Multi-sensor fused badminton rapid scoring method and system

Abstract

The invention discloses a multi-sensor fused badminton rapid scoring method and system, and relates to the technical field of intelligent perception and data processing, and the method comprises the steps: deploying a projection assembly at a target site, carrying out site projection, and executing multi-sensor cooperative collection based on a ball hitting period, and executing parallel self-attention recognition and scoring processing based on scoring correlation, judgment rules and dispute characteristics, generating a scoring result and displaying the scoring result on a platform interface, so as to solve the problems that in the prior art, a standardized site is difficult to deploy quickly, an intelligent penalty technical system under random site deployment is lacked, and the working efficiency is high. In order to solve the technical problem that the sports requirements under site limitation are difficult to meet, the invention provides the rapid badminton scoring method which is rapid in deployment, fuses various sensing information and has adaptive environment compensation and dispute ball three-dimensional verification capabilities, so that the sports requirements of random site deployment and high-precision automatic scoring are met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent sensing and data processing, in particular to a badminton rapid scoring method and system based on multi-sensor fusion. BACKGROUND

[0002] In the badminton game, the traditional field relies on fixed ground markings, which is limited to indoor venues or pre-set field use, greatly limiting the flexibility of the game scene. When playing outdoors or in non-standard fields, there are often disputes over out-of-bound determination due to the lack of clear boundaries, affecting the game experience and fairness.

[0003] At the same time, there is a lack of standardized scoring methods, especially in the process of non-standardized badminton games in non-standardized fields, which cannot achieve automatic and rapid judgment, and some intelligent devices can only detect single data such as ball speed, lacking support for core penalties such as out-of-bound and violation.

[0004] Therefore, there is an urgent need for a technology system that can be flexibly deployed, quickly generate a standard field, and realize intelligent judgment, to break through the field restrictions and solve the problem of game requirements and judgment in outdoor and non-standard fields. SUMMARY

[0005] The present application provides a badminton rapid scoring method and system based on multi-sensor fusion, which is used to solve the technical problems of the prior art, such as the difficulty in quickly deploying standardized fields, the lack of intelligent judgment technology system under random field deployment, and the difficulty in meeting the game requirements under field restrictions.

[0006] In view of the above problems, the present application provides a badminton rapid scoring method and system based on multi-sensor fusion.

[0007] In a first aspect, the present application provides a badminton rapid scoring method based on multi-sensor fusion, which comprises: deploying a projection component in a target field, and projecting a projection field in the target field through multi-lens linkage control, wherein the projection field has adaptive participation compensation based on the field terrain and environmental light and shadow; driving a sensing array to perform multi-sensor cooperative collection based on a ball hitting cycle to determine a sensing number array; developing an automatic scorer on a game platform, receiving the sensing number array for each ball hitting cycle, and performing parallel self-attention recognition and scoring processing based on scoring correlation, decision rules and controversial features to generate a scoring result and display it on the platform interface; if the controversial feature is a non-empty set, perform three-dimensional trajectory reconstruction assisted scoring verification of the ball movement loop.

[0008] In a second aspect, the application provides a badminton quick scoring system based on multi-sensor fusion, which comprises: a field projection module, which is used to deploy a projection assembly on a target field to project a projection field on the target field through multi-lens linkage control, wherein the projection field has adaptive reference control compensation based on the field topography and environmental light and shadow; a sensing acquisition module, which is used to perform multi-sensor cooperative acquisition based on a hitting cycle by driving a sensing array to determine a sensing number array; and an automatic scoring module, which is used to develop an automatic scorer on a sports platform, receive the sensing number array for each hitting cycle, and perform parallel self-attention recognition and scoring processing based on scoring correlation, decision rules and controversial features to generate a scoring result and display the scoring result on a platform interface; wherein if the controversial features are a non-empty set, three-dimensional trajectory reconstruction of a ball movement loop is performed to assist in scoring verification.

[0009] The one or more technical solutions provided in the application have at least the following technical effects or advantages: The badminton quick scoring method based on multi-sensor fusion provided by the application can deploy a projection assembly on a target field to project a projection field on the target field through multi-lens linkage control, drive a sensing array to perform multi-sensor cooperative acquisition based on a hitting cycle to determine a sensing number array, develop an automatic scorer on a sports platform, receive the sensing number array for each hitting cycle, and perform parallel self-attention recognition and scoring processing based on scoring correlation, decision rules and controversial features to generate a scoring result and display the scoring result on a platform interface; if the controversial features are a non-empty set, three-dimensional trajectory reconstruction of a ball movement loop is performed to assist in scoring verification. Thus, the technical problem that it is difficult to quickly deploy a standardized field and there is a lack of intelligent penalty technology system under random field deployment and it is difficult to meet the sports demand under field restriction in the prior art is solved. The badminton quick scoring method provided by the application can be quickly deployed, fuse multiple sensing information, have adaptive environmental compensation and three-dimensional verification capability for controversial balls, and achieve random field deployment and high-precision automatic scoring. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A badminton quick scoring method flowchart based on multi-sensor fusion is provided for the application.

[0011] Figure 2 A badminton quick scoring system structure diagram based on multi-sensor fusion is provided for the application.

[0012] Legend of reference signs: field projection module 11, sensing acquisition module 12, and automatic scoring module 13. DETAILED DESCRIPTION

[0013] The application provides a badminton quick scoring method and system through multi-sensor fusion, to solve the technical problems in the prior art that it is difficult to quickly deploy standardized venues, there is a lack of intelligent penalty technology system under the deployment of the venue, and it is difficult to meet the sports needs under the venue restrictions.

[0014] Embodiment one: as shown, the application provides a multi-sensor fusion badminton quick scoring method, which comprises: Figure 1 S1: deploying a projection component in the target venue to project a projection venue in the target venue through multi-lens linkage control, wherein the projection venue has adaptive reference control compensation based on the venue terrain and environmental light and shadow.

