Then, the data is reviewed and manually annotated by humans afterward by comparing it with the video.

[0022] The second category is pattern recognition methods based on traditional machine learning or deep learning. These methods typically acquire massive amounts of motion images or sensor datasets with full manual annotations as training samples, and use neural network models to perform statistical fitting and probability classification on predefined standard basketball motion features.

[0023] However, the aforementioned existing technologies have the following fundamental technical shortcomings when facing complex basketball scenarios: First, deep learning-based black-box recognition methods are highly dependent on the completeness of the training samples. In actual basketball games, athletes' movements are often severely distorted due to defensive confrontations and physical exertion, and different athletes have highly individualized variations of their movements. When faced with these non-standard deformed movements or new variations not seen in the training set, the recognition accuracy and stability of traditional models will plummet. Furthermore, because their internal decision-making mechanisms are opaque and cannot provide physical evidence for their judgments, they are severely inadequate in critical application scenarios requiring high reliability, such as automated referee assistance and high-level game review.

[0024] Secondly, there is a semantic difference between low-level physical sensor data and the semantics of basketball tactics. The signals collected by sensors are generally expressed as low-order physical quantities such as "a sudden change in pressure value within a certain detection time slot" or "the trajectory of a point in space," while basketball tactics and competition rules describe high-order semantics with high abstraction and contextual characteristics, such as "a crossover dribble was executed" or "a defensive three-second violation." Current technology lacks a bridge that can transform the domain knowledge of basketball experts into objective, computable logic, making it difficult for computers to directly understand the macroscopic tactical meaning behind the low-level physical signals.

[0025] Therefore, how to establish a mapping between physical motion data and basketball tactical behavior, and more accurately identify and analyze a wide range of basketball movements, has become an urgent technical problem to be solved.

[0026] Based on this, embodiments of this application provide a basketball motion recognition method, electronic device, and storage medium, aiming to improve the accuracy of complex basketball motion recognition.

[0027] The basketball motion recognition method, electronic device, and storage medium provided in this application are specifically described through the following embodiments. First, the basketball motion recognition method in this application embodiment is described.

[0028] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0029] Figure 1This is an optional flowchart illustrating the basketball motion recognition method provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S104.

[0030] Step S101: Collect the target athlete's target motion data stream by pre-laying a sports floor tile screen in the target space area; Step S102: Obtain a motion data sequence based on the target motion data stream; wherein, the motion data sequence contains multiple basic motion data, and the triggering time sequence of the multiple basic motion data is represented as time correlation information; Step S103: Perform motion feature detection based on the basic motion data of the motion data sequence to obtain basic motion labels; Step S104: Based on multiple basic action labels and time association information, perform action pattern matching with multiple pre-configured action semantic models, and output advanced action recognition information based on the matching results.

[0031] Steps S101 to S104 as shown in the embodiments of this application involve collecting target motion data streams of a target athlete by pre-laying motion tile screens in the target space area; parsing the target motion data streams to obtain motion data sequences; wherein, the motion data sequences contain multiple basic motion data, and the triggering sequence of the multiple basic motion data is represented as time association information; motion feature detection is performed on the basic motion data of the motion data sequences to obtain basic motion labels; based on the multiple basic motion labels and time association information, motion pattern matching is performed with multiple pre-configured motion semantic models, and advanced motion recognition information is output based on the matching results. Therefore, this application obtains a motion data sequence by parsing the target motion data stream, extracts continuous motion data on the time axis to obtain a motion data sequence, and sets corresponding basic action labels for the continuously triggered basic action data in the motion data sequence. Different athletes may have differences when performing the same complex basketball action, but the basic action type is the same. Therefore, obtaining basic action labels with definite meaning can eliminate the interference of individual differences among athletes and the differences caused by the deformation of the opponent to a certain extent. Since complex basketball actions are often composed of multiple short-term, highly dynamic basic behaviors, feature matching can be performed based on multiple consecutive basic action labels and time-related information, and action pattern matching can be performed with multiple pre-configured action semantic models. Each action semantic model corresponds to a complex basketball action. Through this temporal logic matching mechanism, accurate identification of complex basketball actions can be achieved.

[0032] In step S101 of some embodiments, the target athlete's target motion data stream is collected using a sports floor tile screen pre-laid in the target space area.

[0033] The sports floor tile screen is equipped with an array of LED beads, optical sensors, and pressure sensors. The optical sensors emit detection light through the sports floor tile screen towards the target area. During the movement of multiple athletes, the athletes modulate the detection light to generate feedback light that is reflected back to the optical sensors, forming optical sensing signals. During each athlete's movement, the pressure generated by the sports floor tile screen is transmitted to the pressure sensors, forming pressure sensing signals. Based on the optical and pressure sensing signals, the posture of each athlete is identified to determine the real-time target motion data stream corresponding to each athlete. The LED bead display array can be used to provide athletes with feedback motion data or motion instruction information.

[0034] Optical sensors can be densely packed arrays of infrared photocells with centimeter-level spacing, acquiring precise two-dimensional coordinates, velocity vectors, and movement trajectories of the athlete's feet, the ball, and opponents. Pressure sensors can also be arranged in arrays to acquire mechanical characteristic data such as the athlete's plantar pressure distribution, pressure center trajectory, and impact force timing. Ultimately, this allows for the acquisition of movement trajectory data for each athlete within a target spatial area. This movement trajectory data typically includes core parameters such as each athlete's three-dimensional or two-dimensional physical coordinates, instantaneous velocity vectors, and acceleration over a continuous time series.

[0035] It's important to note that the optical sensor is a component inside the sports floor tile screen used to generate light signals. The optical sensor emits probe light towards the target area, and this light penetrates the surface material of the sports floor tile screen to enter the playing space. Its function is to establish an active optical detection mechanism, outputting light signals from the field itself to provide a stable lighting or structured light foundation for subsequent acquisition of athlete position information. Using an internal light source, rather than relying on ambient light, helps reduce the interference of changes in external lighting conditions on data acquisition quality.

