A badminton landing point determination and evidence method and system based on visual trajectory

Abstract

The application relates to the technical field of computer vision processing, and particularly relates to a badminton landing point determination and evidence providing method and system based on a visual trajectory. The method obtains sparse sampling visual trajectory data corresponding to a badminton video, and establishes a mapping relationship between an image space and a regular court space based on court calibration parameters; ground-touch candidate detection is performed on the sparse sampling trajectory, a ground-touch time is determined in combination with a detection step adjustment strategy and a ground-touch state determination mechanism, a target landing point is identified according to corresponding trajectory motion characteristics; the target landing point is subjected to coordinate conversion and regular region matching based on the mapping relationship, a badminton landing point result is obtained, regular determination information and corresponding evidence playback data are further generated, and a landing point determination result is output. The application realizes accurate determination of a badminton ground-touch time and a landing point position and traceable evidence of a determination process, thereby improving the accuracy, reliability and credibility of badminton landing point determination.

Description

Technical Field

[0001] This application relates to the field of computer vision processing technology, and in particular to a method and system for determining and proving the landing point of a badminton shuttlecock based on visual trajectory. Background Technology

[0002] In related technologies, with the rapid development of computer vision and intelligent sports technology, video analytics-based intelligent referee and training assistance systems for badminton are increasingly being applied to scenarios such as match officiating and training statistics. These systems typically acquire badminton video footage using camera equipment, utilize target detection and trajectory tracking algorithms to obtain the shuttlecock's trajectory, and establish a mapping relationship between the image space and the regular court space by combining court location parameters. This enables the detection of the shuttlecock's landing point, determination of whether it is in or out of bounds, and auxiliary evidence collection. Some existing solutions further integrate trajectory prediction, perspective mapping, and deep learning detection technologies to improve the automation level of landing point determination.

[0003] However, existing methods for determining the landing point of badminton shuttlecocks still have certain limitations in practical applications. Due to the limited computing resources of terminal devices, traditional methods often rely on consecutive frame hits, fixed thresholds, or single motion features for landing detection. When there is motion blur, occlusion, or increased sampling intervals, trajectory breaks can easily occur, leading to missed landing events. Furthermore, the detection step size and trigger parameters are usually configured in a fixed way, making it impossible to adaptively adjust according to different sampling densities and computational loads, thus making it difficult to balance real-time performance and detection accuracy. In addition, existing methods typically use a uniform court area model to simultaneously complete landing detection and rule-based penalties, making it difficult to effectively distinguish between the trigger observation area and the actual valid playing area, easily leading to misjudgments or missed detections in scenarios such as edge shots and shots on the line. Moreover, the evidence-gathering process after landing, such as slow-motion replay and rule-based field map rendering, may consume inference resources, affecting the real-time operation of the system; in multi-ball training scenarios, stationary shuttlecocks and trajectory jumps may also cause false landing detections.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0005] The main objective of this application is to propose a method and system for determining and proving the landing point of a badminton shuttlecock based on visual trajectory, so as to achieve accurate determination of the time of contact and landing point of the shuttlecock and traceable evidence of the determination process, thereby improving the accuracy, reliability and credibility of the determination of the landing point of the badminton shuttlecock.

[0006] To achieve the above objectives, one aspect of this application proposes a method for determining and proving the landing point of a badminton shuttlecock based on visual trajectory, the method comprising the following steps: Acquire sparsely sampled visual trajectory data corresponding to badminton videos; the sparsely sampled visual trajectory data includes video frame identifier data, ball center position coordinate data, detection time series data, and trajectory confidence data; Based on the field calibration parameters and the sparse sampling trajectory data, a mapping relationship between the image space and the regular field space is established; The sparse sampled trajectory data is subjected to ground contact candidate detection, and the ground contact time is determined by combining the detection step size adjustment strategy and the ground contact state determination mechanism. The contact point is identified based on the trajectory motion characteristics corresponding to the contact moment, and the target contact point is obtained. Based on the target contact point and the mapping relationship, coordinate transformation is performed, and regular area matching is conducted to obtain the badminton shuttlecock landing point result; Based on the badminton shuttlecock landing point results, rule judgment information is generated, corresponding evidence playback data is constructed, and landing point judgment results are output.

[0007] In some embodiments, establishing the mapping relationship between the image space and the regular site space based on the site calibration parameters and the sparse sampling trajectory data includes: Based on the badminton court corner point calibration data, the homography transformation parameters between the image space and the regular court space are determined; Based on the homography transformation parameters and the field corner point calibration results, the main playing area and the outer contact area corresponding to the regular field space are determined; the outer contact area is used for contact observation access control, and the main playing area is used for in-bounds, out-of-bounds and line-pressing judgments. Based on the homography transformation parameters, coordinate mapping is performed on the sparsely sampled visual trajectory data to obtain trajectory coordinate data in a regular field space. Based on the main competition area, the extended contact area, and the trajectory coordinate data, a mapping relationship is established between the image space and the rule field space.

[0008] In some embodiments, the step of performing ground contact candidate detection on the sparse sampled trajectory data, combined with a detection step size adjustment strategy and a ground contact state determination mechanism, to determine the ground contact time includes: Based on the sparse sampled trajectory data, the badminton trajectory change features are extracted to generate a ground contact candidate detection sequence; Based on the difference between the current detection step size and the preset benchmark step size, the cumulative hit threshold, missed detection tolerance threshold and observation waiting parameters corresponding to the ground contact detection are adjusted to generate step size-adapted detection parameters. Based on the step size adaptation detection parameters, the state of the candidate ground contact detection sequence is determined to determine the ground contact observation state; wherein, before entering the ground contact observation state, when the center of the badminton shuttlecock is detected to be located in the outer ground contact area, the number of valid hits is accumulated, and the accumulated number of valid hits is maintained when a missed detection occurs; when the number of valid hits meets the accumulated hit threshold, it is determined to enter the ground contact observation state; Based on the motion trajectory segment corresponding to the ground-touching observation state, ground-touching candidate moments are detected, and the ground-touching moment is determined according to the motion characteristics corresponding to the candidate moments.

[0009] In some embodiments, detecting candidate ground contact times based on the motion trajectory segment corresponding to the ground contact observation state, and determining the ground contact time based on the motion characteristics corresponding to the candidate times, includes: Based on the motion trajectory segment, extract the vertical coordinate extreme points, local velocity change points and trajectory convergence points to generate a set of candidate ground contact moments; Based on the set of ground contact candidate moments, the corresponding pre-contact deceleration features, post-contact rebound features, trajectory convergence features, and displacement stability features are extracted to obtain candidate evaluation features. Based on the candidate evaluation features, the set of ground contact candidate times is scored and sorted to determine the target ground contact candidate time; Based on the target touchdown candidate time combined with the touchdown state boundary conditions, when the shuttlecock center is detected to leave the outward touchdown area, a valid post-touch sample is obtained, or the number of consecutive missed detections exceeds the missed detection tolerance threshold, the touchdown observation window is closed, and the target touchdown candidate time with the highest current score is determined as the touchdown time.

[0010] In some embodiments, acquiring the motion trajectory segment includes: Before the shuttlecock's center is first detected to enter the outward contact area, trajectory sampling points with a downward trend are continuously extracted from the sparsely sampled visual trajectory data to obtain the pre-contact falling tail segment data. In response to the first detection that the center of the badminton shuttlecock enters the outward contact area, the data of the pre-contact drop segment is frozen, and the trajectory segment within the area after the center of the badminton shuttlecock enters the outward contact area is acquired. The data of the pre-touch fall tail segment is merged with the trajectory segment within the region according to the detection time sequence to obtain the motion trajectory segment.

[0011] In some embodiments, the step of identifying the point of contact based on the trajectory motion features corresponding to the moment of contact to obtain the target point of contact includes: Based on the trajectory segment corresponding to the moment of contact with the ground, extract the change features of the badminton shuttlecock's center position to generate a candidate set of contact points; Based on the candidate set of contact points, extract the corresponding vertical extreme value features, velocity change features and trajectory convergence features to generate contact point evaluation features; Based on the evaluation features of the contact point, the candidate contact point set is matched and filtered to determine the candidate location of the target contact point; The target contact point is obtained by performing position correction based on the candidate positions of the target contact point and the image coordinate information corresponding to the contact time.

[0012] In some embodiments, the step of performing coordinate transformation based on the target contact point and the mapping relationship, and performing regular region matching to obtain the badminton shuttlecock landing point result includes: Based on the image coordinates corresponding to the target contact point and the mapping relationship, coordinate mapping is performed to obtain the coordinates of the regular field. Based on the rules, the field coordinates are spatially matched with the main competition area to determine the area attributes corresponding to the target touch point. Boundary determination is performed based on the positional relationship between the rule-based field coordinates and the field edge lines to determine the edge line state corresponding to the target contact point; The badminton shuttlecock landing point is generated based on the region attributes and the edge line status.