[0015] In one embodiment, the target venue refers to a fixed or temporary venue to be used for badminton sports, and the projection component refers to an integral whole composed of one or more projection lenses, mounting brackets, power supply and network communication units, etc.

[0016] In the specific implementation process, the venue corner points and key sideline coordinates are determined according to the venue information when deploying, and are mapped to a unified world coordinate system; then the effective projection area that can be covered by a single lens is calculated according to the projection distance and projection ratio analysis, and the installation pose of the lens and the bracket is planned in combination with the shielding risk and the installation height.

[0017] In the application, the first mode or the second mode for projection is determined according to the venue requirements, for example: if the first mode is adopted, the projection lens is deployed at both ends of the goal limit, and each end is composed of two projection lenses.

[0018] Then, in this deployment condition, the projection parameters are decided, and then the deployed multiple projection lenses are controlled to perform venue projection control. At the same time, based on the venue terrain and environmental light and shadow, further control compensation is performed, such as compensating the chroma of the projection color in the outdoor high-brightness environment to improve the contrast and clarity of the projection.

[0019] For example, the obtained badminton venue line frame, functional partition and auxiliary visual elements (such as landing hot area, serving area prompt, sideline highlight, etc.) are accurately mapped to the ground or bearing plane in the form of graphic overlay, thereby forming a perceivable and correctable projection venue.

[0020] ​Further, the step S1 includes: deploying a projection component according to a projection requirement, wherein the projection component includes a projection lens and a support; if it is a first mode projection of a small range, deploying the projection component according to a first deployment mode; if it is a second mode projection of a large range, deploying the projection component according to a second deployment mode; wherein, with a goal as a boundary, the first deployment mode is to deploy the projection component at a first edge end position and a second edge end position of the boundary, and each edge end position is deployed with a double projection lens; and the second deployment mode is to mirror deploy the projection component at a double side position based on the boundary.

[0021] In the embodiment, the projection component is deployed according to a projection requirement in a target field, wherein the projection requirement refers to determining the configuration requirement of the projection according to the range of the badminton court, the accuracy of the display content, and different requirements of the projection coverage range. The projection component is a hardware unit for realizing the projection function, and its core components include a projection lens and a support. The projection lens is used to generate and project an optical image of a visual area of a competition field, and the support is used to fix and support the projection lens to ensure the stability of the projection lens at a set position and the accuracy of the projection direction.

[0022] In the preferred implementation process, a fixed support or an adjustable support can be selected according to the size of the field and the environmental conditions to adapt to the arrangement requirements of different competition scenes.

[0023] First, if it is a first mode projection of a small range, the projection component is deployed according to a first deployment mode. In the embodiment, the small range refers to a small field area covered by the projection, for example, it is suitable for a single badminton competition scene or a temporary training area, and the arrangement of the projection equipment focuses on accurately covering the key competition area rather than the entire field.

[0024] In this mode, the first deployment mode is adopted, and the implementation is based on the goal boundary, that is, the boundary position of the middle net post in the badminton court is used as a separation reference to distinguish the two end areas of the field. In the specific implementation, the projection lens is arranged at the edge positions of the boundary, which are respectively referred to as a first edge end position and a second edge end position, and two projection lenses are deployed at each edge end position, that is, a double projection lens configuration.

[0025] Further preferably, the double lens arrangement has the advantage of realizing overlapping projection of the same side area, thereby enhancing the brightness and clarity of the projection picture, and to some extent, reducing the influence of the distortion of a single lens projection on the overall image.

[0026] Second, if it is a second mode projection of a large range, the projection component is deployed according to a second deployment mode. In the embodiment, the large range refers to a large field area covered by the projection, for example, it is suitable for a double badminton competition scene or a full-field coverage projection of a formal event.

[0027] In this mode, the second deployment mode is adopted, in which the projection lenses are deployed in a mirror image on the basis of the boundary on both sides, that is, the projection lenses are symmetrically arranged on both sides of the center line of the field (i.e., the boundary where the net is located), so that the projection devices on both sides are in a symmetrical relationship in spatial layout. The advantage is that it can cover both sides of the field at the same time, form balanced projection effect, and reduce the brightness difference and geometric distortion of the left and right pictures through the mirror-symmetrical projection angle.

[0028] In actual operation, each group of lenses can be adjusted to the corresponding focal length and projection angle to ensure that the projection identification in each area is clear and identifiable when covering a large range of field.

[0029] Further, the step S1 includes introducing an upper controller, wherein the upper controller is a cooperative central controller of the projection assembly; after the deployment of the projection assembly is completed, triggering the upper controller, executing a participation control decision based on image projection according to a first control node to determine a first participation control group, wherein the first participation control group at least includes a projection parameter and a field corner point; and performing cooperative projection control on the distributedly deployed projection lenses according to the first participation control group.

[0030] In the present application, the upper controller is a cooperative central controller of the projection assembly, which is a central controller for unified scheduling and management of multiple projection assemblies. Its functions include not only parameter calculation, but also task allocation and cooperative control execution, to ensure that the projection assemblies in different deployment positions form stable, synchronized and target-demand-compliant projection effects.

[0031] In the specific embodiment of the present application, the logical architecture of the upper controller is determined, specifically including a first control node, a second compensation node and a third compensation node, wherein the first control node executes a participation control decision, the second compensation node executes a participation control compensation decision for the case where the terrain affects the projection quality, and the third compensation node executes a participation control compensation decision for the case where the environment light affects the projection quality. The execution nodes are cascaded as the logical architecture, which is further trained to convergence in a sample-driven manner to serve as a built upper controller, and communicates with the projection assembly.