[0036] When athletes move on the court surface, their footsteps exert pressure on the sports tile screen. This pressure is transmitted through the screen structure to pressure sensors deployed inside, generating pressure sensing signals. These pressure sensors detect the direct contact between the athlete and the court surface. This allows for the acquisition of physical interaction information between the athlete and the court, recording mechanical characteristics such as foot placement and movement rhythm. The design is based on the fact that optical signals can be interfered with by body posture or occlusion, while pressure signals directly reflect the athlete's contact with the court. These two sensing modes complement each other, improving the reliability of position determination.

[0037] In team sports, traditional external camera acquisition mechanisms are easily affected by fluctuations in ambient lighting, overlapping players, and visual obstruction caused by tight formations, making it impossible to accurately locate the positions of each athlete and consequently, to accurately assess the range and intensity of each athlete's influence. However, the sports floor tile screen provided in this application, through its hardware configuration mechanism that integrates active light sensing and mechanical perception, not only overcomes the limitations of traditional external visual capture technologies in terms of perception distortion and coordinate fragmentation under high-density personnel obstruction and variable lighting conditions, but also effectively eliminates perception blind spots and measurement errors by transforming raw heterogeneous sensor signals into structured, highly expressive motion trajectory data. This provides more accurate underlying data for analyzing the types of athlete movements.

[0038] In step S102 of some embodiments, a motion data sequence is obtained by parsing the target motion data stream. The parsed motion data sequence contains multiple basic motion data, and the triggering sequence of the multiple basic motion data is represented as time correlation information. A motion data sequence refers to a queue of motion events arranged in chronological order, and basic motion data refers to micro-motion segments obtained by segmenting a continuous queue of motion events. The time correlation information is used to characterize the chronological order, duration, and time window of these motion segments on the time axis.

[0039] In a specific embodiment, a typical complex movement in basketball, the step-back jump shot, will be used as an example for detailed explanation. The motion data sequence is obtained by parsing the target motion data stream, which contains multiple basic motion data, including the forward push-off and step-back motion, the rear foot braking motion, the two-foot charging motion, and the jump motion. The triggering order of these basic motion data on the timeline, the duration of each segment, and the maximum allowable time interval between segments—these key timing parameters collectively constitute the temporal correlation information.

[0040] In practical applications, since complex basketball movements are often composed of multiple highly dynamic and instantaneous basic behaviors, if pattern recognition is performed directly on a continuous stream of motion data, traditional algorithms are prone to causing the entire waveform to become disordered due to changes in the athlete's personal style and deformation of movements during competition, thus failing to accurately determine the type of movement.

[0041] The technical solution in this step uses time-related information to break down complex basketball movements into a set of well-structured and interconnected basic movement data sequences in the time dimension. This means that in subsequent steps, instead of dealing with a chaotic whole data stream, it is only necessary to verify the segmented, feature-clear independent segments one by one. This significantly reduces the algorithmic complexity of processing complex continuous data and lays a practical foundation for achieving high-accuracy movement pattern matching.

[0042] In step S103 of some embodiments, motion feature detection is performed on the basic motion data of the motion data sequence to obtain basic motion labels. Multiple micro-motion segments in the motion data sequence are labeled with motion tags to obtain basic motion labels. Basic motion labels refer to the classification identification information assigned to the corresponding basic motion data after motion feature detection. These labels can be recorded in the form of characters, codes, or vectors, and are used to clearly characterize the specific motion type corresponding to the basic motion data.

[0043] In step S104 of some embodiments, based on multiple basic action tags and temporal association information, action pattern matching is performed with multiple pre-configured action semantic models, and advanced action recognition information is output based on the matching results. The pre-configured action semantic model refers to an action judgment model pre-stored in a knowledge base, organized by multiple standard feature tags according to strict temporal causal logic and spatiotemporal threshold constraints. Each action semantic model defines a specific complex basketball action and corresponds to basic action tag combination rules, temporal constraints, and logical relationships. Advanced action recognition information refers to action event descriptions with clear basketball tactical action semantics, which may include descriptions of time, spatial location, and dominant limb.

[0044] Please see Figure 2 In some embodiments, step S104 may include, but is not limited to, steps S201 to S205: Step S201: Obtain the advanced motion dataset; wherein, the advanced motion dataset is configured with multiple motion execution data, and each motion execution data corresponds to a basketball motion type; Step S202: Perform physical action description based on the execution data of each action of the same basketball action type to obtain the corresponding physical description information; Step S203: Based on the physical description information, the action is decomposed to obtain multiple consecutive action sub-events; wherein, the multiple consecutive action sub-events are used to characterize the action execution process; Step S204: Perform action label mapping processing based on multiple consecutive action sub-events to obtain action feature labels corresponding to each action sub-event; wherein, each action feature label corresponds to a specific physical implementation condition. Step S205: Based on the physical realization conditions corresponding to multiple action feature labels and multiple consecutive action sub-events, parameter association processing is performed to construct an action semantic model corresponding to the basketball action type.

[0045] In step S201 of some embodiments, an advanced action dataset is obtained, wherein the advanced action dataset is configured with multiple action execution data, each action execution data corresponding to a type of basketball action. This advanced action dataset refers to a basketball tactical action material library stored in a knowledge base for model building, which stores multiple action execution data. Action execution data refers to the data record of one action execution process of a specific basketball action type, such as a step-back jump shot, a crossover dribble, or a double screen cut.

[0046] In step S202 of some embodiments, a physical action description is performed based on the execution data of each action of the same basketball action type to obtain corresponding physical description information. The physical description information is structured information describing the action using strict physical language and biomechanical terminology.

