A motion data acquisition system, a swimming trajectory analysis method and a storage medium

Abstract

The application relates to the technical field of swimming technology analysis, in particular to a motion data acquisition system, a swimming track analysis method and a storage medium. The system comprises a multi-channel image acquisition unit, a plurality of image acquisition devices and a synchronous triggering device, the synchronous triggering device is circuit-connected with each image acquisition device, is used for controlling each image acquisition device to synchronously acquire image data of a target human body in a motion process according to a synchronous acquisition instruction; a control unit, a control machine, the control machine is circuit-connected with the synchronous triggering device, is used for sending the synchronous acquisition instruction and receiving original image data transmitted by each image acquisition device; a processing unit, a processor, the processor is circuit-connected with the control machine, is used for uniformly processing spatial coordinates of the original image data, extracting human body key points and calculating spatial motion tracks. The application eliminates the time synchronization error of multi-channel acquisition, improves the data precision and the track reconstruction quality, and meets the high-precision requirement of swimming and other harsh scenes.

Description

Technical Field

[0001] This invention relates to the field of swimming technique analysis technology, and in particular to a motion data acquisition system, a swimming trajectory analysis method, and a storage medium. Background Technology

[0002] Sports data acquisition systems are core infrastructure equipment in the fields of scientific training and biomechanical analysis in competitive sports. They acquire kinematic parameters such as position and posture during human movement through non-contact visual acquisition and other technologies. Swimming data acquisition, as a special branch, needs to cover the entire process both on and under water. It has much stricter requirements for the temporal synchronization accuracy and spatial coordinate consistency of multi-channel data than land sports, which directly determines the reliability of subsequent trajectory analysis.

[0003] Existing multi-channel swimming motion data acquisition systems generally adopt an architecture where multiple independent image acquisition devices are directly connected to a host computer. The host computer sends acquisition commands to each device individually via network or serial port. After acquisition, offline data processing is performed. The core drawback is the lack of an independent synchronization triggering mechanism, which makes it impossible to achieve accurate synchronous acquisition of multi-channel images. The inherent delay and time deviation in the transmission of host computer commands lead to mismatches in motion images acquired from different channels at the same moment. This results in large errors in spatial coordinate unification and jumps in motion trajectory stitching. At the same time, the separation of data acquisition and processing makes it difficult to support the high-precision, high-real-time swimming motion trajectory analysis requirements.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the general background of this disclosure and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] In view of at least one of the above technical problems, the present invention provides a motion data acquisition system, a swimming trajectory analysis method and a storage medium, which adopts an integrated independent synchronous triggering multi-channel image acquisition and integrated processing architecture to eliminate multi-channel acquisition time synchronization error and improve the accuracy of motion data acquisition and trajectory reconstruction quality.

[0006] This invention provides a motion data acquisition system, comprising: A multi-channel image acquisition unit includes multiple image acquisition devices and a synchronization triggering device. The synchronization triggering device is connected to the circuit of each of the image acquisition devices and is used to control each of the image acquisition devices to synchronously acquire image data of the target human body during the movement process according to the synchronization acquisition command. The control unit includes a controller, which is circuitically connected to the synchronization triggering device and is used to send the synchronization acquisition command and receive the raw image data transmitted by each of the image acquisition devices. The processing unit includes a processor connected to the control circuit, which is used to perform spatial coordinate unification, human body key point extraction, and spatial motion trajectory calculation on the original image data.

[0007] Furthermore, the processor has the following built-in features: The data fusion module is used to perform visual calibration on the acquisition channels corresponding to each of the image acquisition devices, establish spatial coordinate relationships between the acquisition channels, and obtain multi-channel fused data. The data extraction module is used to identify key points of the target human body and extract the coordinate data of the key points; The trajectory calculation module is used to calculate the spatial motion trajectory of the target human body based on the coordinate data of the key points and the multi-channel fused data.