[0032] After the deployment of the projection assembly is completed, the upper controller is triggered, for example, through manual instructions, system signals or automatic detection conditions, to start the operation of the upper controller and enter the projection control decision phase.

[0033] Specifically, according to the first control node, that is, the initial decision point in the control logic flow, a set of reference and control parameters for guiding the projection behavior is generated according to the field environment, lens position and projection demand, as the first participation control group.

[0034] The first control group includes, but is not limited to, projection parameters and court corner points. Projection parameters include, but are not limited to, lens focal length, projection brightness, projection angle, resolution, and projection image size, used to ensure the matching degree between the projected image and the target area. Court corner points refer to the coordinates of key geometric boundary points of the badminton court, including the four corner points and other important marker points, used for geometric correction and positioning mapping of the image to ensure accurate alignment between the projected markers and the actual court lines.

[0035] Subsequently, the host controller distributes the parameter information from the first control group to each projection lens and adjusts its projection status in real time, so that the lenses in different positions maintain consistency and complementarity in terms of image position, size, brightness and color, thereby forming a continuous and seamless projection image throughout the venue to meet the needs of competition scoring and visualization.

[0036] Furthermore, step S1 of this application includes: introducing a first condition based on site topography and a second condition based on ambient light and shadow; determining whether the first condition is met; if not, triggering a second compensation node connected to the first control node to perform adaptive terrain calibration compensation on the first control group and determine a second compensation control group; determining whether the second condition is met; if not, triggering a third compensation node connected to the second compensation node to perform projection brightness compensation on the second compensation control group and determine a third compensation control group.

[0037] In one embodiment, a first condition based on site topography is introduced, where topography refers to the flatness, local warping, and deformation of the resilient floor of the target site. This first condition measures the sensitivity of the projected geometry to topographic deviations, and its criteria include, but are not limited to: whether corner reprojection errors exceed corresponding thresholds, whether grid alignment residuals are outside the allowable range, and whether there are excessive undulations in the ground elevation / slope scan.

[0038] The second condition is used to measure the consistency of projected luminance in the current environment. Its criteria include, but are not limited to: whether the average illuminance and illuminance gradient of the site exceed the threshold, whether the color temperature fluctuation exceeds the threshold, and whether the coverage ratio of the shadow mask on the key line segment exceeds the limit; where ambient light and shadow refer to factors such as illuminance, color temperature, dynamic shadows and reflected glare.

[0039] In a preferred embodiment, the first condition and the second condition are calculated online by the geometric sensing chain and the photometric sensing chain, respectively, and the binary results of satisfying / not satisfying the condition and the corresponding metric values ​​are periodically output.

[0040] Subsequently, it is first determined whether the first condition is met. If the first condition is not met, the geometric compensation process begins: based on the marker point deviation returned by the calibration camera and the elevation map from the terrain scan, the increment of the non-planar mapping (e.g., thin plate splines or piecewise bicubic splines) is solved, and the pixel-level geometric lookup table is updated to correct local distortions. The control parameters within the first parameter control group are adjusted, and a second compensation parameter control group is generated after compensation is completed. This compensation process is automated based on a pre-trained and constructed second compensation node.

[0041] For example, the reprojection error threshold can be set to ≤1–2 pixels, and the grid alignment residual root mean square value converges to ≤0.5 pixels to be considered as passing.

[0042] Furthermore, it is determined whether the second condition is met. If not, the third compensation node connected to the second compensation node is triggered to perform projection brightness compensation on the second compensation parameter control group and determine the third compensation parameter control group.

[0043] For example, the third compensation node, oriented towards photometric consistency, calculates block-by-block or pixel-by-pixel gain and gamma correction tables based on the current illuminance distribution, color temperature estimation, and shadow mask. It also jointly solves the fusion weight map in overlapping lens regions to suppress brightness steps at seams. If necessary, it performs cross-device consistency correction on white balance to ensure that color difference in the same area is below a perceptible threshold. To avoid flicker, photometric parameters are updated smoothly using temporal low-pass filtering, and an upper limit constraint is set on the spatial gradient to ensure a natural brightness transition. After compensation, the third compensation parameter control group is output.

[0044] For example, the threshold for triggering the rate of change of illuminance can be set to >10% / s and the threshold for shadow coverage of key line segments can be set to >8% of the area. When either one is triggered, brightness compensation will be initiated. After compensation, the brightness uniformity of the seam area is required to be ≥80% and the color difference ΔE≤2, which means that the second condition is satisfied.

[0045] S2: By driving the sensor array, perform multi-sensor collaborative acquisition based on the hitting cycle to determine the sensor array.

[0046] Furthermore, performing multi-sensor collaborative data acquisition based on the striking cycle, step S2 of this application includes: According to the hitting cycle, the sensor array is triggered to perform sensor loop detection; wherein, the sensor types that make up the sensor array include at least microwave radar, piezoelectric film, acoustic array and IMU, wherein the microwave radar performs ball speed detection, the piezoelectric film performs ball hitting force detection, the acoustic array performs landing point detection, and the IMU performs ball motion attitude detection.

[0047] In this application, the hitting cycle refers to the time period from when the shuttlecock is hit to when it is hit again, which usually covers the entire process of the shuttlecock's flight in the air, the moment it lands or is hit.

[0048] The sensor array is a collection of sensing units formed by arranging and connecting various types of sensors according to a certain logic. Its function is to collect key physical quantities in real time during the competition.

[0049] In this embodiment, the hitting cycle is used as the trigger condition; that is, the time interval between when the shuttlecock is hit and when it is hit again is used as the sensing loop. After the sensing array completes data acquisition for one hitting cycle, the data is integrated into the sensing array.

[0050] Specifically, based on the sensor array, information from one hitting cycle is collected to evaluate the state of that hit, and the above process is repeated until the end of the match.

[0051] The sensing types comprising the sensor array include at least microwave radar, piezoelectric thin film, acoustic array, and IMU.