[0047] Please see Figure 3 In some embodiments, step S202 may include, but is not limited to, steps S301 to S304: Step S301: Perform motion subject analysis based on motion execution data to obtain the motion execution subject; Step S302: Perform action type analysis based on action execution data to obtain action type information; Step S303: Based on the action execution data, identify the action intent and determine the action intent information; Step S304: Based on the action execution subject, action type information and action intention information, a text structure template is generated to obtain the physical description information of the corresponding action execution data.

[0048] In step S301 of some embodiments, motion subject analysis is performed based on motion execution data to obtain the motion execution subject. The specific limbs or body parts that perform the key physical operations of the motion can be identified and extracted from the motion execution data. By analyzing the sensor signal characteristics, kinematic response amplitude, or force distribution pattern corresponding to each limb part, the physical execution unit that plays a dominant role in the motion is determined. The motion execution subject refers to the identification information obtained after motion subject analysis, representing the specific limbs or body parts that perform the key physical operations of the motion. For example, determining that the crossover step breakthrough motion is performed by the right foot as the pivot foot performing a lateral push-off, or determining that the shooting motion is performed by the right hand performing a flick and release.

[0049] In step S302 of some embodiments, motion type analysis is performed based on motion execution data to obtain motion type information. Using preset classification rules or pattern recognition algorithms, motion pattern features in the motion execution data are judged and classified to determine the motion category corresponding to the data record. Motion type information refers to the classification identifier information output after motion type analysis, used to identify the motion type to which the motion execution data belongs.

[0050] In step S303 of some embodiments, action intent is identified based on action execution data to determine action intent information. This process can be based on kinematic features, force characteristics, and contextual information in the action execution data, combined with basketball tactical rules and principles of sports biomechanics, to infer the tactical purpose or physical goal pursued by the action. Action intent information refers to semantic information characterizing the tactical intent of the action execution. For example, identifying the execution intent of a crossover step-through as using lateral acceleration to get rid of the defender, or identifying the execution intent of a step-back jump shot as creating space backward and transitioning to a shooting posture.

[0051] In step S304 of some embodiments, a text structure template is generated based on the action execution subject, action type information, and action intent information to obtain the physical description information of the corresponding action execution data. The text structure template refers to a text template that formats, semantically associates, and generates text based on a preset physical description template. The text structure template specifies the grammatical structure and element arrangement rules of the physical description. The physical description information refers to the information generated by the text structure template that can completely characterize the essential physical features and tactical intent of the basketball action type.

[0052] Through steps S301 to S304, unstructured, purely physical motion execution data is transformed into a structured semantic carrier containing limb parts, motion categories, and tactical intentions. This provides standardized and clear descriptive content for subsequent motion decomposition and feature label mapping based on physical description information, significantly improving the accuracy and completeness of basketball motion semantic model construction.

[0053] In step S203 of some embodiments, the action is decomposed based on physical description information to obtain multiple consecutive action sub-events, which are used to characterize the action execution process. An action sub-event refers to a data segment or logical unit that corresponds to a specific physical causal link in the action execution process after decomposition. Multiple consecutive action sub-events can be combined in sequence to completely characterize the action execution process of this basketball action type.

[0054] In a specific embodiment, based on the physical description information of the cross step breakthrough, the action can be decomposed into multiple consecutive action sub-events such as the lateral force pulse sub-event, the centroid lateral acceleration sub-event, the direction consistency verification sub-event, and the effective displacement sub-event. This makes the complete cross step breakthrough action decomposed into a physical causal chain with strict causal relationship.

[0055] In step S204 of some embodiments, action label mapping is performed based on multiple consecutive action sub-events to obtain action feature labels corresponding to each action sub-event. Each action feature label corresponds to a specific physical implementation condition. An action feature label is a symbolic identifier assigned to an action sub-event after mapping, used to identify the category of measurable physical features to which the sub-event belongs. For example, a "lateral force pulse sub-event" is mapped to a "lateral force pulse label," and a "center of mass lateral acceleration sub-event" is mapped to a "center of mass lateral acceleration label." A physical implementation condition refers to the quantitative standard by which the measurable physical feature represented by the action feature label can be detected and judged at the data level. For example, the physical implementation condition corresponding to the "lateral force pulse label" is "the lateral component of the pressure center of the left / right foot increases by more than 300% of the baseline within 100 milliseconds."

[0056] In step S205 of some embodiments, parameter association processing is performed based on the physical realization conditions corresponding to multiple action feature labels and multiple consecutive action sub-events to construct an action semantic model corresponding to the basketball action type. The action semantic model refers to a structured knowledge template that can completely describe a specific basketball action type after parameter association processing. It integrates the action feature label sequence, the physical realization conditions corresponding to each label, and the spatiotemporal constraint parameters of the action sub-events.

[0057] Please see Figure 4 In some embodiments, the action semantic model is configured with time-series constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds. Step S205 may include, but is not limited to, steps S401 to S404: Step S401: Configure timing constraint parameters based on the order of occurrence of multiple consecutive action sub-events and the maximum allowed time window; Step S402: Configure spatial geometric constraint parameters based on the spatial motion trajectories and vector angles of multiple consecutive action sub-events during the action execution process; Step S403: Configure physical quantity thresholds based on the mechanical critical values, acceleration critical values, and displacement critical values ​​contained in the physical realization conditions corresponding to each action feature label; Step S404: The temporal constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds are associated and mapped with the corresponding action feature labels and action sub-events, respectively, to construct an action semantic model corresponding to the basketball action type.