[0008] This invention also provides a method for analyzing swimming trajectories, comprising: Simultaneously acquire multi-channel image data of the target human body during its movement; Visual calibration is performed on each acquisition channel of the multi-channel image data to establish spatial coordinate relationships between each acquisition channel, thereby obtaining multi-channel fused data. Identify key points of the target human body and extract the coordinate data of the key points; The spatial motion trajectory of the target human body is calculated based on the coordinate data of the key points and the multi-channel fused data.

[0009] Furthermore, when performing the visual calibration on each of the acquisition channels, a reference point with known actual spatial coordinates is selected. Based on the reference object with a known spatial position, the mapping relationship between pixel coordinates and actual spatial coordinates is calibrated, and the mapping relationship between pixel coordinates and actual spatial coordinates is established.

[0010] Furthermore, visual calibration is performed on each of the acquisition channels to obtain multi-channel fused data, including: Based on Snell's law of refraction, a water-air bilayer medium refraction propagation model was constructed to adapt to underwater shooting scenarios in swimming pools. The pixel coordinates of the underwater acquisition channel are pre-corrected using the refraction propagation model. Based on the corrected pixel coordinates, a spatial coordinate association is established between each acquisition channel to achieve multi-channel image data fusion and obtain the multi-channel fused data.

[0011] Furthermore, after performing refraction pre-correction on the pixel coordinates of the underwater acquisition channel, water surface correction is also included: The boundary between the lane line and the water surface is identified in real time through the acquisition channel located on the water, the actual water surface height of the current frame is calculated and the vertical reference plane is updated. Substitute the actual water surface height into the refraction propagation model, dynamically adjust the calculation logic of water layer thickness and light incident angle, and update the refraction correction parameters. Frame-by-frame dynamic pre-correction is performed on the pixel coordinates of human key points in each frame of the underwater acquisition channel.

[0012] Furthermore, identifying key points of the target human body includes: Collect sample data containing multiple swimming strokes, and manually annotate the sample data to form an annotated sample dataset; The labeled sample dataset is divided into a training set and a validation set according to a preset ratio; The human keypoint detection model is trained based on the training set, and the parameters can be verified and optimized using the validation set. The trained human key point detection model is used to detect key points on target human bodies in the various swimming strokes.

[0013] Furthermore, after extracting the coordinate data of the key points and before calculating the spatial motion trajectory of the target human body, the process also includes: Based on predefined normal range of human skeleton length, skeleton proportion constraints, joint motion angle limits and swimming motion timing constraints, abnormal coordinate data that does not conform to human anatomy and kinematic laws are eliminated and corrected. The detection confidence scores of each key point output by the human key point detection algorithm are obtained, and combined with the occlusion ratio of each acquisition channel, multiple sets of coordinate data of the same key point are dynamically weighted and fused in the overlapping observation area of ​​adjacent acquisition channels to obtain a unique valid coordinate.

[0014] Furthermore, calculating the spatial motion trajectory of the target human body includes: The coordinate data of each key point are converted into actual spatial coordinates in the global coordinate system; The actual spatial coordinates of each acquisition channel are spliced ​​together according to a unified time axis, and the discontinuous segments are filled by cubic spline interpolation to obtain the spliced ​​motion trajectory. The spliced ​​motion trajectory is smoothed by Kalman filtering to obtain the spatial motion trajectory that covers the entire swimming process of the target human body. Based on the described spatial motion trajectory, swimming-specific technical indicators are quantitatively calculated.

[0015] The present invention also provides a storage medium, wherein the computer instructions are used to cause the computer to execute the swimming trajectory analysis method described above.