[0052] Specifically, the microwave radar utilizes the Doppler effect or time-of-flight principle of microwave signals to detect the speed changes of a badminton shuttlecock during flight in a non-contact manner, making it suitable for high-frequency sampling to obtain speed curves. That is, the microwave radar calculates the shuttlecock's speed at every instant through continuous ranging and Doppler frequency shift, ensuring high accuracy and a high sampling rate for the speed data.

[0053] Piezoelectric thin films are sensing elements based on the piezoelectric effect. When a badminton shuttlecock comes into contact with or collides with a racket, the thin film generates a tiny electrical signal. The amplitude of this signal is proportional to the contact force, and therefore can be used to quantify the force of the hit. In other words, the electrical signal generated by the piezoelectric thin film at the moment of impact, after amplification and filtering, can be directly used for force calculation.

[0054] An acoustic array is an array structure composed of multiple microphones arranged in a specific geometric distribution. By analyzing the arrival time difference and spectral characteristics of the sound, the exact coordinates of the shuttlecock's landing point can be determined, making it suitable for determining whether the shuttlecock is out of bounds and identifying the landing point. In other words, the acoustic array, combined with triangulation, determines the precise landing point of the shuttlecock, maintaining high recognition accuracy even in complex indoor echo environments.

[0055] An IMU (Inertial Measurement Unit) is an inertial sensing component composed of sensors such as accelerometers and gyroscopes. It can acquire the attitude information of a badminton shuttlecock in real time during its flight, including pitch, roll, and yaw angles, to analyze the shuttlecock's trajectory and rotation in the air. In other words, the IMU continuously outputs attitude data throughout the flight and integrates it with velocity and landing point information, providing reliable input data support for subsequent trajectory reconstruction and automatic scoring.

[0056] S3: Develop an automatic scorer on the sports platform. For each hitting cycle, receive the sensor array and perform parallel self-attention recognition and scoring processing based on scoring correlation, decision rules and disputed features to generate scoring results and display them on the platform interface. If the disputed features are a non-empty set, perform three-dimensional trajectory reconstruction of the ball's motion loop to assist in scoring verification.

[0057] In one embodiment, the automatic scorer receives and processes a sensor array acquired by multiple sensors in each shot cycle online.

[0058] In this application, the sports platform refers to an integrated hardware and software system with data access, model inference, and visualization capabilities, which can be deployed on referee terminals or cloud inference nodes; the automatic scorer refers to a set of scoring decision modules running on the platform, including logical functions such as preprocessing, parallel self-attention recognition, rule fusion, dispute detection, and visualization output. The sensor array is the structured input of the aforementioned sensor channels under a unified time reference, containing elements such as speed, force, landing point, and posture, as well as their quality indicators, used to trigger the scoring process for that cycle.

[0059] In a specific implementation, the automatic scorer first aligns the multi-channel data, i.e. the sensor array, to the zero point of the hitting cycle. By performing time-domain alignment, amplitude normalization, and coordinate system unification, it ensures that the outputs of different sensors can be analyzed in the same feature space.

[0060] Preferably, to meet real-time requirements, the above standardization process adopts a sliding window and incremental update strategy to keep the preprocessing latency of a single cycle within a predetermined threshold.

[0061] In a further embodiment, the automatic scorer performs parallel self-attention recognition and scoring processing on the sensor array based on scoring correlation, decision rules, and disputed features.

[0062] In one feasible implementation provided by this application, parallel self-attention recognition refers to constructing a multi-layered self-attention network, with each layer focusing on different semantic subspaces of the input using different query vectors: the first is scoring relevance, which learns the cross-modal coupling relationship between speed, landing point, force, and posture, and outputs scoring relevance features to measure whether the cycle constitutes a valid scoring event; the second is decision rules, which encode the rules into learnable or hard-constrained attention templates, including clauses such as out-of-bounds / in-bounds, legality of the serve, and net touch / second hit, and outputs decision rule features to constrain predictions to not deviate from the competition rules; the third is a disputed feature branch, which focuses on situations such as near the sideline, insufficient confidence in the landing point, inconsistency between speed and landing point, and conflict between acoustic and radar time series, and outputs a set of disputed features.

[0063] Subsequently, the automated scorer fuses scoring relevance features with decision rule features to generate the first scoring data. This data specifically includes the event type for that period (e.g., valid score, out-of-bounds, replay, etc.), the responsible party (Party A / Party B), and the confidence score, along with explanatory tags (e.g., key rules triggered, most contributing sensor features) to improve auditability.

[0064] If the output of the controversial feature branch is an empty set or its comprehensive score is lower than the set threshold, the first scoring data will be directly used as the scoring result for that period and displayed and accumulated in the interface.

[0065] If the controversial feature is a non-empty set and its score exceeds the controversy threshold, the 3D trajectory reconstruction of the ball's motion loop is automatically triggered to assist in scoring verification, outputting a trajectory sequence containing spatial coordinates, velocity vectors, rotation parameters, and their uncertainties. After reconstruction, the minimum distance and contact time are calculated to obtain secondary scoring data.

[0066] Subsequently, the automatic scorer fuses the first and second scoring data to generate the final score.

[0067] In an exemplary implementation, if the consistency check of the two scoring data passes, the high-confidence result after reconstruction assistance is directly adopted; if they are inconsistent and both confidence levels are below the threshold, a flag indicating that manual review is required can be output to ensure fairness.

[0068] The final scoring results include event category, winner / loser attribution, confidence level, key evidence that triggered the event, and replay index, and are written to the audit log for post-match review.

[0069] Furthermore, in developing an automatic scorer, step S3 of this application includes: deploying a first self-attention layer based on scoring relevance; deploying a second self-attention layer based on decision rules; deploying a third self-attention layer based on controversial features; cascading the first, second, and third self-attention layers to construct a scoring unit; and cascading the scoring unit with a 3D reconstruction unit to generate the automatic scorer, wherein the 3D reconstruction unit is constructed based on adversarial network supervised training.