[0013] In some embodiments, determining the boundary state corresponding to the target contact point based on the positional relationship between the rule-defined field coordinates and the field edge lines includes: Based on the rules and field coordinates, the corresponding edge detection objects are determined, and the set of outer frame edge lines corresponding to the main competition area is obtained. Calculate the distance between the coordinates of the regular site and each edge in the set of outer frame edges to obtain edge distance data; Based on the comparison between the edge distance data and the preset pressure threshold, the pressure state corresponding to the target contact point is determined; Based on the pressure line state, generate the edge line state information corresponding to the target contact point.

[0014] In some embodiments, generating rule-based judgment information based on the badminton shuttlecock landing point result, constructing corresponding evidence replay data, and outputting the landing point judgment result includes: Based on the badminton shuttlecock landing point result, the corresponding rule judgment type is determined, and rule judgment information is generated; the rule judgment information includes the landing point area, the sideline status, and the judgment basis. Based on the video frame data and site mapping data corresponding to the ground contact time, ground contact point playback data and rule scene mapping data are generated respectively. Based on the connection and fusion of the touch point playback data and the rule scene mapping data, a dual-channel evidence playback data is constructed. The badminton shuttlecock landing point determination result is output by integrating the dual-channel evidence playback data with the rule determination information.

[0015] To achieve the above objectives, another aspect of this application proposes a badminton shuttlecock landing point determination and evidence collection system based on visual trajectory, the system comprising: The data acquisition module is used to acquire sparsely sampled visual trajectory data corresponding to badminton videos; The spatial mapping module is used to establish a mapping relationship between the image space and the regular site space based on the site calibration parameters and the sparse sampling trajectory data. The ground contact detection module is used to perform ground contact candidate detection on the sparse sampled trajectory data, and determine the ground contact time by combining the detection step size adjustment strategy and the ground contact state determination mechanism. The touch point recognition module is used to identify the touch point based on the trajectory motion characteristics corresponding to the touch moment, and obtain the target touch point; The landing point determination module is used to perform coordinate transformation based on the target contact point and the mapping relationship, and to perform regular area matching to obtain the badminton shuttlecock landing point result; The evidence output module is used to generate rule judgment information based on the badminton shuttlecock landing point result, construct corresponding evidence playback data, and output the landing point judgment result.

[0016] To achieve the above objectives, another 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 above-described method.

[0017] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0018] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0019] The embodiments of this application include at least the following beneficial effects: This application provides a method and system for determining and proving badminton landing points based on visual trajectories. This scheme obtains sparsely sampled visual trajectory data corresponding to badminton videos and establishes a mapping relationship between image space and regular court space by combining court location parameters, thereby realizing the association between video spatial trajectory and actual match court coordinates and improving the accuracy of badminton trajectory analysis and landing point positioning; by performing ground contact candidate detection on the visual trajectory and determining the ground contact time by combining the detection step size adjustment strategy and ground contact state determination mechanism, it can reduce the impact of sampling interval and movement speed in complex motion states. The system addresses the deviation in contact time caused by factors such as changes, improving the stability of landing point detection. Furthermore, it identifies the contact point based on the trajectory motion characteristics corresponding to the contact time, and determines the badminton landing point through coordinate transformation and rule area matching. This achieves automated judgment of the landing point area, reducing subjective errors caused by manual observation and improving judgment efficiency and consistency. Simultaneously, it generates rule judgment information based on the landing point results and constructs evidence replay data, enabling traceability and visual evidence presentation of the landing point judgment process. This facilitates referee review, match analysis, and dispute resolution, thereby improving the accuracy, real-time performance, and reliability of the intelligent judgment system for badminton matches. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for determining and providing evidence of a badminton shuttlecock landing point based on visual trajectory, provided in an embodiment of this application. Figure 2 yes Figure 1 A flowchart illustrating step S2 in the process; Figure 3 yes Figure 1 A flowchart illustrating step S3 in the process; Figure 4 yes Figure 1 A flowchart illustrating step S4 in the process; Figure 5 yes Figure 1 A flowchart illustrating step S5 in the process; Figure 6 This is a schematic diagram of a badminton shuttlecock landing point determination and evidence collection system based on visual trajectory provided in an embodiment of this application; Figure 7 This is a schematic diagram of the state transition of the grounding state machine provided in an embodiment of this application; Figure 8 This is a schematic diagram of the combined evaluation of the pre-touch fall tail freeze and the in-field hit segment provided in the embodiments of this application; Figure 9 This is a schematic diagram of the main competition area, the outer contact area, and the line-judging distance provided in the embodiments of this application; Figure 10This is a schematic diagram illustrating the linkage between the detection step size and the ground-touching access control parameters provided in the embodiments of this application. Detailed Implementation

[0021] 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 of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0022] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0023] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0024] 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.

[0025] This application provides a method and system for determining and providing evidence of badminton landing points based on visual trajectories. This method acquires sparsely sampled visual trajectory data corresponding to badminton videos and establishes a mapping relationship between image space and regular court space by combining court location parameters. This establishes a connection between the video space trajectory and the actual match court coordinates, improving the accuracy of badminton trajectory analysis and landing point positioning. By performing ground contact candidate detection on the visual trajectory and combining a detection step size adjustment strategy and a ground contact state determination mechanism to determine the ground contact time, it can reduce the deviation in ground contact time caused by factors such as sampling intervals and changes in movement speed under complex motion conditions, improving the stability of landing point detection. Furthermore, based on the trajectory motion characteristics corresponding to the ground contact time, it identifies the ground contact point and determines the badminton landing point result through coordinate transformation and regular region matching, achieving automated determination of the landing point area, reducing subjective errors caused by manual observation, and improving determination efficiency and consistency. Simultaneously, it generates rule determination information based on the landing point result and constructs evidence playback data, realizing traceability and visual evidence presentation of the landing point determination process, facilitating referee review, match analysis, and dispute resolution, thereby improving the accuracy, real-time performance, and reliability of the intelligent badminton match determination system.

[0026] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0027] Figure 1 This is an optional flowchart of a method for determining and proving the landing point of a badminton shuttlecock based on visual trajectory, provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S1 to S6: S1: Obtain sparse sampled visual trajectory data corresponding to the badminton video; the sparse sampled visual trajectory data includes video frame identifier data, ball center position coordinate data, detection time series data, and trajectory confidence data.

[0028] In this embodiment, video frames to be detected are read from the real-time video stream captured by a monocular camera or from locally stored video files according to a pre-configured detection step size, and the badminton trajectory inference algorithm is invoked to perform ball detection on the corresponding video frames. The detection process does not require inference to be performed on every single frame of the video; instead, video frames are selected at detection step size intervals. For example, with a video frame rate of 30 frames per second and a detection step size of 3, ball detection is performed every 3 frames to obtain discrete ball center sampling points reflecting the badminton motion process while reducing computational resource consumption.

[0029] For each detected badminton target, the corresponding video frame identifier, detection time, and the horizontal and vertical coordinates of the shuttlecock's center in the image coordinate system are recorded. The target confidence level output by the detection algorithm is used as the trajectory confidence level of the corresponding trajectory point, thus forming sparsely sampled visual trajectory data in chronological order. For missed frames caused by motion blur, occlusion, or unstable detector output, no manual interpolation is performed on the missed locations, nor is a virtual shuttlecock center point constructed. Instead, the missing state at that detection time is retained, allowing subsequent ground contact determination to be processed based on the actual detection results and the actual sampling interval, thereby avoiding interference from interpolated trajectories on the actual ground contact motion characteristics.

[0030] Furthermore, before the trajectory points enter the subsequent touchdown determination process, obviously abnormal detection points can be screened based on trajectory confidence and the continuity of motion between adjacent valid trajectory points. For example, detection points that have undergone abnormally large positional jumps relative to the previous valid trajectory point and significantly deviate from the recent normal movement step length can be excluded from valid touchdown analysis samples; trajectory points that have been concentrated in a fixed area for a long time and lack a continuous falling process can also be identified as stationary ball interference, thereby reducing the impact of stationary badminton shuttlecocks and detected jumps on the subsequent touchdown state determination in multi-ball scenarios.

[0031] S2: Based on the field calibration parameters and sparse sampling trajectory data, establish the mapping relationship between the image space and the regular field space; Among them, reference Figure 2 As shown, based on the site calibration parameters and sparse sampled trajectory data, a mapping relationship between the image space and the regular site space is established, including: S21: Based on the badminton court corner point calibration data, determine the homography transformation parameters between the image space and the regular court space; S22: Based on the homography transformation parameters and the field corner point calibration results, determine the main playing area and the outer contact area corresponding to the regular field space; the outer contact area is used for contact observation access control, and the main playing area is used for in-bounds, out-of-bounds and line-pressing judgments. S23: Based on the homography transformation parameters, coordinate mapping is performed on sparsely sampled visual trajectory data to obtain trajectory coordinate data in a regular field space; S24: Based on the main competition area, the extended contact area, and the trajectory coordinate data, establish a mapping relationship between the image space and the regular field space.