[0016] The technical solution of this invention can achieve the following technical effects: The overall architecture of multi-channel image acquisition, centralized control and integrated processing with integrated independent synchronous triggering mechanism solves the core problems of large time synchronization error and poor data matching caused by independent triggering of multi-channel acquisition devices. The acquisition sequence of all image acquisition devices is directly and uniformly controlled by the synchronous triggering device, eliminating the transmission delay and time deviation caused by the host computer sending instructions separately, and ensuring that the motion images acquired by different channels are aligned in the time dimension.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a logical schematic diagram of the motion data acquisition system in an embodiment of the present invention; Figure 2 This is a schematic diagram of the processor in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the swimming trajectory analysis method in an embodiment of the present invention; Figure 4 This is a schematic diagram of the process for obtaining multi-channel fused data in an embodiment of the present invention; Figure 5 This is a schematic diagram of the water surface correction process in an embodiment of the present invention; Figure 6 This is a flowchart illustrating the process of calculating the spatial motion trajectory of the target human body in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] 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 invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] This invention provides a method such as Figure 1 and Figure 2 The motion data acquisition system shown includes: The multi-channel image acquisition unit includes multiple image acquisition devices and a synchronization triggering device. The synchronization triggering device is connected to the circuit of each image acquisition device and is used to control each image acquisition device to synchronously acquire image data of the target human body during the movement process according to the synchronization acquisition command. The control unit includes a controller circuit connected to a synchronous triggering device, used to send synchronous acquisition commands and receive raw image data transmitted by each image acquisition device; The processing unit includes a processor, which is connected to the control circuit and is used to perform spatial coordinate unification, human body key point extraction, and spatial motion trajectory calculation on the original image data.

[0023] The working principle of this invention is as follows: This system is designed for data acquisition in a 50-meter standard swimming lane at a professional swimming training center. The data acquisition system comprises a multi-channel image acquisition unit, a control unit, and a processing unit. The multi-channel image acquisition unit can include multiple global shutter industrial cameras as image acquisition devices and an FPGA synchronous trigger board as a synchronous triggering device. The synchronous triggering device is connected to the external hardware trigger interface circuit of each industrial camera via shielded coaxial cables. The control unit can use a high-performance industrial computer running Windows as the control machine, connected to the synchronous triggering device circuit via a PCIe high-speed interface. The processing unit uses a multi-core CPU processor integrated within the same industrial computer as the core processing component, connected to the control machine via a high-speed bus circuit inside the motherboard. During actual operation, the control... The machine first sends a unified synchronous acquisition command to the synchronous triggering device. After receiving the command, the synchronous triggering device simultaneously outputs the same level trigger signal to all connected industrial cameras, precisely controlling each industrial camera to start image acquisition at the same time, eliminating time synchronization errors between different channels, and ensuring that the multi-channel image data of the target human swimming process are aligned in the time dimension. The raw image data acquired by each industrial camera can be transmitted to the control computer in real time via gigabit Ethernet. The control computer transmits all the received channel raw image data synchronously to the processor through the motherboard's internal bus. The processor sequentially performs spatial coordinate unification processing, target human key point extraction processing, and spatial motion trajectory calculation processing on the multi-channel raw image data, and finally outputs continuous spatial motion trajectory data of the target human covering the entire swimming process.

[0024] In some embodiments of the present invention, such as Figure 2 As shown, the processor has the following built-in features: The data fusion module is used to perform visual calibration on the acquisition channels of each image acquisition device, establish spatial coordinate relationships between each acquisition channel, and obtain multi-channel fused data. The data extraction module is used to identify key points of the target human body and extract the coordinate data of the key points; The trajectory calculation module is used to calculate the spatial motion trajectory of the target human body based on the coordinate data of key points and multi-channel fused data.