[0058] In step S401 of some embodiments, timing constraint parameters are configured based on the sequential order of occurrence of multiple consecutive action sub-events and the maximum allowed time window. Timing constraint parameters refer to a set of parameters used to limit the temporal sequence of action sub-events and the maximum allowed time interval between adjacent action sub-events. The sequential order of occurrence refers to the time sequence in which action sub-events must be triggered according to physical causal logic. For example, in a crossover step breakthrough action, the lateral force pulse sub-event, as the starting point of force generation, must occur before the lateral acceleration sub-event at the center of mass.

[0059] In step S402 of some embodiments, spatial geometric constraint parameters are configured based on the spatial motion trajectories and vector angles of multiple consecutive action sub-events during the execution of the action. The spatial motion trajectory refers to the positional change path formed by the action sub-event within the target spatial region during execution. The vector angle refers to the directional angle between the physical quantity vectors corresponding to different action sub-events, such as the angle between the force direction vector and the center of mass movement direction vector. Spatial geometric constraint parameters are quantified parameters used to limit the range of geometric conditions that the aforementioned spatial motion trajectories and vector angles must satisfy. For example, in the construction of the semantic model for the cross-step breakthrough action, based on the direction of the ground reaction force generated by the lateral force pulse sub-event and the center of mass movement direction formed by the lateral acceleration sub-event, a spatial geometric constraint parameter is configured to ensure that the force direction and the body movement direction are consistent, thereby eliminating pseudo-actions where the force direction and movement direction contradict each other.

[0060] In step S403 of some embodiments, physical quantity thresholds are configured based on the mechanical critical values, acceleration critical values, and displacement critical values ​​included in the physical implementation conditions corresponding to each action feature tag. The mechanical critical value refers to the minimum or maximum limit value that the physical quantity of force or torque involved in the action execution must reach. The displacement critical value refers to the minimum or maximum displacement that the center of mass or a specific part must generate within a specific time during the action execution. The acceleration critical value refers to the minimum or maximum limit value that the acceleration generated by the center of mass or a specific limb part during movement must meet. The physical quantity threshold refers to the quantitative judgment standard formed after standardizing the above-mentioned critical values, used to determine whether an action feature tag is triggered.

[0061] In step S404 of some embodiments, the temporal constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds are associated and mapped with the corresponding action feature labels and action sub-events, respectively, to construct an action semantic model corresponding to the basketball action type. Here, it means binding and associating the configured temporal constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds with the corresponding action feature labels and action sub-events in the action semantic model to form a structured action template with complete judgment logic.

[0062] Through steps S401 to S404, by configuring temporal constraint parameters to limit the causal response time characteristics of action sub-events, configuring spatial geometric constraint parameters to limit the directional consistency and spatial positional relationship during action execution, and configuring physical quantity thresholds to clarify the detectable physical judgment criteria for each action feature label, and by associating and mapping the above multi-dimensional constraint parameters with the corresponding action feature labels and action sub-events, the constructed basketball action semantic model can form a rigorous judgment logic in the time dimension, spatial geometric dimension, and physical quantity dimension, thereby significantly improving the recognition accuracy when performing action pattern matching based on real-time data streams.

[0063] Through steps S201 to S205, basketball tactical knowledge is transformed into an objective, rigorous, and computable standardized logical model, enabling any basketball action to be deconstructed into a set of features composed of measurable physical quantities and organized according to strict temporal causal logic, thus providing a foundation for subsequent feature label detection and semantic model matching based on real-time data streams.

[0064] Please see Figure 5 In some embodiments, step S104 may include, but is not limited to, steps S501 to S504: Step S501: Perform time series matching based on time correlation information and multiple time series constraint parameters to obtain the time series matching result; Step S502: Perform spatial trajectory analysis based on the basic motion data corresponding to multiple basic motion labels to obtain spatial feature data, and perform spatial range matching based on the spatial feature data and multiple spatial geometric constraint parameters to obtain spatial matching results; Step S503: Based on the basic motion data corresponding to multiple basic motion tags, mechanical features are extracted to obtain motion physical quantity data. Then, based on the motion physical quantity data, critical values ​​are compared with multiple physical quantity thresholds to obtain physical quantity matching results. Step S504: Based on the temporal matching results, spatial matching results, and physical quantity matching results, the target action model is obtained by matching from multiple action semantic models, and advanced action recognition information is output based on the target action model.

[0065] In step S501 of some embodiments, timing matching is performed based on time association information and multiple timing constraint parameters to obtain a timing matching result. Time association information refers to the timing relationship data between basic action tags, such as timestamps, sequence, and time intervals. Multiple timing constraint parameters are quantization parameters pre-configured in each action semantic model to limit the maximum allowed time interval and strict sequence between adjacent action sub-events. The timing matching result is the judgment information output after comparing the triggering order and time interval, representing whether the real-time action sequence conforms to or does not conform to the timing constraints of a specific action semantic model in the time dimension.

[0066] In step S502 of some embodiments, spatial trajectory analysis is performed based on the basic action data corresponding to multiple basic action tags to obtain spatial feature data. Then, spatial range matching is performed between the spatial feature data and multiple spatial geometric constraint parameters to obtain a spatial matching result. Spatial feature data refers to data information obtained after spatial trajectory analysis that characterizes the spatial dimensional characteristics during action execution, such as the lateral movement trajectory of the center of mass, the force direction vector, or the displacement path curve. Spatial geometric constraint parameters refer to geometric condition parameters pre-configured in the action semantic model to limit the spatial relationships and directional consistency between action sub-events, such as the range of vector angles or spatial proximity requirements. The spatial matching result refers to the judgment information output after comparison, characterizing whether the real-time action sequence conforms to or does not conform to the geometric constraints of a specific action semantic model in the spatial dimension.