[0032] In this embodiment, corner point calibration data for a badminton court is acquired. This data includes the pixel coordinates of the four corner points in the video image and the regular coordinates of each corner point within a standard badminton court. Homography transformation parameters between the image space and the regular court space are determined based on the correspondence between the corner points, enabling the court location in the original video image to be transformed to a standard bird's-eye view of the court. These homography transformation parameters eliminate the influence of camera angle and perspective distortion on subsequent trajectory position determination.

[0033] Based on homography transformation parameters and court corner calibration results, a main playing area is determined within the regular court space. This main playing area is then extended outwards by a predetermined distance along its boundary to form an extended touchdown area. The main playing area corresponds to the official court boundary and is used for subsequent in-bounds, out-of-bounds, and line-touching judgments. The extended touchdown area is only used to determine whether a shuttlecock has entered the touchdown observation range and whether a touchdown status assessment should be initiated. For example, the area can be extended outwards proportionally to the court width, allowing shuttlecocks near the sidelines to enter the touchdown observation process before their final in-bounds / out-of-bounds attribute is determined, thus achieving spatial decoupling between the touchdown observation area and the competition rule penalty area.

[0034] The effective sphere center image coordinates are extracted from the sparsely sampled visual trajectory data according to the detection time sequence, and then transformed to a regular field space one by one using homography transformation parameters to obtain the corresponding trajectory coordinate data. After coordinate transformation, each trajectory point still retains the correspondence with the original video frame, detection time, and trajectory confidence; no additional coordinate interpolation is performed for missed detection times to maintain the original sparse sampling characteristics of the trajectory data.

[0035] By associating the trajectory coordinate data of the main playing area, the extended touchdown area, and the regular playing area, a mapping relationship is established between the image space and the regular playing area. Based on this mapping relationship, the position of any valid ball center can be converted from image coordinates to regular playing area coordinates, and it can be further determined whether the corresponding trajectory point enters the extended touchdown area and whether the final touchdown point is located in the main playing area or an adjacent playing area sideline. Therefore, subsequent touchdown detection uses the extended touchdown area as the observation gate, while the landing point judgment uses the main playing area as the rule benchmark, improving the sensitivity of edge ball touchdown detection while maintaining the accuracy of the playing area judgment boundary.

[0036] S3: Perform ground contact candidate detection on sparse sampled trajectory data, and determine the ground contact time by combining the detection step size adjustment strategy and the ground contact state determination mechanism; Among them, reference Figure 3As shown, ground contact candidate detection is performed on sparsely sampled trajectory data. Combining a detection step size adjustment strategy and a ground contact state determination mechanism, the ground contact time is determined, including: S31: Extract the trajectory change features of badminton shuttlecocks based on sparse sampled trajectory data and generate ground contact candidate detection sequences; S32: Based on the difference between the current detection step size and the preset baseline step size, adjust the cumulative hit threshold, missed detection tolerance threshold and observation waiting parameters corresponding to the ground contact detection to generate step size-adapted detection parameters; S33: Based on the step size adaptation detection parameters, the state of the candidate ground contact detection sequence is determined to determine the ground contact observation state; wherein, before entering the ground contact observation state, when the center of the badminton shuttlecock is detected to be in the outer ground contact area, the number of valid hits is accumulated, and when a missed detection occurs, the accumulated number of valid hits is maintained; when the number of valid hits meets the accumulated hit threshold, the ground contact observation state is determined to be entered. S34: Detect candidate ground contact moments based on motion trajectory segments corresponding to the ground contact observation state, and determine the ground contact moment based on the motion characteristics corresponding to the candidate moments.

[0037] like Figure 7 As shown, Figure 7 This describes the transition of the touch-to-ground state machine from the standby state to the touch-to-ground observation state. At the start of each round, the touch-to-ground state enters the standby state, and the effective trajectory points entering the expanded touch-to-ground area are accumulated for hits. When a missed detection occurs at the corresponding detection moment, only the missing state is retained, no interpolation is performed, and the accumulated hit count is not reset to zero. In the standby state, the normal path, the kill path, and the sticky path are triggered and judged in parallel. The normal path is mainly judged based on the accumulated hit threshold and vertical downward movement characteristics; the kill path is mainly judged based on a low accumulated threshold and high-speed falling characteristics; and the sticky path is mainly judged based on low displacement aggregation and weak falling characteristics. When any path meets the corresponding trigger condition, the touch-to-ground state is transitioned from the standby state to the touch-to-ground observation state.

[0038] After entering the ground contact observation state, the system continues to wait for several detection steps to supplement the trajectory samples before and after the ground contact. Based on the formed motion trajectory segments, multiple candidate ground contact moments are extracted. These candidate moments are then comprehensively evaluated using motion characteristics such as deceleration, rebound, convergence, and displacement stability to determine the target ground contact candidate moment. When the ground contact evidence meets the confirmation conditions, or when the observation window closes due to conditions such as the sphere leaving the outer ground contact area or consecutive missed detections exceeding the tolerance threshold, the corresponding ground contact moment is output. After the ground contact moment is determined, a preset short cooling phase begins. During this cooling phase, new ground contact triggers are temporarily not accepted to avoid the same ground contact event being repeatedly identified due to rebound, stillness, or trajectory jitter. After the cooling phase ends, the next round of ground contact detection resumes.

[0039] like Figure 8 As shown, Figure 8 This describes the process of combining the pre-touch tail segment freeze with the in-court segment evaluation. Sparse sampling trajectory points of the shuttlecock are acquired along the time axis. Hollow circles represent validly detected shuttlecock center trajectory points, crosses indicate missed detections at corresponding detection moments, and solid circles represent the trajectory points corresponding to the shuttlecock's first entry into the outer contact area. For missed detection moments, only the missing state is retained; no coordinate interpolation is performed, and no virtual shuttlecock center point is constructed. Before the shuttlecock first enters the outer contact area, valid trajectory points with a downward trend are continuously collected, forming the pre-touch tail segment. When the shuttlecock's center is detected to have entered the outer contact area for the first time, the pre-touch tail segment is frozen and saved, and valid trajectory points within the outer contact area continue to be collected, forming the in-court hit segment.

[0040] Furthermore, the frozen pre-contact descent tail segment and the in-field hit segment are merged according to the detection time sequence to construct a unified evaluation kinematic sequence. This allows the ground contact candidate evaluation to simultaneously retain evidence of continuous descent before ground contact as well as motion change information such as deceleration, rebound, convergence, or low-displacement stabilization after ground contact. Based on the unified evaluation kinematic sequence, motion features such as vertical coordinate extrema, local velocity changes, and trajectory convergence are extracted. These features are then combined with deceleration, rebound, and displacement stabilization states before and after the candidate moment to evaluate ground contact evidence, thereby determining the ground contact candidate moment and the corresponding ground contact moment.

[0041] In this embodiment, based on sparsely sampled trajectory data, trajectory change features such as the change in the shuttlecock's center position, vertical displacement, step length of adjacent trajectory points, and direction of movement are extracted according to the detection time sequence to generate a candidate ground contact detection sequence. Before the shuttlecock enters the outward-expanding ground contact area, the tail segment of the pre-ground contact trajectory with a downward trend is continuously maintained; when the shuttlecock's center is first detected to enter the outward-expanding ground contact area, this tail segment is frozen and associated with subsequent in-court trajectory segments. For intermediate missed detection moments, no interpolation is performed to fill in the missing points, so as to preserve the original sparse trajectory and pre-ground contact falling evidence.

[0042] Furthermore, based on the difference between the current detection step size and the preset baseline step size, the cumulative hit threshold, the missed detection tolerance threshold, and the observation waiting parameters are adjusted in a coordinated manner to obtain step size-adapted detection parameters. When the detection step size increases, the cumulative hit count required to enter ground contact observation is reduced accordingly, and the missed detection tolerance count and waiting steps are adjusted; when the detection step size decreases, the cumulative hit requirement is appropriately increased to keep the actual observation time under different detection step sizes basically consistent.

[0043] like Figure 10 As shown, Figure 10This indicates the linkage between the detection step size and the grounding threshold parameters. As the detection step size gradually increases from a small value, the fixed cumulative threshold remains unchanged, while the linked cumulative threshold and the number of observation waiting steps adopted in this application decrease accordingly with the increase of the detection step size. For example, when the detection step size is adjusted from once every 3 frames to once every 6 frames, the cumulative hit threshold required to enter the grounding observation state can be lowered, and the number of observation waiting steps can be reduced simultaneously, thereby avoiding the difficulty in triggering grounding observation in a timely manner due to the increase in sampling interval and the reduction in the number of effective detection points. By linking and adjusting the cumulative hit threshold, the missed detection tolerance threshold, and the observation waiting parameters according to the current detection step size, the grounding threshold conditions under different sampling densities can be kept in line with the actual time scale.