[0025] The processor adopts a modular software architecture with built-in data fusion, data extraction, and trajectory calculation modules. These modules achieve high-speed data interaction through memory sharing, avoiding latency caused by cross-process data transmission and improving overall system processing efficiency. During system initialization, the data fusion module performs visual calibration on the acquisition channels of each industrial camera. By establishing spatial coordinate relationships between these channels, it transforms the originally independent and scattered multi-channel image data into multi-channel fused data based on the same global coordinate system. This fundamentally solves the problem of inconsistent spatial references for image data from different perspectives, providing a unified and reliable spatial foundation for subsequent high-precision trajectory calculation. The data extraction module uses a pre-trained human keypoint recognition model to batch process multi-channel image data, automatically identifying key points such as the head, torso, and limbs of the target human body and accurately extracting the coordinate data corresponding to each keypoint. No manual annotation is required throughout the process, significantly improving the automation and speed of data processing. The trajectory calculation module combines the keypoint coordinate data output by the data extraction module with the multi-channel fused data output by the data fusion module, using spatial mapping and multi-segment trajectory stitching algorithms to calculate the three-dimensional spatial motion trajectory of the target human body.

[0026] Based on the same inventive concept as the motion data acquisition system in the foregoing embodiments, this invention also provides a swimming trajectory analysis method, such as... Figures 3 to 6 As shown, it includes: Simultaneously acquire multi-channel image data of the target human body during its movement; Visual calibration is performed on each acquisition channel of the multi-channel image data to establish spatial coordinate relationships between the acquisition channels and obtain multi-channel fused data. Identify key points of the target human body and extract the coordinate data of the key points; Based on the coordinate data of key points and multi-channel fused data, the spatial motion trajectory of the target human body is calculated.

[0027] Implemented in the swimming lane using the aforementioned motion data acquisition system, the entire process requires no wearable sensors on the target swimmer, avoiding interference from wearable devices and ensuring the authenticity and objectivity of the motion data. First, multi-channel image data of the target swimmer is simultaneously acquired via a multi-channel image acquisition unit. This simultaneous acquisition is controlled by an independent synchronization triggering device, ensuring that the motion images acquired from different channels at the same moment are perfectly aligned in time, reducing time synchronization errors and guaranteeing the basic accuracy of subsequent trajectory calculations from the data source. After acquisition, visual calibration is performed on each acquisition channel to establish the accuracy of each acquisition channel. The spatial coordinate association between channels unifies image data originally scattered in local coordinate systems at different perspectives into the same global coordinate system, resulting in multi-channel fused data. This lays a unified and reliable spatial foundation for subsequent cross-channel trajectory stitching. Then, the system automatically identifies key points such as the head, torso, and limbs of the target human body and extracts the coordinate data corresponding to each key point. No manual annotation intervention is required throughout the process, which greatly improves the efficiency of data processing and the consistency of results. Finally, the extracted key point coordinate data is combined with the multi-channel fused data to calculate the continuous spatial motion trajectory of the target human body covering the entire process of starting from the water, taking off in the air, swimming underwater, turning and touching the wall.

[0028] In some embodiments of the present invention, when visually calibrating each acquisition channel, a reference point with known actual spatial coordinates is selected. Based on the reference object with a known spatial position, the mapping relationship between pixel coordinates and actual spatial coordinates is calibrated, and the mapping relationship between pixel coordinates and actual spatial coordinates is established.

[0029] Visual calibration is performed after system deployment and before formal motion data collection. This invention includes horizontal and vertical calibration. In horizontal calibration, the operator selects the start and end positions of the lane on the display interface of each camera, records their corresponding pixel coordinates, and inputs the actual physical coordinates of these two positions in the pool's global coordinate system. Based on this correspondence, a mapping model from pixel coordinates to actual physical coordinates is established, thereby realizing the coordinate conversion of any pixel point within the camera's field of view, providing a spatial scale reference for subsequent calculation of the actual motion trajectory. In vertical calibration, the operator calibrates the longitudinal pixel coordinates of the water surface position on the camera display interface, and further calibrates the longitudinal pixel coordinates of any reference point, while simultaneously inputting the actual physical coordinates of the corresponding positions. Based on this correspondence, a mapping model from longitudinal pixel coordinates to actual physical coordinates is established, realizing the scale conversion and unification of the longitudinal coordinates of key human body points.