[0070] In one embodiment, a first self-attention layer is deployed based on scoring relevance. This first self-attention layer is used to automatically learn cross-modal coupling features directly related to whether a scoring event occurs from a multimodal sensor array. Inputs include velocity sequences aligned by the hit cycle, force features, acoustic landing point estimates, and attitude quaternions. These are first linearly embedded and mapped to a unified 3D feature space, with positional encoding introduced to preserve temporal order information. Subsequently, a multi-head self-attention mechanism is used to perform global dependency checks on different modal segments. The output is a scoring relevance feature vector containing separability information for events such as in-bounds / out-of-bounds, net-touch / non-net-touch, and valid / invalid hits.

[0071] Similarly, a second self-attention layer is deployed based on decision rules. This second layer explicitly encodes the competition rules into learnable or semi-hard constraint attention templates, which are used to constrain and supplement the features generated by the first layer. Rule representation can employ sparse logic graphs or finite state machine fragments, mapped to a set of rule vectors by a rule embedder and used as a query. Preferably, for hard rules that must be satisfied (e.g., the serve landing point must be within bounds), a prohibition / enforcement mask is applied before the feedforward layer using a gating unit; for probabilistic clauses (e.g., whether to re-serve after a net touch), a penalty is applied during the training phase using a soft constraint loss. This layer outputs a decision rule feature vector, accompanied by interpretable labels and confidence levels for the triggering rules.

[0072] Similarly, a third self-attention layer is deployed based on controversial features. This third layer focuses on identifying difficult case patterns with ambiguous boundaries, sensor conflicts, or insufficient evidence to trigger subsequent reconstruction verification. The input is an intermediate attention map from the first and second layers. The attention distribution is calculated through an exceptionally sensitive query design (e.g., magnifying the edge neighborhood, low-confidence landing points, and velocity-landing point inconsistency regions), and the output is a set of controversial features and the corresponding controversy score.

[0073] Subsequently, the first, second, and third self-attention layers are cascaded to construct a scoring unit. The output of the scoring unit includes elements such as event category, attributor, and confidence level, and is accompanied by a rule trigger list and key evidence index to support subsequent auditing and review.

[0074] Furthermore, based on the above construction mechanism and decision-making logic, a sample training method is used to supervise the training of the scoring unit until convergence, so as to ensure the accuracy of automated decision-making.

[0075] Subsequently, the scoring unit and the 3D reconstruction unit are cascaded to generate a complete inference chain for the automatic scorer. The 3D reconstruction unit is activated when a dispute is triggered. Based on the twin space of the projection field, it integrates radar velocity, acoustic TDOA landing point, IMU attitude, and projection geometric constraints to solve the sphere's 3D trajectory and its topological relationship with the edge / ground.

[0076] In this application, the three-dimensional reconstruction unit adopts the adversarial supervised training principle. That is, based on the above logic principle, the generator-discriminator architecture is used for supervised training of automatic three-dimensional trajectory generation until the discriminator determines that the generation accuracy meets the physical fit. The trained generator is then used as the three-dimensional reconstruction unit.

[0077] Furthermore, step S3 of this application includes: establishing communication interaction between the projection component, the sensor array, and the motion platform; transmitting the sensor array back to the motion platform, activating the automatic scorer, performing layer attention checks in parallel, and determining the first self-attention feature, the second self-attention rule, and the third self-attention feature; fusing the first self-attention feature and the second self-attention rule to generate first scoring data; if the third self-attention feature is an empty set, using the first scoring data as the first scoring result of the first cycle.

[0078] In one embodiment, the projection component, the sensor array, and the motion platform establish communication interaction. That is, the three communicate via wired or wireless links to send commands and transmit data back, and introduce timestamps and integrity checks at the transmission layer to ensure the consistency and traceability of cross-device data.

[0079] The interaction between the projection component and the motion platform is used to synchronize the projection geometry / photometric parameters and status information; the interaction between the sensor array and the motion platform is used to transmit the original sensing sequence back at high frequency for subsequent scoring and decision-making processes.

[0080] Specifically, the sensor array, i.e., the multimodal structured data packet aggregated according to the hitting cycle, including elements such as speed, force, landing point, attitude and confidence level, is transmitted back to the motion platform. Then, the automatic scorer is activated to make a scoring decision on the sensor array for that hitting cycle.

[0081] Subsequently, the attention check is performed in parallel execution layers. Specifically, the first, second, and third self-attention layers extract self-attention features from the sensor array to determine the first self-attention feature, which is used to characterize the scoring relevance and focuses on cross-modal coupling evidence of velocity, landing point, force, and attitude. The second self-attention rule is determined to characterize the matching degree of the rule constraint and uses the encoded competition terms as a query to perform cross-attention on the base features. The third self-attention feature is determined to characterize controversial patterns and emphasizes abnormal regions such as edge neighborhoods, low-confidence landing points, or intermodal conflicts.

[0082] Furthermore, the first self-attention feature and the second self-attention rule are integrated, that is, the scoring correlation vector and the rule matching vector are gated and weighted or conditionally inferred to output the preliminary event type, attribution party and confidence level of the period, along with the most critical rule triggering clause and evidence index, for subsequent auditing and interpretable presentation, so as to determine the first scoring data.

[0083] To improve stability, a penalty term is applied to inconsistent evidence during the fusion process, and the weights are adaptively adjusted based on channel quality, thereby obtaining high-confidence conclusions when the rules are clear and the evidence is sufficient.