[0044] Furthermore, compared to a fixed cumulative threshold, this application achieves this by dynamically adapting the detection step size to the access control parameters. This allows for maintaining the effective time range required for ground contact observation even when the detection frequency and computational load are reduced. For example, as the detection step size increases, even if the number of waiting steps decreases accordingly, the corresponding video time interval remains close to the original detection configuration. This avoids triggering delays or missed ground contact detections caused by a fixed threshold under sparse sampling conditions, achieving a balance between computational resource consumption and ground contact detection sensitivity.

[0045] Subsequently, the state of the candidate ground-touching detection sequence is determined based on the step-size adaptation detection parameters. Before entering the ground-touching observation state, when the center of the badminton shuttlecock is detected to be within the extended ground-touching area, the number of valid hits is incremented. If subsequent missed detections occur due to motion blur, occlusion, or other reasons, the currently accumulated number of valid hits is maintained and not reset or decremented. As valid trajectory points continue to enter the extended ground-touching area, when the accumulated number of valid hits reaches the accumulated hit threshold corresponding to the current detection step size, the ground-touching observation threshold condition is met, and the ground-touching state is changed from the standby state to the ground-touching observation state. This process avoids repeated interruptions in the accumulation process due to occasional missed detections under interval frame detection conditions.

[0046] After entering the ground contact observation state, trajectory data corresponding to several detection steps are collected, and the pre-contact descent tail segment is merged with the in-field trajectory segment to form a ground contact evaluation sequence. Candidate ground contact moments are extracted from this sequence, including longitudinal extreme points, local peak points, and knee points formed by the gradual convergence of movement step lengths. These are then combined with motion characteristics such as bounce, deceleration, convergence, and adhesion stability for a comprehensive judgment. The candidate with the most sufficient ground contact evidence is selected as the ground contact moment. If observation window closing signals occur, such as the ball leaving the outward-expanding ground contact area, obtaining effective post-contact sampling, or continuous missed detections exceeding the tolerance threshold, the current optimal candidate is determined as the ground contact moment.

[0047] Specifically, candidate moments of ground contact are detected based on the motion trajectory segments corresponding to the observed ground contact state, and the ground contact moment is determined based on the motion characteristics corresponding to the candidate moments, including: Based on motion trajectory segments, extract vertical coordinate extreme points, local velocity change points, and trajectory convergence points to generate a set of candidate ground contact moments; Based on the set of candidate ground contact moments, the corresponding pre-contact deceleration features, post-contact rebound features, trajectory convergence features, and displacement stability features are extracted to obtain candidate evaluation features. Based on the candidate evaluation features, the set of ground contact candidate times is scored and ranked to determine the target ground contact candidate time; The target touchdown candidate time is verified by combining the touchdown state boundary conditions. When the shuttlecock center is detected to have left the outward touchdown area, a valid post-touch sample is obtained, or the number of consecutive missed detections exceeds the missed detection tolerance threshold, the touchdown observation window is closed, and the target touchdown candidate time with the highest score is determined as the touchdown time.

[0048] In this embodiment, after entering the touchdown observation state, the frozen pre-touch fall tail segment and the in-field trajectory segment are merged in chronological order to form the motion trajectory segment corresponding to the touchdown analysis. Based on the motion trajectory segment, the positions where the vertical coordinates of the ball center reach extreme values, the positions where local motion velocity changes significantly, and the convergence positions formed by the gradual decrease of the trajectory step length during the fall are detected. Vertical coordinate extreme points, local velocity change points, and trajectory convergence points are obtained, and their corresponding times are combined into a set of touchdown candidate times. In this embodiment, the maximum vertical coordinate point, the local vertical peak point, and the velocity convergence knee point are used as touchdown candidates.

[0049] Furthermore, trajectory change characteristics before and after each candidate touchdown moment are extracted. Pre-touch deceleration characteristics are obtained by observing whether the velocity or step length of adjacent trajectories gradually decreases before the candidate moment; post-touch rebound characteristics are obtained by observing whether the ball's center of motion changes from downward to upward after the candidate moment; and displacement stability characteristics are obtained by observing whether multiple trajectory points form a low-displacement cluster after touchdown. For cases where the rebound is not significant, the characteristic of the trajectory converging from a high-speed descent to a low-speed stable state is also used for auxiliary evaluation.

[0050] Subsequently, a comprehensive score was calculated based on the pre-contact deceleration characteristics, post-contact rebound characteristics, and displacement stability characteristics corresponding to each candidate moment. The candidate moments were then ranked according to their scores, and the candidate with the highest score was selected as the target contact moment. For contact trajectories with obvious rebound, the evaluation value of rebound evidence can be improved; for cases of slight rebound, sliding, or sticking, the contact moment determination is supported by deceleration, trajectory convergence, and displacement stability characteristics, thereby reducing the reliance on a single rebound characteristic.

[0051] Finally, the target touchdown candidate moments are validated based on the touchdown state boundary conditions. When it is detected that the shuttlecock's center has left the extended touchdown area, a valid trajectory sample has been obtained after touchdown, or the number of consecutive missed detections exceeds the missed detection tolerance threshold corresponding to the current step size, the current touchdown observation window is determined to be closed. At this point, there is no need to wait for new soft evidence; instead, the target touchdown candidate moment with the highest current score is determined as the final touchdown moment from the candidate set that has already passed the kinematic condition screening. This avoids long-term state machine waiting or no output results for the touchdown event due to incomplete touchdown evidence.

[0052] Specifically, the acquisition of motion trajectory segments includes: Before the shuttlecock's center enters the outward-expanding contact area for the first time, trajectory sampling points with a downward trend are continuously extracted from sparsely sampled visual trajectory data to obtain the pre-contact falling tail segment data. In response to the first detection that the center of the badminton shuttlecock has entered the outward contact area, the data of the final segment of the fall before contact is frozen, and the trajectory segment within the area after the center of the badminton shuttlecock enters the outward contact area is obtained. The data of the falling tail segment before impact is merged with the trajectory segment within the region according to the detection time series to obtain the motion trajectory segment.

[0053] In this embodiment, before the shuttlecock's center enters the outward-expanding contact area for the first time, sparsely sampled visual trajectory data is continuously read according to the detection time sequence, and the changes in the center position of adjacent valid trajectory points are analyzed. When multiple valid trajectory points show a continuous trend of moving towards the ground, the corresponding trajectory sampling points are sequentially written into the pre-contact fall tail buffer to obtain the pre-contact fall tail data. For intermediate missed detections caused by motion blur or occlusion, virtual trajectory points are not inserted, and the already saved fall trajectory is not deleted due to a single missed detection, thereby preserving the real fall process of the shuttlecock before entering the contact area as much as possible under the condition of interval frame detection.

[0054] When the shuttlecock's center is first detected entering the extended contact area, this moment is designated as the boundary between the pre-contact trajectory and the trajectory within the area. The current pre-contact fall tail segment data is immediately frozen, preventing the saved trajectory sampling points from being overwritten or updated by subsequent data. Simultaneously, starting from the first valid trajectory point entering the extended contact area, the shuttlecock's center trajectory within the extended contact area is continuously acquired, forming a trajectory segment within the area. Therefore, even if only a few valid sampling points are obtained after the shuttlecock enters the extended contact area, the frozen pre-contact fall tail segment can still supplement the pre-contact motion information, avoiding insufficient contact evidence due to sparse trajectories within the area.

[0055] Based on the detection time corresponding to each trajectory sampling point, the frozen pre-touch fall tail segment data is stitched together with the trajectory segments within the region in chronological order to form a continuous motion trajectory segment. Information such as the video frame identifier, detection time, and shuttlecock center position corresponding to each trajectory point is retained. The motion trajectory segment simultaneously includes the shuttlecock's fall process before entering the extended contact area and its near-ground motion process after entering, which can be used to subsequently extract contact candidate features such as vertical coordinate extreme points, local velocity change points, and trajectory convergence points. By combining the pre-touch tail segment with the trajectory within the region, the impact of fragmented pre-touch motion information caused by sparse sampling on contact time identification can be reduced.

[0056] S4: Identify the point of contact based on the trajectory motion characteristics corresponding to the moment of contact, and obtain the target point of contact; Among them, reference Figure 4 As shown, the point of contact is identified based on the trajectory motion characteristics corresponding to the moment of contact, resulting in the target point of contact, including: S41: Extract the shuttlecock's center position change features based on the trajectory segment corresponding to the moment of contact with the ground, and generate a candidate set of contact points; S42: Based on the candidate set of contact points, extract the corresponding vertical extreme value features, velocity change features and trajectory convergence features to generate contact point evaluation features; S43: Based on the evaluation features of the contact point, match and filter the candidate set of contact points to determine the candidate location of the target contact point; S44: Based on the candidate locations of the target contact point and the image coordinate information corresponding to the contact time, position correction is performed to obtain the target contact point.