[0067] In step S503 of some embodiments, mechanical features are extracted based on the basic motion data corresponding to multiple basic motion tags to obtain motion physical quantity data. Then, the motion physical quantity data is compared with multiple physical quantity thresholds to obtain a physical quantity matching result. Motion physical quantity data refers to data information obtained through mechanical feature extraction that can quantify the physical load and motion intensity during motion execution, such as the lateral component growth factor of the plantar pressure center, the vertical acceleration amplitude, or the horizontal displacement distance of the center of mass. Physical quantity thresholds refer to various critical value parameters pre-configured in the motion semantic model used to determine whether a motion feature tag is triggered, including mechanical critical values, acceleration critical values, and displacement critical values. The physical quantity matching result refers to the judgment information output after comparison, representing whether the real-time motion sequence conforms to or does not conform to the threshold constraints of a specific motion semantic model in the physical quantity dimension.

[0068] In step S504 of some embodiments, a target action model is obtained by matching from multiple action semantic models based on temporal matching results, spatial matching results and physical quantity matching results, and advanced action recognition information is output based on the target action model.

[0069] The temporal matching result, spatial matching result, and physical quantity matching result refer to the compliance judgment information independently output by the aforementioned steps in the time dimension, spatial geometric dimension, and physical quantity dimension, respectively. The target action model refers to the action semantic model selected from multiple pre-configured action semantic models after multi-dimensional comprehensive judgment, and which is fully satisfied by the real-time action sequence in terms of temporal constraints, spatial geometric constraints, and physical quantity thresholds. Advanced action recognition information refers to the action event description with clear basketball tactical action semantics, ultimately output based on the target action model. For example, if the temporal matching result, spatial matching result, and physical quantity matching result of a real-time action sequence all satisfy all the constraints of the crossover step breakthrough action semantic model, then the action semantic model corresponding to the crossover step breakthrough is matched from multiple action semantic models as the target action model, and advanced action recognition information containing execution time, spatial position, action direction, and action subject information is output.

[0070] Through steps S501 to S504, temporal matching is performed to verify the causal response time characteristics of the action sub-events, spatial range matching is performed to verify the directional consistency and spatial position relationship during the action execution process, and critical value comparison is performed to verify whether the mechanical and kinematic physical quantities meet the model standards. The independent matching results of the three dimensions are jointly judged to determine the target action model. This ensures that the action recognition process meets the three objective constraints of time, space, and physical quantities, significantly improving the discrimination accuracy of complex basketball action recognition and effectively eliminating the influence of personal style differences, action deformation, or noise interference.

[0071] Please see Figure 6 In some embodiments, step S104 may also include, but is not limited to, steps S601 to S603: Step S601: In response to obtaining the target action model by matching from multiple action semantic models, extract the target athlete's identity information, action execution timestamp, and action position data corresponding to the target action model; Step S602: Obtain action corpus information matching the target action model from a preset basketball action corpus; wherein, the basketball action corpus is configured with vocabulary corpus information specifically for basketball. Step S603: Generate structured text based on the target athlete's identity information, action execution timestamp, action location data, and action corpus information, and output advanced action recognition information.

[0072] In step S601 of some embodiments, in response to obtaining a target action model by matching from multiple action semantic models, the identity information of the target athlete, the action execution timestamp, and the action location data corresponding to the target action model are extracted.

[0073] The target action model refers to a specific basketball action semantic model that is selected from multiple pre-configured action semantic models and determined to conform to the current real-time action sequence after multi-dimensional joint judgment including temporal matching, spatial matching, and physical quantity matching. The action execution timestamp refers to the system time stamp information that records the moment the target action model is triggered, represented in the form of an absolute or relative timestamp. Action position data refers to the spatial position information of the target athlete's center of mass or a specific body part in the basketball court coordinate system when the target action occurs.

[0074] In step S602 of some embodiments, action corpus information matching the target action model is obtained from a preset basketball action lexicon. The basketball action lexicon contains vocabulary information specifically for basketball. The preset basketball action lexicon refers to a pre-built and stored corpus database specifically for the basketball field, containing a collection of professional terms related to basketball tactics, technical actions, and rule descriptions. Action corpus information refers to standardized professional vocabulary or phrase descriptions retrieved from the basketball action lexicon that match the basketball action type corresponding to the target action model, such as basketball professional terms like "crossover dribble," "lateral push-off," and "pivot foot" corresponding to the crossover dribble action model.

[0075] In step S603 of some embodiments, structured text is generated based on the target athlete's identity information, action execution timestamp, action location data, and action corpus information, and advanced action recognition information is output. Structured text refers to the formatted organization, semantic splicing, and text encoding of multi-source heterogeneous information according to a preset text template or grammatical rules. Advanced action recognition information refers to the high-level basketball action event description information output after the structured text generation, containing complete semantic elements, including action category, executing entity, spatiotemporal context, and basketball terminology.

[0076] This step semantically integrates the judgment results of the target action model with professional basketball corpus and multi-dimensional contextual information, so that the final output action recognition information has a standardized, understandable text form that conforms to the professional basketball expression norms.

[0077] Through steps S601 to S603, professional action corpus information matching the target action model is obtained from the preset basketball action lexicon, and structured text is generated based on the above multi-dimensional information to output advanced action recognition information. This makes the final output advanced action recognition information include the identity of the executing subject, precise spatiotemporal context, and semantic description of professional terms in the basketball field, which significantly improves the completeness, readability, and domain professionalism of the output information.

[0078] Please see Figure 7In some embodiments, a portion of accuracy may be sacrificed to increase the tolerance for recognizing complex basketball movements. Step S104 may also include, but is not limited to, steps S701 to S702: Step S701: In response to the fact that the spatial matching result or the physical quantity matching result does not completely match any action semantic model, the difference is calculated based on the spatial matching result and the physical quantity matching result, and the spatial geometric constraint parameters or physical quantity thresholds to obtain the parameter deviation value. Step S702: Based on the parameter deviation value meeting the preset tolerance range, and based on the spatial matching result and physical quantity matching result, the action semantic model with the highest similarity is matched and determined as the target action model.