[0084] Subsequently, the third self-attention feature is verified. If the third self-attention feature is an empty set, it indicates that the dispute score is lower than the trigger threshold and no mark that needs to be reviewed is generated. Then, it is determined that no disputed pattern was detected in this period, and the first scoring data is used as the first scoring result of the first period.

[0085] In this case, skip the 3D reconstruction and secondary verification steps, directly write the first scoring data, and visualize it on the sports platform interface with minimal flashing. At the same time, record the timestamp and evidence summary for post-match review.

[0086] Furthermore, step S3 of this application includes: if the output of the third self-attention layer is a non-empty set, activating the three-dimensional reconstruction unit; according to the three-dimensional reconstruction unit, using the twin space based on the projection field as the basis, performing three-dimensional trajectory reconstruction generation to determine the ball motion loop; generating second scoring data for the ball motion loop; and fusing the first scoring data and the second scoring data as the first scoring result of the first cycle.

[0087] In one embodiment, if the output of the third self-attention layer is a non-empty set, indicating that a valid label is given on any sub-condition such as edge neighborhood, modal conflict or insufficient evidence, and its dispute score exceeds a preset threshold, then it is considered that there is a disputed mode in the hitting cycle, and the three-dimensional reconstruction unit is activated to enter the enhanced verification process.

[0088] Subsequently, based on the three-dimensional reconstruction unit, using the twin space based on the projection field as a basis, three-dimensional trajectory reconstruction is performed to generate and determine the ball's motion loop.

[0089] Specifically, the twin space refers to a three-dimensional geometric and photometric parameterized space that corresponds one-to-one with the real projected field. It includes at least the field plane, the voxelized representation of the boundary lines, the net plane, the reference coordinate system, and the projected geometric mapping. Its parameters are derived from the parameterized snapshots after the projection component calibration and terrain / photometric compensation. Further, a three-dimensional reconstruction of the badminton trajectory is performed to determine the ball's motion loop, that is, the closed-loop event sequence from the initial state of the hit to the first landing point (or the point of contact with the net / secondary contact with the ground) within this cycle, which is used to characterize the key evidence for validity and attribution determination.

[0090] In a further embodiment, for the ball's motion loop, second scoring data is generated, which in this application refers to an independent judgment result based on a three-dimensional trajectory, including at least the event type (inbound / outbound, net contact, second hit, etc.), the party attributing the ball, the critical contact moment, the minimum distance, and the corresponding uncertainty.

[0091] Subsequently, the first scoring data and the second scoring data are combined to form the first scoring result for the first cycle.

[0092] In one specific implementation, the first scoring data generated by the aforementioned parallel self-attention is used as the prior, and the second scoring data obtained from the aforementioned three-dimensional reconstruction is used as the likelihood evidence. The posterior conclusion is calculated by combining their respective confidence levels and quality weights.

[0093] When both are consistent and the posterior confidence is higher than the display threshold, the final attribution and event type are given directly; when there is a conflict, the reconstructed consistency score, the minimum distance of the boundary line and its confidence interval are used as the arbitration elements; if the posterior confidence is still lower than the conservative threshold, the review flag is output and the evidence index is retained.

[0094] Finally, the first scoring result is written into the scoring context of this period, triggering the scoreboard update and interface visualization overlay, while recording the timestamp and fusion weight for post-match review.

[0095] In summary, the above process ensures that strong evidence is provided through 3D reconstruction of the twin space when disputes are detected, thereby improving the accuracy and interpretability of the scoring results in the first cycle.

[0096] Furthermore, step S3 of this application includes: performing periodic polling processing based on the acquisition of data from the sensor array and the scoring of the automatic scorer during the hitting cycle, until the Nth scoring result of the Nth cycle is obtained; integrating the first scoring result up to the Nth scoring result as a total score result, and visualizing it in a pop-up window on the display interface of the sports platform.

[0097] In one embodiment, periodic polling refers to a fixed process of acquisition-inference-output, with the ball-hitting cycle as the smallest processing unit, which starts the next round of acquisition-inference-output after the end of each cycle is triggered; wherein, acquisition includes acquiring sensor array data for the cycle and obtaining the sensor array for the cycle; inference includes a scoring decision process based on an automatic scorer; and output includes generating the scoring result for the cycle and writing it into the state machine.

[0098] Where N is a natural number representing the count of the currently completed valid shot cycles. Preferably, invalid cycles (such as false triggers or incomplete data) are marked and not included in N to ensure the traceability and consistency of the cumulative results.

[0099] Subsequently, the first scoring result up to the Nth scoring result is integrated as the total score. Integration refers to performing deterministic folding on {the first scoring result, ..., the Nth scoring result} in chronological order to generate a total score that includes the current score, game / set result, and technical statistics (such as the number of valid rallies, average rally duration, and sideline dispute rate).

[0100] Furthermore, if there are periods marked as requiring manual review, a prompt will be retained in the total score results, and a direct link to review the previous period will be provided.

[0101] Subsequently, the total score results are visualized in a pop-up window on the sports platform's display interface. In one feasible implementation, a lightweight pop-up window above the main score bar displays the event category, attributor, and confidence level of the latest update. Through the above-described polling, integration, and visualization process, stable scoring and total result output from the 1st to the Nth cycle are achieved, meeting the application requirements of real-time, fairness, and traceability.

[0102] The multi-sensor fusion badminton rapid scoring method provided in this application has the following technical effects: Adaptive Projection Field Construction: Projection components are deployed on the target field, and the projection field is controlled through multi-lens linkage. Adaptive parameter control compensation (including terrain calibration compensation and projection brightness compensation) is performed based on the field terrain and ambient light and shadow. This ensures that the projection field accurately matches the actual field, unaffected by terrain undulations and light changes, providing a stable benchmark for subsequent trajectory judgment and landing point detection. Multi-Sensor Collaborative Acquisition: A sensor array composed of microwave radar, piezoelectric film, acoustic array, and IMU is driven to collaboratively acquire data according to the hitting cycle, obtaining data such as ball speed, hitting force, landing point, and motion posture, forming a sensor array. This multi-dimensional capture of badminton motion information avoids the limitations of single-sensor methods, improves data comprehensiveness and accuracy, and provides rich evidence for scoring.