[0057] In this embodiment, after determining the moment of contact with the ground, multiple valid ball center positions are extracted from the motion trajectory within a predetermined range before and after the moment of contact. The positional changes of the ball center in the vertical direction and between adjacent sampling points are analyzed chronologically to form a candidate set of contact points. This candidate set may include the ball center position corresponding to the moment of contact, the nearest vertical coordinate extreme value position, and the position where the trajectory transitions from a rapid descent to a low displacement state, in order to avoid contact point offset caused by using only a single sampling point.

[0058] Furthermore, evaluation features of the contact point are extracted for each candidate position in the candidate set of contact points. Specifically, vertical extreme value features are obtained based on whether the candidate position corresponds to the vertical coordinate extreme value during the shuttlecock's descent; velocity change features are obtained based on whether the velocity of adjacent trajectory points before and after the candidate position significantly decreases or changes direction; and trajectory convergence features are obtained based on whether the step length of the shuttlecock's center of motion gradually decreases from a large state and forms a low displacement state near the ground. For contact processes with insignificant rebound, velocity convergence and low displacement stability features can be used to supplement vertical extreme value evidence to adapt to contact patterns such as slight bounce or sticky ground.

[0059] Subsequently, the vertical extreme value characteristics, velocity change characteristics, and trajectory convergence characteristics corresponding to each candidate position are comprehensively matched to eliminate candidate positions that only exhibit trajectory jitter, isolated jumps, or lack evidence of pre-impact descent. Positions that correspond temporally to the determined impact time and simultaneously possess clear characteristics of descent termination or motion state transition are prioritized for retention, thus determining the target impact point candidate positions. If multiple candidates meet the requirements, candidates with a smaller temporal distance from the impact time and more complete motion change characteristics are prioritized to improve the consistency between the impact point and the impact time.

[0060] Finally, based on the candidate locations of the target contact points, the original sparse sampling trajectory data corresponding to the contact time is traced back to obtain the real image sphere center coordinates corresponding to that location. These coordinates are then corrected by combining the positions with adjacent valid sampling points before and after the contact time. For missed contact times, instead of constructing new contact points through manual interpolation, the real detection coordinates closest to the contact time and satisfying the contact motion characteristics are prioritized as the correction basis, ultimately yielding the target contact point. This maintains the correspondence between the target contact point and the original video frames, providing a traceable image location basis for subsequent regular field coordinate transformation and evidence playback.

[0061] S5: Based on the target contact point and mapping relationship, perform coordinate transformation and regular area matching to obtain the badminton shuttlecock landing point result; Among them, reference Figure 5 As shown, coordinate transformation is performed based on the target contact point and mapping relationship, and regular region matching is performed to obtain the badminton shuttlecock landing point result, including: S51: Based on the image coordinates and mapping relationship corresponding to the target contact point, perform coordinate mapping to obtain the coordinates of the regular field; S52: Based on the coordinates of the rule field and the main competition area, spatial matching is performed to determine the area attributes corresponding to the target touch point; S53: Determine the boundary based on the positional relationship between the regular field coordinates and the field edge lines to determine the edge line status corresponding to the target contact point; S54: Generate badminton shuttlecock landing point results based on region attributes and sideline status.

[0062] like Figure 9 As shown, Figure 9This describes the relationship between the main playing area, the extended touch zone, and the line-judging distance. Based on the court's four corner markings, the main playing area is determined within the standard court space, and an extended touch zone is formed around it. The extended touch zone is used to determine whether the shuttlecock has entered the touch zone, while the main playing area is used to determine the in-bounds, out-of-bounds, and scoring attributes of the target touch point. They perform different judgment functions. After the target touch point undergoes a standard court coordinate transformation, its area attributes are first determined based on its positional relationship with the main playing area. Then, line-judging is performed on the four outer edge lines corresponding to the main playing area. The center line and service line are not included in the line-judging to avoid misjudgments caused by the internal standard lines.

[0063] Furthermore, the shortest distance between the rule coordinates corresponding to the target touch point and the adjacent outer edge line is calculated, and this shortest distance is compared with a preset line-crossing threshold. The line-crossing threshold can be determined by considering the edge line width, the equivalent range of the shuttlecock in a bird's-eye view, and the judgment margin. When the shortest distance is not greater than the preset line-crossing threshold, the target touch point is judged as being on the line; when the shortest distance is greater than the preset line-crossing threshold, it is further determined as in-bounds or out-of-bounds depending on whether the target touch point is located within the main playing area. Thus, by distinguishing the touchdown observation area, the playing penalty area, and the line-crossing judgment range, the observation range of edge touchdown events can be expanded while maintaining the accuracy of the playing rule boundaries.

[0064] In this embodiment, the coordinates of the center of the sphere at the target contact point in the original video image are obtained. Then, a pre-established mapping relationship between the image space and the regular field space is invoked to transform the image coordinates to a standard bird's-eye view field coordinate system, yielding the corresponding regular field coordinates. This transformation eliminates the influence of camera angle and perspective distortion on the determination of the landing point, enabling the target contact point to be determined according to a unified standard field scale for subsequent rule-based determination.

[0065] The system spatially matches the coordinates of the rule-defined playing area with the main playing area to determine whether the target's point of contact is inside or outside the main playing area, thus assigning it the corresponding area attribute. If the coordinates are inside the main playing area, the area attribute is determined to be within bounds; if they are outside the main playing area, it is initially determined to be out of bounds. Furthermore, the position of the coordinates in the longitudinal direction of the playing area can be used to further determine the half of the court to which the target's point of contact belongs, providing spatial basis for generating complete landing point determination information.

[0066] Boundary determination is based on the positional relationship between the standard court coordinates and the court sidelines. Specifically, the distance relationship between the target contact point and the corresponding sideline is determined only for the four sidelines of the standard court's outer frame. This distance is then combined with the sideline width, the equivalent range under the top-view view of the shuttlecock, and a preset judgment margin to determine whether a line is crossed. If the target contact point is within the line-crossing judgment range of the corresponding sideline, the sideline status is determined to be crossed; otherwise, it is determined to be not crossed. The center line and service line are not considered for line-crossing judgment during this process, thus avoiding false line-crossing results caused by internal standard lines.

[0067] The system generates badminton shuttlecock landing points based on a combination of area attributes and sideline status. If the target contact point meets the line-on condition, a line-on landing point is prioritized; if it does not meet the line-on condition and is located within the main playing area, an in-bounds landing point is generated; if it is located outside the main playing area, an out-of-bounds landing point is generated, and the system associates the target contact point's half of the court with the rule-defined court location. This creates structured landing point information that includes in-bounds, line-on, and out-of-bounds statuses, providing clear geometric criteria for subsequent rule-based judgments and evidence replay.

[0068] Specifically, boundary determination is performed based on the positional relationship between the regular field coordinates and the field edge lines to determine the edge line state corresponding to the target contact point, including: Based on the coordinates of the rule field, determine the corresponding edge detection objects and obtain the set of outer frame edge lines corresponding to the main competition area; Calculate the distances between the site coordinates and the edges in the outer frame edge set to obtain edge distance data; The line pressure status corresponding to the target contact point is determined by comparing the edge distance data with the preset line pressure threshold. Generate edge line status information corresponding to the target contact point based on the pressure line status.

[0069] In this embodiment, the sideline detection object is first determined based on the regular court coordinates corresponding to the target contact point, and the set of outer frame sidelines corresponding to the main playing area is obtained. Specifically, the four outer sidelines of the main playing area are used as boundary determination objects, and the center line and service line are not included in the line detection range to avoid misjudgment when the target contact point is close to the inner regular lines. For the target contact point, the outer frame sidelines that are adjacent to it in spatial position can be determined first, while the four outer frame sidelines are retained for uniform distance comparison.

[0070] Furthermore, the shortest spatial distance between the coordinates of the regular playing field and each edge line in the set of outer frame edge lines is determined to obtain the corresponding edge line distance data. From this data, the outer frame edge line closest to the target touch point is identified. This distance calculation is performed on actual edge line segments of finite length, rather than extending the edge lines infinitely. This ensures that the edge line distance accurately reflects the positional relationship between the target touch point and the actual playing boundary, providing a geometric basis for subsequent line-pressing judgments.

[0071] Subsequently, the sideline distance data is compared with a preset line-pressing threshold. The line-pressing threshold is set by considering the width of the court sidelines, the equivalent range of the shuttlecock in a bird's-eye view, and the allowable judgment margin. When the distance between the target contact point and any outer edge line is not greater than the line-pressing threshold, it is determined to be in a line-pressing state; when the distance to all sidelines exceeds the line-pressing threshold, it is determined to be in a non-line-pressing state. This method incorporates the space occupied by the shuttlecock itself into the boundary judgment, rather than judging solely based on whether the center of the shuttlecock lands exactly on the sideline.