[0079] In step S701 of some embodiments, in response to the spatial matching result or physical quantity matching result not fully matching any action semantic model, a difference calculation is performed based on the spatial matching result and physical quantity matching result, and the spatial geometric constraint parameter or physical quantity threshold, to obtain a parameter deviation value.

[0080] Spatial matching results refer to the judgment information output after comparing spatial feature data with spatial geometric constraint parameters, characterizing whether the actual spatial motion conforms to the geometric conditions of the model. Physical quantity matching results refer to the judgment information output after comparing the motion physical quantity data with the physical quantity threshold values, characterizing whether the actual physical quantity reaches the critical standard of the model.

[0081] In response to a situation where the spatial matching result or physical quantity matching result does not fully match any action semantic model, it means that the action sequence detected in real time fails to meet the constraints defined by the action semantic model in at least one dimension, including the temporal, spatial, or physical quantity dimensions. Therefore, the actual detected spatial feature data or motion physical quantity data is subtracted from the preset spatial geometric constraint parameters or physical quantity thresholds in the corresponding action semantic model to quantify the degree of deviation between the actual parameters and the model standard.

[0082] The parameter deviation value refers to the quantitative value obtained after difference calculation, which is used to characterize the degree of difference between the actual motion parameters and the theoretical standard of the motion semantic model. For example, the angle difference between the actual vector angle and the angle required by the model, or the numerical difference between the actual acceleration amplitude and the critical value of the model.

[0083] For example, when it is determined that the spatial matching result or physical quantity matching result of a real-time action sequence fails to fully meet the strict constraints of the cross-step breakthrough action semantic model, this step is triggered. The difference between the angle between the actual detected force direction and the center of mass movement direction and the spatial geometric constraint parameter of less than 30 degrees configured in the model is calculated, or the difference between the actual displacement and the displacement critical value of more than 0.7 meters configured in the model is calculated, so as to obtain the parameter deviation value of the action sequence relative to the standard given by the action semantic model.

[0084] In step S702 of some embodiments, based on the parameter deviation value meeting the preset tolerance range, and based on the spatial matching result and physical quantity matching result, the action semantic model with the highest similarity is matched and determined as the target action model.

[0085] The tolerance range refers to a pre-defined numerical range that allows for reasonable deviations of actual motion parameters from the standard motion semantic model. This range comprehensively considers factors such as differences in individual athlete styles, motion deformation under competitive conditions, and sensor data acquisition errors, and is used to maintain reasonable recognition flexibility when strict matching fails. The target motion model here refers to the specific motion semantic model that is finally determined after the above tolerance screening and similarity ranking, and is judged to be the closest to the current real-time motion sequence within the reasonable tolerance range.

[0086] For example, if the parameter deviation of a real-time action sequence relative to the action semantic model of the cross-step breakthrough is within a preset tolerance range, and the similarity is higher than that of other action semantic models, then the action semantic model of the cross-step breakthrough is identified as the target action model, thereby avoiding direct judgment as recognition failure due to action deformation caused by the athlete's personal style or slight confrontation interference.

[0087] By sacrificing some accuracy through steps S701 to S702, the motion recognition process no longer fails directly due to the failure to meet a strict threshold in a single dimension. Instead, it can perform flexible matching within a reasonable deviation range. This significantly improves the adaptability and robustness to differences in athletes' personal styles, motion deformation in competitive environments, and sensor noise interference, avoiding the missed recognition problems that may be caused by strict matching mechanisms.

[0088] Please see Figure 8 In some embodiments, the embodiments of this application further include, but are not limited to, steps S801 to S804: Step S801: The optical sensor emits detection light into the target space area through the moving floor tile screen; Step S802: During the movement of multiple target athletes, the detection light is modulated by the target athletes to generate feedback light that is reflected to the optical sensor, forming an optical sensing signal. Step S803: During the movement of each target athlete, the pressure generated by the sports floor tile screen is transmitted to the pressure sensor, forming a pressure sensing signal. Step S804: Based on optical sensing signals and pressure sensing signals, perform pose recognition on each target athlete to determine the real-time target motion data stream corresponding to each target athlete.

[0089] In step S801 of some embodiments, an optical sensor emits a detection light beam through the sports floor tile screen toward the target spatial area. An optical sensor is a sensing device deployed at a specific location on the sports field that can actively emit and receive light signals to sense the shape and position information of objects in space, such as a lidar, infrared transmitter / receiver module, or depth camera. The detection light beam is a light wave signal actively emitted by the optical sensor to detect the presence, outline, and position of objects in space. By using an optical sensor to emit detection light beams through the sports floor tile screen toward the target athlete's activity area, a spatial perception foundation based on active optics is established, providing a prerequisite for subsequently receiving the light signals reflected by the athlete.

[0090] In step S802 of some embodiments, during the movement of multiple target athletes, the target athletes modulate the detection light to generate feedback light that is reflected to the optical sensor, forming an optical sensing signal. Modulation here refers to the process by which the target athlete's body or limbs alter the physical properties of the detection light, such as its propagation path, phase, frequency, or intensity, including optical interactions such as blocking, reflection, scattering, or absorption. During the movement of multiple target athletes, their bodies modulate the detection light, forming feedback light carrying information about the athlete's spatial position and form. The optical sensor receives this feedback light and forms an optical sensing signal, thereby achieving non-contact optical acquisition of the athlete's spatial information.

[0091] In step S803 of some embodiments, during the movement of each target athlete, the pressure generated by the sports floor tile screen is transmitted to the pressure sensor, forming a pressure sensing signal. During the movement of each target athlete, the pressure generated by the contact between the sole of their foot and the sports floor tile screen is transmitted and captured by the pressure sensor, forming a pressure sensing signal, thereby realizing the direct measurement and recording of the athlete's foot biomechanical information.