[0103] A three-layer self-attention automatic scoring system was developed, comprising a first self-attention layer (scoring relevance), a second self-attention layer (judgment rules), and a third self-attention layer (controversial features). These layers are cascaded and combined with a 3D reconstruction unit (based on an adversarial network). A hierarchical attention mechanism accurately extracts key scoring information, while the 3D reconstruction unit resolves controversial calls, improving the automation and reliability of scoring. Dynamic scoring processing flow: For each stroke cycle, the automatic scoring system performs attention checks in parallel, fusing scoring relevance and judgment rules to generate preliminary scoring data. If controversial features exist, the 3D reconstruction unit is activated to generate the ball's motion loop, fusing the two sets of data to determine the final scoring result. This achieves rapid scoring while reducing misjudgments through a dispute resolution mechanism, balancing efficiency and accuracy to meet the high-paced demands of badminton matches.

[0104] Example 2: Based on the same inventive concept as the multi-sensor fusion badminton rapid scoring method in the foregoing examples, such as... Figure 2 As shown, this application provides a multi-sensor fusion badminton rapid scoring system, the system comprising: The site projection module 11 is used to deploy projection components on the target site and project the projection site on the target site with multi-lens linkage control. The projection site has adaptive parameter control compensation based on the site terrain and ambient light and shadow. The sensor acquisition module 12 is used to drive the sensor array to perform multi-sensor collaborative acquisition based on the hitting cycle and determine the sensor array. Automatic scoring module 13 is used to develop an automatic scorer on the sports platform. For each hitting cycle, it receives the sensor array and performs parallel self-attention recognition and scoring processing based on scoring correlation, decision rules and disputed features, generates scoring results and displays them on the platform interface. If the controversial feature is a non-empty set, the three-dimensional trajectory reconstruction of the ball's motion loop is used to assist in scoring verification.

[0105] Furthermore, the field projection module 11 is used to perform the following steps: deploying a projection component on the target field according to the projection requirements, wherein the projection component includes a projection lens and a support; if it is a small-scale first mode projection, the field is deployed according to a first deployment method; if it is a large-scale second mode projection, the field is deployed according to a second deployment method; wherein, with the goal as the boundary, the first deployment method is to deploy at the first and second edge positions of the boundary line, and each edge position is equipped with dual projection lenses; the second deployment method is to deploy in a mirror image on both sides of the boundary.

[0106] Furthermore, the site projection module 11 is used to perform the following steps: introducing a host controller, wherein the host controller is a collaborative central controller for the projection components; after the deployment of the projection components is completed, triggering the host controller to perform image projection-based parameter control decisions according to the first control node, determining a first parameter control group, wherein the first parameter control group includes at least projection parameters and site corner points; and performing collaborative projection control on the distributed projection lenses according to the first parameter control group.

[0107] Furthermore, the site projection module 11 is used to perform the following steps: introducing a first condition based on the site terrain and a second condition based on ambient light and shadow; determining whether the first condition is met; if not, triggering a second compensation node connected to the first control node to perform adaptive terrain calibration compensation on the first control group and determine a second compensation control group; determining whether the second condition is met; if not, triggering a third compensation node connected to the second compensation node to perform projection brightness compensation on the second compensation control group and determine a third compensation control group.

[0108] Furthermore, the sensing acquisition module 12 is used to perform the following steps: triggering the sensing array to perform sensing loop detection according to the hitting cycle; wherein, the sensing types constituting the sensing array include at least microwave radar, piezoelectric film, acoustic array and IMU, wherein the microwave radar performs ball speed detection, the piezoelectric film performs hitting force detection, the acoustic array performs landing point detection, and the IMU performs ball motion attitude detection.

[0109] Furthermore, the automatic scoring module 13 is used to perform the following steps: deploying a first self-attention layer based on scoring relevance; deploying a second self-attention layer based on decision rules; deploying a third self-attention layer based on controversial features; cascading the first, second, and third self-attention layers to construct a scoring unit; cascading the scoring unit with a 3D reconstruction unit to generate the automatic scorer, wherein the 3D reconstruction unit is constructed based on adversarial network supervised training.

[0110] Furthermore, the automatic scoring module 13 is used to perform the following steps: the projection component, the sensor array, and the motion platform establish communication interaction; the sensor array is transmitted back to the motion platform, the automatic scorer is activated, and layer attention detection is performed in parallel to determine the first self-attention feature, the second self-attention rule, and the third self-attention feature; the first self-attention feature and the second self-attention rule are fused to generate the first scoring data; if the third self-attention feature is an empty set, the first scoring data is used as the first scoring result of the first cycle.

[0111] Furthermore, the automatic scoring module 13 is used to perform the following steps: if the output of the third self-attention layer is a non-empty set, activate the three-dimensional reconstruction unit; according to the three-dimensional reconstruction unit, based on the twin space of the projection field as the basis, perform three-dimensional trajectory reconstruction generation to determine the ball motion loop; generate second scoring data for the ball motion loop; and fuse the first scoring data and the second scoring data as the first scoring result of the first cycle.

[0112] Furthermore, the automatic scoring module 13 is used to perform the following steps: with each hitting cycle, it performs periodic polling processing based on the acquisition of data from the sensor array and the scoring of the automatic scorer until the Nth scoring result of the Nth cycle is obtained; it integrates the first scoring result up to the Nth scoring result as a total score result and displays it in a pop-up window on the display interface of the sports platform.