[0072] Finally, the sideline status information corresponding to the target contact point is generated based on the line-pressing status. When it is determined to be line-pressing, the line-pressing status and the corresponding nearest outer edge line are recorded; when it is determined to be non-line-pressing, the non-line-pressing status is recorded, and this result is combined with the area attribute of the target contact point relative to the main playing area for subsequent determination of whether the landing point is in or out of bounds. This forms a clear sideline determination result and avoids false line-pressing events caused by the center line and service line participating in the determination.

[0073] S6: Generate rule judgment information based on the badminton shuttlecock landing point result, construct corresponding evidence playback data, and output the landing point judgment result.

[0074] Specifically, rule-based judgment information is generated based on the badminton shuttlecock's landing point, and corresponding evidence playback data is constructed. The landing point judgment results are then output, including: S61: Determine the corresponding rule judgment type based on the badminton shuttlecock landing point result, and generate rule judgment information; the rule judgment information includes the landing point area, the sideline status, and the judgment basis. S62: Based on the video frame data and site mapping data corresponding to the ground contact time, generate ground contact point playback data and rule scene mapping data respectively; S63: Based on the correlation and fusion of touch point playback data and rule scenario mapping data, construct dual-channel evidence playback data; S64: Based on the dual-channel evidence playback data and rule judgment information, integrate and process the data to output the badminton shuttlecock landing point judgment result.

[0075] In this embodiment, the corresponding rule judgment type is first determined based on the badminton shuttlecock's landing point. Specifically, according to the half of the court where the target contact point belongs, its positional relationship with the main playing area, and the status of the sidelines, the landing point is divided into in-bounds, on-line, or out-of-bounds, and corresponding rule judgment information is generated. This rule judgment information includes at least the landing point area, the sideline status, and the judgment basis, such as recording the half of the court where the target contact point belongs, whether it is located in the main playing area, whether it is within the outer sideline's on-line range, and the corresponding judgment reason. When additional rule events such as net crossing are detected, the regular landing point judgment can also be corrected, thereby generating an interpretable judgment record.

[0076] Furthermore, using the moment of contact as the time reference, continuous real video frames before, during, and after contact are extracted from the video buffer data. Contact point playback data is then generated with the target contact point as the observation center to demonstrate the shuttlecock's approach to the ground, its contact, and its post-contact motion. Simultaneously, based on the court mapping data, the target contact point is mapped onto a bird's-eye view of a regular court. The landing position, corresponding outer frame lines, and whether the shuttlecock is in or out of bounds are marked on a pre-generated master map of the regular court, generating regular scene mapping data.

[0077] Subsequently, the touchpoint playback data and rule-based scene mapping data are correlated along a unified timeline to construct dual-channel evidence playback data. One channel displays a slow-motion replay of the original video centered on the touchpoint, ensuring that each playback frame corresponds to the original video frame. The other channel displays a rule-based field map, showing the relative position of the touchpoint from a full-field perspective at the beginning of the playback, and then switching to a local zoom-in centered on the touchpoint, highlighting the nearest outer frame and corresponding judgment markers. By presenting the data simultaneously through the two channels, both original video evidence and rule-based spatial geometric evidence can be provided at the same time.

[0078] Finally, the dual-channel evidence playback data is correlated and integrated with the rule-based judgment information to form the badminton shuttlecock landing point judgment result, and outputs the target contact point, landing area, sideline status, judgment basis, and corresponding evidence playback content. The evidence data can be constructed in a processing path independent of the main inference process, and pre-constructed frames are provided to the playback end through a bounded queue. This ensures that the generation of evidence screens does not block subsequent video acquisition and shuttlecock detection, thereby maintaining the processing efficiency of the continuous ground contact detection process while ensuring the traceability and interpretability of the judgment result.

[0079] In some embodiments, different ground contact evidence modes can be configured according to the deployment scenario corresponding to the badminton shuttlecock landing point determination, including a lenient evidence mode and a strict evidence mode. The lenient evidence mode is suitable for real-time scoring and evidence replay scenarios where there is no interference from multiple stationary shuttlecocks. While maintaining the determination of ground contact motion characteristics, it can cancel or reduce the spatiotemporal suppression intensity for stationary shuttlecocks on the ground, and allow sticky ground triggers to pass a higher vertical span threshold when preset ground contact motion conditions are met. The strict evidence mode is suitable for multi-spinner training and statistical scenarios. It establishes corresponding spatiotemporal suppression regions for sticky ground trajectory clusters lacking effective pre-contact landing evidence, suppressing repeated ground contact triggers generated at the same location within a preset time window to reduce misjudgments caused by stationary shuttlecocks, repeated detections, and trajectory jumps.

[0080] In one embodiment, the present invention is applied to a real-time scoring scenario. The system takes real-time match video or local video files captured by a monocular camera as input and uses an interval frame method to acquire the sparsely sampled visual trajectory of the badminton shuttlecock, thereby reducing the computational resource consumption of continuous frame-by-frame detection. For example, the video frame rate is set to 30 frames per second, and the detection step size is set to perform badminton shuttlecock detection once every 3 frames. Before video processing begins, the four corner points of the court are first calibrated. Based on the corner point calibration results, a homography mapping relationship between the image space and the regular court space is established, and a main playing area and an extended touch-the-ground area relative to the main playing area are generated in the regular court space. The extended touch-the-ground area is used to trigger touch-the-ground observation, and the main playing area is used for subsequent in-bounds, out-of-bounds, and line-crossing judgments, thereby decoupling the touch-the-ground detection area from the rule-based penalty area.

[0081] During video processing, the coordinates of the shuttlecock's center are periodically acquired according to the detection step size, and sparse sampling trajectories are formed in chronological order. While the shuttlecock's center has not yet entered the outward-expanding contact area, trajectory sampling points with a downward trend are continuously extracted to form the pre-contact falling tail segment. When the shuttlecock's center is first detected entering the outward-expanding contact area, the pre-contact falling tail segment is frozen, and the number of valid hits within the outward-expanding contact area is accumulated. If a short-term missed detection occurs due to motion blur, occlusion, or other reasons, the accumulated hit count remains unchanged and is not reset due to a single missed detection, thus preventing the contact observation process from failing to start due to trajectory gaps under sparse sampling conditions. Taking a smash trajectory as an example, when the accumulated hit count reaches a trigger threshold adapted to the current detection step size, and the vertical downward movement of the pre-contact trajectory and the speed of movement near the ground meet the corresponding trigger conditions, the contact state is changed from a standby state to a contact observation state.

[0082] After entering the ground contact observation state, the system continues to wait for approximately two detection steps to collect supplementary trajectory data before and after ground contact. The frozen pre-contact descent tail segment and the in-field trajectory segment are then merged in chronological order to form a ground contact assessment sequence. Subsequently, multiple candidate ground contact moments are extracted from the ground contact assessment sequence, including vertical coordinate extreme points, local velocity change points, and convergence knee points formed by the gradual decrease in descent trajectory step length. The trajectory motion characteristics before and after each candidate moment are analyzed. Specifically, deceleration features are extracted based on whether the pre-contact velocity decreases; rebound features are extracted based on whether the post-contact motion direction changes from downward to upward; convergence features are extracted based on whether the descent trajectory gradually converges from a larger step length to a smaller step length; and viscous stability features are extracted based on whether a low-displacement aggregation state forms at the center of the ball after ground contact. Based on these features, each candidate ground contact moment is comprehensively scored, and the candidate with the highest evaluation result is selected as the target ground contact moment. When the target candidate meets the ground contact confirmation conditions, the corresponding ground contact moment is directly determined; for cases where the rebound is not significant, ground contact confirmation can be supported by convergence and viscous stability features. If the evaluation characteristics have not yet met the normal confirmation conditions, but there are already observation window closing signals such as the ball center leaving the outer contact area or continuous missed detections exceeding the tolerance threshold, then the highest valid candidate with the current evaluation result will be determined as the contact moment, in order to avoid the contact detection process having no output for a long time.

[0083] Once the moment of contact is determined, the target point of contact at that moment is obtained, and the image coordinates of the point of contact are converted to regular court coordinates based on the court homography mapping relationship. Then, the regular court coordinates are matched with the main playing area to determine the half of the court to which the target point of contact belongs and its in-bounds or out-of-bounds attribute. Only the four outer sidelines corresponding to the main playing area are considered for line-pressing judgment; the center line and service line are not included in the boundary detection range. When the positional relationship between the target point of contact and any outer sideline meets the line-pressing condition, it is judged as line-pressing; otherwise, it is determined as in-bounds or out-of-bounds based on its positional relationship with the main playing area. If net position calibration information is pre-configured, the trajectory before contact can be traced back. When a stable half-court crossing of the shuttlecock is detected and its trajectory passes under the net, a net crossing judgment is generated, and the regular landing point scoring result is corrected accordingly. Finally, a structured scoring result is generated, including the target half-court, landing point area, sideline status, scoring party, and the reason for the judgment.