[0092] In step S804 of some embodiments, pose recognition is performed on each target athlete based on optical sensing signals and pressure sensing signals to determine the real-time target motion data stream corresponding to each target athlete. Pose recognition refers to the process of jointly identifying and determining the body posture and spatial position of the target athlete by fusing the spatial position information reflected by optical sensing signals and the plantar biomechanical information reflected by pressure sensing signals, through a preset multi-source data fusion algorithm or solution model. The spatial contour and position coordinates of the athlete are obtained based on optical sensing signals, while the plantar pressure center and force state of the athlete are obtained based on pressure sensing signals. Pose recognition is performed on each target athlete through multi-source information fusion, thereby determining the real-time target motion data stream corresponding to each target athlete, providing the original data foundation for subsequent motion feature detection and semantic parsing.

[0093] Through steps S801 to S804, the complementary fusion of optical and mechanical dual-modal data effectively overcomes the problems of information loss and blind spots in complex basketball scenarios caused by a single sensor, improves the accuracy and robustness of pose recognition, and provides high-quality, multi-dimensional and spatiotemporally synchronized raw data support for subsequent motion feature detection, semantic model matching and advanced motion recognition based on motion data sequences, ensuring the integrity and reliability of data in the process of transforming raw physical signals into high-level motion semantics.

[0094] This application embodiment collects target motion data streams of target athletes by pre-laying sports floor tiles in the target space area; it then parses the target motion data streams to obtain motion data sequences; wherein, the motion data sequences contain multiple basic motion data, and the triggering time sequence of the multiple basic motion data is represented as time association information; it performs motion feature detection based on the basic motion data of the motion data sequences to obtain basic motion labels; based on the multiple basic motion labels and time association information, it performs motion pattern matching with multiple pre-configured motion semantic models, and outputs advanced motion recognition information based on the matching results. Therefore, this application obtains a motion data sequence by parsing the target motion data stream, extracts continuous motion data on the time axis to obtain a motion data sequence, and sets corresponding basic action labels for the continuously triggered basic action data in the motion data sequence. Different athletes may have differences when performing the same complex basketball action, but the basic action type is the same. Therefore, obtaining basic action labels with definite meaning can eliminate the interference of individual differences among athletes and the differences caused by the deformation of the opponent to a certain extent. Since complex basketball actions are often composed of multiple short-term, highly dynamic basic behaviors, feature matching can be performed based on multiple consecutive basic action labels and time-related information, and action pattern matching can be performed with multiple pre-configured action semantic models. Each action semantic model corresponds to a complex basketball action. Through this temporal logic matching mechanism, accurate identification of complex basketball actions can be achieved.

[0095] This application also provides a basketball motion recognition system that can implement the above-described basketball motion recognition method. The system includes: The motion data acquisition module is used to collect the target motion data stream of the target athlete by using a motion tile screen that is pre-laid in the target space area; The motion data parsing module is used to parse the target motion data stream to obtain a motion data sequence; the motion data sequence contains multiple basic motion data, and the triggering time sequence of the multiple basic motion data is represented as time correlation information; The action label matching module is used to detect action features based on basic action data of the action data sequence to obtain basic action labels; The basketball motion recognition module is used to match motion patterns with multiple pre-configured motion semantic models based on multiple basic motion labels and time-related information, and output advanced motion recognition information based on the matching results.

[0096] The specific implementation of this basketball motion recognition system is basically the same as the specific implementation of the basketball motion recognition method described above, and will not be repeated here.

[0097] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the basketball motion recognition method described above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0098] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 902 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 902 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and called and executed by the processor 901 using the basketball action recognition method of the embodiments of this application. The input / output interface 903 is used to implement information input and output; The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904); The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0099] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the basketball motion recognition method described above.

[0100] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0101] The basketball motion recognition method, electronic device, and storage medium provided in this application collect target motion data streams of target athletes by pre-laying sports floor tiles in a target space area; parse the target motion data streams to obtain motion data sequences; wherein, the motion data sequences contain multiple basic motion data, and the triggering time sequence of the multiple basic motion data is represented as time association information; perform motion feature detection based on the basic motion data of the motion data sequences to obtain basic motion labels; perform motion pattern matching with multiple pre-configured motion semantic models based on the multiple basic motion labels and time association information, and output advanced motion recognition information based on the matching results. Therefore, this application obtains a motion data sequence by parsing the target motion data stream, extracts continuous motion data on the time axis to obtain a motion data sequence, and sets corresponding basic action labels for the continuously triggered basic action data in the motion data sequence. Different athletes may have differences when performing the same complex basketball action, but the basic action type is the same. Therefore, obtaining basic action labels with definite meaning can eliminate the interference of individual differences among athletes and the differences caused by the deformation of the opponent to a certain extent. Since complex basketball actions are often composed of multiple short-term, highly dynamic basic behaviors, feature matching can be performed based on multiple consecutive basic action labels and time-related information, and action pattern matching can be performed with multiple pre-configured action semantic models. Each action semantic model corresponds to a complex basketball action. Through this temporal logic matching mechanism, accurate identification of complex basketball actions can be achieved.

[0102] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0103] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0104] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0105] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0106] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0107] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. The coupling or direct coupling or communication connection between the shown or discussed units may be through some interfaces, or indirect coupling or communication connection between the apparatus or units, and may be electrical, mechanical, or other forms.