[0113] Through the foregoing detailed description of the multi-sensor fusion-based rapid scoring method for badminton, those skilled in the art can clearly understand the multi-sensor fusion-based rapid scoring method and system for badminton in this embodiment. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant details can be found in the method section. The above description of the disclosed embodiments enables those skilled in the art to implement or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-sensor fusion-based rapid scoring method for badminton, characterized in that, The method includes: Projection components are deployed at the target site, and multi-lens linkage control is used to project the projection field onto the target site. The projection field has adaptive parameter control compensation based on the site terrain and ambient light and shadow. By driving the sensor array, multi-sensor collaborative acquisition based on the hitting cycle is performed to determine the sensor array; An automatic scorer is developed on the sports platform. For each hitting cycle, the system receives the sensor array and performs parallel self-attention recognition and scoring processing based on scoring correlation, decision rules and disputed features to generate scoring results and display them on the platform interface. If the controversial feature is a non-empty set, the three-dimensional trajectory reconstruction of the ball's motion loop is used to assist in scoring verification.

2. The multi-sensor fusion-based rapid scoring method for badminton as described in claim 1, characterized in that, According to the projection requirements, a projection assembly is deployed at the target site, wherein the projection assembly includes a projection lens and a support. If it is a small-scale first-mode projection, the site deployment shall be carried out according to the first deployment method; If it is a large-scale second mode projection, the site deployment shall be carried out according to the second deployment method; Among them, with the goal as the boundary, the first deployment method is to deploy the projector at the first and second ends of the boundary line, with dual projection lenses deployed at each end; the second deployment method is to deploy the projector in a mirror image on both sides of the boundary line.

3. The multi-sensor fusion-based rapid scoring method for badminton as described in claim 2, characterized in that, A host controller is introduced, wherein the host controller is a central controller for the collaborative control of the projection components; After the deployment of the projection component is completed, the host controller is triggered to execute the image projection-based parameter control decision according to the first control node and determine the first parameter control group, wherein the first parameter control group includes at least projection parameters and site corner points; Based on the first parameter control group, coordinated projection control is performed on the distributed projection lenses.

4. The multi-sensor fusion-based rapid scoring method for badminton as described in claim 3, characterized in that, Introduce a first condition based on site topography and a second condition based on ambient light and shadow; Determine whether the first condition is met. If not, trigger the second compensation node connected to the first control node to perform adaptive terrain calibration compensation on the first control group and determine the second compensation control group. Determine whether the second condition is met. If not, trigger the third compensation node connected to the second compensation node to perform projection brightness compensation on the second compensation parameter control group and determine the third compensation parameter control group.

5. The multi-sensor fusion-based rapid scoring method for badminton as described in claim 1, characterized in that, Perform multi-sensor collaborative data acquisition based on the hitting cycle, including: Based on the ball striking cycle, the sensor array is triggered to perform sensor loop detection; The sensor array comprises at least a microwave radar, a piezoelectric film, an acoustic array, and an IMU. The microwave radar performs ball speed detection, the piezoelectric film performs ball impact force detection, the acoustic array performs landing point detection, and the IMU performs ball motion attitude detection.

6. The multi-sensor fusion-based rapid scoring method for badminton as described in claim 1, characterized in that, Develop an automatic scoring system, including: Deploy the first self-attention layer based on scoring correlation; Deploy a second layer of self-attention based on the judgment rules; Deploy a third self-attention layer based on controversial features; The first self-attention layer, the second self-attention layer, and the third self-attention layer are cascaded to construct a scoring unit; The scoring unit and the 3D reconstruction unit are cascaded to generate the automatic scorer, wherein the 3D reconstruction unit is constructed based on adversarial network supervised training.

7. The multi-sensor fusion-based rapid scoring method for badminton as described in claim 6, characterized in that, The projection component, the sensor array, and the motion platform establish communication interaction. The sensor array is transmitted back to the motion platform to activate the automatic scorer and perform layer attention detection in parallel to determine the first self-attention feature, the second self-attention rule and the third self-attention feature. The first self-attention feature and the second self-attention rule are combined to generate the first scoring data; If the third self-attention feature is an empty set, the first scoring data is used as the first scoring result of the first cycle.

8. The multi-sensor fusion-based rapid scoring method for badminton as described in claim 7, characterized in that, If the output of the third self-attention layer is a non-empty set, activate the 3D reconstruction unit; Based on the three-dimensional reconstruction unit, using the twin space based on the projection field as a basis, three-dimensional trajectory reconstruction is performed to generate and determine the ball's motion loop; For the ball's motion loop, generate second scoring data; The first scoring data and the second scoring data are combined to form the first scoring result for the first period.

9. The multi-sensor fusion-based rapid scoring method for badminton as described in claim 8, characterized in that, As the ball is hit, the sensor array collects data and the automatic scorer scores, and periodic polling is performed until the Nth score of the Nth period is obtained. The first scoring result is integrated up to the Nth scoring result, and the total score is displayed in a pop-up window on the display interface of the sports platform.

10. A multi-sensor fusion badminton rapid scoring system, characterized in that, The system is used to execute the multi-sensor fusion rapid scoring method for badminton as described in any one of claims 1-9, the system comprising: The site projection module is used to deploy projection components on the target site and project the projection site on the target site with multi-lens linkage control. The projection site has adaptive parameter control compensation based on the site terrain and ambient light and shadow. The sensor acquisition module is used to drive the sensor array to perform multi-sensor collaborative acquisition based on the hitting cycle and determine the sensor array. The automatic scoring module is used to develop an automatic scorer on the sports platform. For each hitting cycle, it receives the sensor array and performs parallel self-attention recognition and scoring processing based on scoring correlation, decision rules and disputed features, generates scoring results and displays them on the platform interface. If the controversial feature is a non-empty set, the three-dimensional trajectory reconstruction of the ball's motion loop is used to assist in scoring verification.

Images