[0084] Furthermore, when a further reduction in computational load is needed, the detection step size can be adjusted from once every 3 frames to once every 6 frames. At this point, based on the correlation between the detection step size and the ground contact parameter, the cumulative hit threshold required to enter the ground contact observation state is correspondingly reduced, for example, from approximately 4 hits to 2 hits. The waiting step after entering the ground contact observation state is also adjusted from 2 steps to 1 step, ensuring the actual observation time remains approximately 6 video frames. Thus, while reducing the badminton shuttlecock detection frequency, the ground contact observation parameter can synchronously adapt to changes in sampling density, reducing computational resource consumption and avoiding missed ground contact events due to reduced sampling points, achieving a balance between detection efficiency and ground contact determination sensitivity in real-time scoring scenarios.

[0085] In one embodiment, the present invention is applied to a scenario of reviewing disputed balls and demonstrating penalties. The touchdown detection process is consistent with that in Embodiment 1, with the video frame rate set to 30 frames per second, and sparse sampling trajectories obtained by performing badminton ball detection every 3 frames. After the touchdown state machine completes the touchdown confirmation and obtains the target touchdown point and corresponding landing point judgment result, the main inference path continues to execute subsequent video frame acquisition, ball detection, and trajectory tracking for the next round. At the same time, an independent evidence replay processing path is started to process the video data and rule field data corresponding to the current touchdown event. Thus, touchdown detection and evidence image generation are configured in different execution paths, avoiding the consumption of processing resources in the main inference process by evidence image synthesis, thereby ensuring that subsequent badminton ball detection is not interrupted due to the evidence task of the current round during continuous matches.

[0086] Specifically, in the evidence playback processing path, the determined touchdown moment is used as the time reference. Continuous source video frames—before, immediately after, and after the touchdown—are extracted from the video buffer data, and touchdown point playback data is generated according to a preset cutout length. For example, approximately 60 frames of video footage can be selected to form a left-side focus-locked slow-motion channel. During playback, the area near the target touchdown point is used as the primary observation range, allowing the continuous presentation of the shuttlecock's approach to the ground, its touchdown, and its subsequent movement. The slow-motion channel displays the touchdown process at a reduced playback speed, but each video frame in the cutout still corresponds to a real frame in the source video. Slowing down is not achieved by repeatedly inserting the same frame, thus maintaining the temporal correspondence between the video frames and the actual touchdown process and improving the traceability of the touchdown event.

[0087] Simultaneously, based on the site landmarking and the mapping relationship between image space and regular site space, the target touch point is transformed into a bird's-eye view of the regular site, and a right-side regular scene mapping channel is constructed using a pre-generated bird's-eye view master map of the site. In the first half of the playback, the overall position of the target touch point in the standard site is displayed from a full-view perspective, allowing viewers to intuitively determine its half of the field and its position relative to the site boundary. In the second half of the playback, a local magnified view centered on the target touch point is switched, highlighting the outer frame line closest to the target touch point, while overlaying rule judgment labels such as "within boundary," "on line," or "out of boundary." For example, in an evidence segment of approximately 60 frames, the view can switch from a full-view perspective to a local magnified view around frame 30, thus forming a continuous evidence presentation process from overall position display to detailed boundary judgment display.

[0088] Finally, the slow-motion playback channel for the left-side contact point and the mapping channel for the right-side rule-based scene are associated along a unified timeline to form dual-column evidence playback data. This data, along with the corresponding landing area, edge line status, and judgment criteria, is then output. The synthesis and playback of the dual-column images are completed within an independent evidence processing path, and pre-built frames are continuously provided to the playback window via a bounded queue, decoupling the evidence processing speed from the main video acquisition and badminton detection process. Even if the evidence image generation speed experiences short-term fluctuations, it will not back-block the main inference path or cause video acquisition buffer backlog. Therefore, original video contact evidence and rule-based field geometry evidence can be provided simultaneously without interrupting subsequent badminton detection, ensuring that the landing point judgment results possess both video process traceability and rule-based penalty interpretability.

[0089] In one embodiment, the present invention is applied to a training course half-court landing point statistics scenario. Considering that multiple stationary badminton shuttlecocks may exist simultaneously on the court during training, and that abnormal trajectories can easily be generated by students picking up shuttlecocks, personnel obstruction, and detection jumps, a strict evidence mode is adopted for ground contact detection. Before the shuttlecock's center trajectory enters the ground contact state determination, trajectory confidence screening is performed: when the displacement of the current detection point relative to the previous valid trajectory point exceeds a preset absolute displacement range and a number of times the recent normal movement step length, it is determined as an abnormal jump point and is removed; trajectory hotspots that are concentrated in a fixed grid area for a long time and whose frequency and proportion of occurrence both exceed the corresponding thresholds are also not included in the valid ground contact analysis. If stationary badminton shuttlecocks on the ground are continuously identified as low displacement trajectory clusters, and the trajectory clusters lack valid evidence of pre-contact landing, a spatiotemporal suppression region is established at the corresponding position, and repeated triggering in the same region is ignored within a suppression time window of approximately 1.8 seconds, thereby reducing the repeated counts caused by stationary shuttlecocks.

[0090] Furthermore, a main statistical region is constructed based on the half-court calibration results, and an extended contact area is formed around it at a ratio of approximately 3%. When the center of the shuttlecock is detected to have entered the extended contact area and accumulated enough hits to meet the corresponding conditions, the contact state is changed from standby to contact observation state. Before entering the contact observation window, the system continues to perform shuttlecock detection according to the normal detection step size; after entering the observation window, the detection frequency during the contact phase is temporarily increased based on the already configured linkage between the detection step size and the access control parameters. For example, the normal detection every 4 frames can be temporarily adjusted to detection every 2 frames to obtain a denser number of effective trajectory points within a limited time range before and after contact, improving the stability of the identification of contact candidate moments and target contact points. After the contact observation ends, the system reverts to the original normal detection step size. This temporary encryption process only applies to the current contact observation window and does not change the existing adaptation relationship between the normal detection step size and the access control threshold.

[0091] After a valid ground contact is detected and included in the half-court landing statistics, a preset cooldown period begins, for example, approximately 2.5 seconds. During this cooldown period, new shuttlecock ground contact detection is paused or suppressed, preventing stationary trajectories, repeatedly shaking trajectories, and duplicate detections occurring before the trainee has finished retrieving the shuttlecock from being recognized as new landing events. After the cooldown period ends, normal shuttlecock detection and the next ground contact statistics resume. By combining trajectory reliability gating, spatiotemporal suppression, and a post-ground contact cooldown mechanism, the interference of abnormal trajectories and stationary shuttlecocks on landing statistics can be reduced in training environments with multiple shuttlecocks and frequent personnel movement.

[0092] Furthermore, to ensure smooth video preview and statistical processing during long-term training, video acquisition, sphere inference and landing point statistics, and screen display are set in decoupled execution paths. The display path only acquires the latest images published by the inference path for display and result overlay, while user operation commands are fed back to the inference path processing through an independent channel, thus avoiding interface refresh blocking sphere detection. When the average frame rate is consistently below approximately 28 frames per second during operation, the normal detection step size can be appropriately increased to reduce the computational load; when the system load decreases, the detection step size can be reduced. If the frame rate still cannot meet the operational requirements after adjustment, the detection step size is restored to its original state, and the cumulative hit threshold, missed detection tolerance threshold, and observation waiting parameters are updated simultaneously to ensure that the ground contact threshold parameters always match the current detection step size. This approach balances the accuracy, anti-interference capability, and long-term operational stability of landing point statistics in the second half of the training scenario.

[0093] Please see Figure 6 This application also provides a badminton shuttlecock landing point determination and evidence collection system based on visual trajectory, the system comprising: The data acquisition module is used to acquire sparsely sampled visual trajectory data corresponding to badminton videos; The spatial mapping module is used to establish a mapping relationship between image space and regular site space based on site calibration parameters and sparse sampling trajectory data. The ground contact detection module is used to detect ground contact candidates in sparse sampled trajectory data and determine the ground contact time by combining the detection step size adjustment strategy and the ground contact state determination mechanism. The contact point recognition module is used to identify the contact point based on the trajectory motion characteristics corresponding to the moment of contact, and obtain the target contact point. The landing point determination module is used to perform coordinate transformation based on the target contact point and mapping relationship, and to perform regular area matching to obtain the badminton shuttlecock landing point result; The evidence output module is used to generate rule judgment information based on the badminton shuttlecock landing point result, construct corresponding evidence playback data, and output the landing point judgment result.

[0094] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0095] 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 above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0096] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0097] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0098] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0099] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0100] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0101] 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.