[0109] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0110] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0111] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0112] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A basketball motion recognition method, characterized in that, The method includes: The target athlete's motion data stream is collected by pre-laying sports floor tiles in the target space area; A motion data sequence is obtained by parsing the target motion data stream; wherein, the motion data sequence contains multiple basic motion data, and the triggering sequence of the multiple basic motion data is represented as time correlation information; Based on the basic motion data of the motion data sequence, motion feature detection is performed to obtain basic motion labels; Based on the multiple basic action tags and the time association information, action pattern matching is performed with multiple pre-configured action semantic models, and advanced action recognition information is output based on the matching results.

2. The method according to claim 1, characterized in that, Before performing action pattern matching with multiple pre-configured action semantic models based on multiple basic action tags and the time association information, and outputting advanced action recognition information based on the matching results, the method further includes pre-configuring the action semantic models, specifically including: Obtain an advanced motion dataset; wherein, the advanced motion dataset is configured with multiple motion execution data, and each motion execution data corresponds to a basketball motion type; Physical action description is performed on each of the action execution data of the same basketball action type to obtain corresponding physical description information; Based on the physical description information, the action is decomposed to obtain multiple consecutive action sub-events; wherein, the multiple consecutive action sub-events are used to characterize the action execution process; Action label mapping is performed based on multiple consecutive action sub-events to obtain action feature labels corresponding to each action sub-event; wherein, each action feature label corresponds to a specific physical implementation condition; Based on the physical implementation conditions corresponding to multiple action feature labels and multiple consecutive action sub-events, parameter association processing is performed to construct the action semantic model corresponding to the basketball action type.

3. The method according to claim 2, characterized in that, The action semantic model is configured with time-series constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds; The step of constructing an action semantic model corresponding to the basketball action type by performing parameter association processing based on the physical implementation conditions corresponding to multiple action feature labels and multiple consecutive action sub-events includes: Configure the timing constraint parameters based on the sequential occurrence order of multiple consecutive action sub-events and the maximum allowed time window; Based on the spatial motion trajectories and vector angles of multiple consecutive action sub-events during the action execution process, configure the spatial geometric constraint parameters; Based on the mechanical critical value, acceleration critical value, and displacement critical value contained in the physical implementation conditions corresponding to each of the aforementioned action feature tags, the physical quantity threshold is configured; The temporal constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds are respectively associated and mapped with the corresponding action feature labels and action sub-events to construct the action semantic model corresponding to the basketball action type.

4. The method according to claim 2, characterized in that, Each of the aforementioned action semantic models is configured with temporal constraint parameters, spatial geometric constraint parameters, and physical quantity thresholds; The process involves matching action patterns based on multiple basic action tags and time-related information with multiple pre-configured action semantic models, and outputting advanced action recognition information based on the matching results, including: Based on the time correlation information and multiple time series constraint parameters, time series matching is performed to obtain the time series matching result; Spatial trajectory analysis is performed based on the basic action data corresponding to multiple basic action tags to obtain spatial feature data. Spatial range matching is then performed based on the spatial feature data and multiple spatial geometric constraint parameters to obtain spatial matching results. Mechanical features are extracted based on the basic motion data corresponding to multiple basic motion tags to obtain motion physical quantity data. Then, the motion physical quantity data is compared with multiple physical quantity thresholds to obtain physical quantity matching results. Based on the temporal matching result, the spatial matching result, and the physical quantity matching result, a target action model is obtained by matching from multiple action semantic models, and the advanced action recognition information is output based on the target action model.

5. The method according to claim 2, characterized in that, The physical action description is performed based on the execution data of each action of the same basketball action type to obtain the corresponding physical description information, including: Based on the action execution data, the motion subject is analyzed to obtain the action execution subject; Based on the action execution data, action type analysis is performed to obtain action type information; Based on the action execution data, action intent is identified to determine action intent information; Based on the action execution subject, the action type information, and the action intent information, a text structure template is generated to obtain the physical description information corresponding to the action execution data.

6. The method according to claim 1, characterized in that, The process involves matching action patterns based on multiple basic action tags and time-related information with multiple pre-configured action semantic models, and outputting advanced action recognition information based on the matching results, including: In response to obtaining a target action model by matching multiple action semantic models, the identity information, action execution timestamp, and action location data of the target athlete corresponding to the target action model are extracted; The system retrieves action corpus information that matches the target action model from a preset basketball action corpus; wherein the basketball action corpus is configured with vocabulary information specifically for basketball. Based on the target athlete's identity information, the action execution timestamp, the action location data, and the action corpus information, structured text is generated, and the advanced action recognition information is output.

7. The method according to claim 4, characterized in that, The step of matching a target action model from multiple action semantic models based on the temporal matching result, the spatial matching result, and the physical quantity matching result, and outputting the advanced action recognition information based on the target action model, further includes: In response to the fact that the spatial matching result or the physical quantity matching result does not completely match any of the action semantic models, a parameter deviation value is obtained by calculating the difference between the spatial matching result and the physical quantity matching result and the spatial geometric constraint parameter or the physical quantity threshold. Based on the parameter deviation value meeting the preset tolerance range, and based on the spatial matching result and the physical quantity matching result, the action semantic model with the highest similarity is matched and determined as the target action model.

8. The method according to claim 1, characterized in that, The sports floor tile screen is equipped with optical sensors and pressure sensors. The process of collecting target athlete movement data streams through the sports floor tile screen pre-laid in the target space area includes: The optical sensor emits detection light into the target space area through the moving floor tile screen; During the movement of multiple target athletes, the detection light is modulated by the target athletes to generate feedback light that is reflected to the optical sensor, forming an optical sensing signal; During the movement of each of the target athletes, the pressure generated by the sports floor tile screen is transmitted to the pressure sensor, forming a pressure sensing signal. Based on the optical sensing signal and the pressure sensing signal, pose recognition is performed on each of the target athletes to determine the real-time target motion data stream corresponding to each target athlete.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the basketball motion recognition method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the basketball motion recognition method according to any one of claims 1 to 8.

Images