[0102] This application provides a method and system for determining and providing evidence of badminton landing points based on visual trajectories. This method acquires sparsely sampled visual trajectory data corresponding to badminton videos and establishes a mapping relationship between image space and regular court space by combining court location parameters. This establishes a connection between the video space trajectory and the actual match court coordinates, improving the accuracy of badminton trajectory analysis and landing point positioning. By performing ground contact candidate detection on the visual trajectory and combining a detection step size adjustment strategy and a ground contact state determination mechanism to determine the ground contact time, it can reduce the deviation in ground contact time caused by factors such as sampling intervals and changes in movement speed under complex motion conditions, improving the stability of landing point detection. Furthermore, based on the trajectory motion characteristics corresponding to the ground contact time, it identifies the ground contact point and determines the badminton landing point result through coordinate transformation and regular region matching, achieving automated determination of the landing point area, reducing subjective errors caused by manual observation, and improving determination efficiency and consistency. Simultaneously, it generates rule determination information based on the landing point result and constructs evidence playback data, realizing traceability and visual evidence presentation of the landing point determination process, facilitating referee review, match analysis, and dispute resolution, thereby improving the accuracy, real-time performance, and reliability of the intelligent badminton match determination system.

[0103] 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.

[0104] 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.

[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] 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 method for determining and proving the landing point of a badminton shuttlecock based on visual trajectory, characterized in that, The method includes the following steps: Acquire sparsely sampled visual trajectory data corresponding to badminton videos; the sparsely sampled visual trajectory data includes video frame identifier data, ball center position coordinate data, detection time series data, and trajectory confidence data; Based on the field calibration parameters and the sparse sampling trajectory data, a mapping relationship between the image space and the regular field space is established; The sparse sampled trajectory data is subjected to ground contact candidate detection, and the ground contact time is determined by combining the detection step size adjustment strategy and the ground contact state determination mechanism. The contact point is identified based on the trajectory motion characteristics corresponding to the contact moment, and the target contact point is obtained. Based on the target contact point and the mapping relationship, coordinate transformation is performed, and regular area matching is conducted to obtain the badminton shuttlecock landing point result; Based on the badminton shuttlecock landing point results, rule judgment information is generated, corresponding evidence playback data is constructed, and landing point judgment results are output.

2. The method according to claim 1, characterized in that, The process of establishing a mapping relationship between image space and regular site space based on site calibration parameters and sparse sampling trajectory data includes: Based on the badminton court corner point calibration data, the homography transformation parameters between the image space and the regular court space are determined; Based on the homography transformation parameters and the field corner point calibration results, the main playing area and the outer contact area corresponding to the regular field space are determined; the outer contact area is used for contact observation access control, and the main playing area is used for in-bounds, out-of-bounds and line-pressing judgments. Based on the homography transformation parameters, coordinate mapping is performed on the sparsely sampled visual trajectory data to obtain trajectory coordinate data in a regular field space. Based on the main competition area, the extended contact area, and the trajectory coordinate data, a mapping relationship is established between the image space and the rule field space.

3. The method according to claim 2, characterized in that, The step of performing ground contact candidate detection on the sparse sampled trajectory data, combined with a detection step size adjustment strategy and a ground contact state determination mechanism, to determine the ground contact time includes: Based on the sparse sampled trajectory data, the badminton trajectory change features are extracted to generate a ground contact candidate detection sequence; Based on the difference between the current detection step size and the preset benchmark step size, the cumulative hit threshold, missed detection tolerance threshold and observation waiting parameters corresponding to the ground contact detection are adjusted to generate step size-adapted detection parameters. Based on the step size adaptation detection parameters, the state of the candidate ground contact detection sequence is determined to determine the ground contact observation state; wherein, before entering the ground contact observation state, when the center of the badminton shuttlecock is detected to be located in the outer ground contact area, the number of valid hits is accumulated, and the accumulated number of valid hits is maintained when a missed detection occurs; when the number of valid hits meets the accumulated hit threshold, it is determined to enter the ground contact observation state; Based on the motion trajectory segment corresponding to the ground-touching observation state, ground-touching candidate moments are detected, and the ground-touching moment is determined according to the motion characteristics corresponding to the candidate moments.

4. The method according to claim 3, characterized in that, The step of detecting candidate ground contact times based on the motion trajectory segments corresponding to the ground contact observation state, and determining the ground contact time based on the motion characteristics corresponding to the candidate times, includes: Based on the motion trajectory segment, extract the vertical coordinate extreme points, local velocity change points and trajectory convergence points to generate a set of candidate ground contact moments; Based on the set of ground contact candidate moments, the corresponding pre-contact deceleration features, post-contact rebound features, trajectory convergence features, and displacement stability features are extracted to obtain candidate evaluation features. Based on the candidate evaluation features, the set of ground contact candidate times is scored and sorted to determine the target ground contact candidate time; Based on the target touchdown candidate time combined with the touchdown state boundary conditions, when the shuttlecock center is detected to leave the outward touchdown area, a valid post-touch sample is obtained, or the number of consecutive missed detections exceeds the missed detection tolerance threshold, the touchdown observation window is closed, and the target touchdown candidate time with the highest current score is determined as the touchdown time.

5. The method according to claim 4, characterized in that, The acquisition of the motion trajectory segment includes: Before the shuttlecock's center is first detected to enter the outward contact area, trajectory sampling points with a downward trend are continuously extracted from the sparsely sampled visual trajectory data to obtain the pre-contact falling tail segment data. In response to the first detection that the center of the badminton shuttlecock enters the outward contact area, the data of the pre-contact drop segment is frozen, and the trajectory segment within the area after the center of the badminton shuttlecock enters the outward contact area is acquired. The data of the pre-touch fall tail segment is merged with the trajectory segment within the region according to the detection time sequence to obtain the motion trajectory segment.

6. The method according to claim 1, characterized in that, The step of identifying the point of contact based on the trajectory motion features corresponding to the moment of contact, to obtain the target point of contact, includes: Based on the trajectory segment corresponding to the moment of contact with the ground, extract the change features of the badminton shuttlecock's center position to generate a candidate set of contact points; Based on the candidate set of contact points, extract the corresponding vertical extreme value features, velocity change features and trajectory convergence features to generate contact point evaluation features; Based on the evaluation features of the contact point, the candidate contact point set is matched and filtered to determine the candidate location of the target contact point; The target contact point is obtained by performing position correction based on the candidate positions of the target contact point and the image coordinate information corresponding to the contact time.

7. The method according to claim 2, characterized in that, The process of performing coordinate transformation based on the target contact point and the mapping relationship, and then performing regular region matching to obtain the badminton shuttlecock landing point result includes: Based on the image coordinates corresponding to the target contact point and the mapping relationship, coordinate mapping is performed to obtain the coordinates of the regular field. Based on the rules, the field coordinates are spatially matched with the main competition area to determine the area attributes corresponding to the target touch point. Boundary determination is performed based on the positional relationship between the rule-based field coordinates and the field edge lines to determine the edge line state corresponding to the target contact point; The badminton shuttlecock landing point is generated based on the region attributes and the edge line status.

8. The method according to claim 7, characterized in that, The boundary determination based on the positional relationship between the rule-based field coordinates and the field edge lines, to determine the edge line state corresponding to the target contact point, includes: Based on the rules and field coordinates, the corresponding edge detection objects are determined, and the set of outer frame edge lines corresponding to the main competition area is obtained. Calculate the distance between the coordinates of the regular site and each edge in the set of outer frame edges to obtain edge distance data; Based on the comparison between the edge distance data and the preset pressure threshold, the pressure state corresponding to the target contact point is determined; Based on the pressure line state, generate the edge line state information corresponding to the target contact point.

9. The method according to claim 1, characterized in that, The process of generating rule-based judgment information based on the badminton shuttlecock landing point result, constructing corresponding evidence replay data, and outputting the landing point judgment result includes: Based on the badminton shuttlecock landing point result, the corresponding rule judgment type is determined, and rule judgment information is generated; the rule judgment information includes the landing point area, the sideline status, and the judgment basis. Based on the video frame data and site mapping data corresponding to the ground contact time, ground contact point playback data and rule scene mapping data are generated respectively. Based on the connection and fusion of the touch point playback data and the rule scene mapping data, a dual-channel evidence playback data is constructed. The badminton shuttlecock landing point determination result is output by integrating the dual-channel evidence playback data with the rule determination information.

10. A badminton shuttlecock landing point determination and evidence collection system based on visual trajectory, characterized in that, The system is applied to the method as described in any one of claims 1-9, the system comprising: The data acquisition module is used to acquire sparsely sampled visual trajectory data corresponding to badminton videos; The spatial mapping module is used to establish a mapping relationship between the image space and the regular site space based on the site calibration parameters and the sparse sampling trajectory data. The ground contact detection module is used to perform ground contact candidate detection on the sparse sampled trajectory data, and determine the ground contact time by combining the detection step size adjustment strategy and the ground contact state determination mechanism. The touch point recognition module is used to identify the touch point based on the trajectory motion characteristics corresponding to the touch moment, and obtain the target touch point; The landing point determination module is used to perform coordinate transformation based on the target contact point and the mapping relationship, and to perform regular area matching to obtain the badminton shuttlecock landing point result; The evidence output module is used to generate rule judgment information based on the badminton shuttlecock landing point result, construct corresponding evidence playback data, and output the landing point judgment result.

Images