[0030] During calibration, natural, fixed standard reference points at the swimming pool site are prioritized as reference points with known actual spatial coordinates. Specifically, standard scale points of lane lines, corner vertices of starting blocks, cross intersections of pool bottom tiles, and endpoints of water level scale lines on the pool sidewalls can be selected. During actual calibration, no fewer than eight sets of non-collinear reference points are selected for each acquisition channel. The actual physical coordinates of each reference point in the pool's global coordinate system are recorded one by one. Simultaneously, the pixel coordinates of each reference point are accurately located in the acquired images of the corresponding channel. Then, the actual spatial coordinates of the same reference point are correlated and matched with the corresponding image pixel coordinates to establish a one-to-one mapping relationship between pixel coordinates and actual spatial coordinates. Using natural, fixed reference points in the swimming pool for calibration ensures that the calibration benchmark is consistent with the actual motion scene. The established pixel-actual spatial coordinate mapping relationship has higher accuracy and stronger stability, and can accurately convert the pixel positions in the image into actual physical spatial positions.

[0031] In some embodiments of the present invention, such as Figure 4 As shown, visual calibration is performed on each acquisition channel to obtain multi-channel fused data, including: Based on Snell's law of refraction, a water-air bilayer medium refraction propagation model was constructed to adapt to underwater shooting scenarios in swimming pools. The pixel coordinates of the underwater acquisition channel are pre-corrected using a refraction propagation model. Based on the corrected pixel coordinates, spatial coordinate relationships are established between each acquisition channel to achieve multi-channel image data fusion and obtain multi-channel fused data.

[0032] First, based on Snell's law of refraction, a water-air dual-medium refraction propagation model adapted to underwater pool shooting scenarios is constructed. The standard refractive indices of both pool water and air are defined, along with the internal parameters and distortion coefficients of the camera lens. For light rays emitted from real underwater objects, the single refraction path at the water-air interface is derived, establishing a forward mapping relationship between underwater real-world coordinates and imaging pixel coordinates. Through iterative optimization, a correction relationship is obtained by inversely solving for real-world coordinates from pixel coordinates. Then, the refraction propagation model is used to perform refraction pre-correction on the pixel coordinates of all reference points in the underwater acquisition channel, eliminating positional shifts caused by light deflection at the water-air interface, resulting in equivalent pixel coordinates close to an ideal, refraction-free state. Finally, a global metric rectangular coordinate system for the pool is locked, with the front edge of the starting platform above water as the horizontal zero point and the still water surface as the vertical zero point. The system selects natural fixed standard reference points in the swimming pool as calibration reference points, and selects no less than eight sets of non-collinear reference points for each acquisition channel. The global true coordinates and corresponding image pixel coordinates are recorded simultaneously. The above-water acquisition channel directly uses the original pixel coordinates, while the underwater acquisition channel can use the equivalent pixel coordinates after refraction pre-correction. Then, the mapping relationship between pixel coordinates and actual spatial coordinates is established through homography matrix. The swimming lane is divided into multiple intervals at fixed intervals to solve parameters independently, and piecewise linear interpolation is used to improve the conversion accuracy of large field of view. Finally, the spatial transformation matrix of each acquisition channel relative to the global coordinate system is solved to complete the unification of multi-channel spatial reference and realize the fusion of multi-channel image data, so as to obtain stable and reliable multi-channel fused data, eliminate the inherent refraction deviation of underwater shooting, improve the accuracy of coordinate mapping, and make the positioning of underwater key points closer to the real physical location.

[0033] In some embodiments of the present invention, such as Figure 5 As shown, after refraction pre-correction of the pixel coordinates of the underwater acquisition channel, water surface correction is also included: The boundary between the lane line and the water surface is identified in real time through the acquisition channel located on the water, the actual water surface height of the current frame is calculated and the vertical reference plane is updated. Substitute the actual water surface height into the refraction propagation model, dynamically adjust the calculation logic of water layer thickness and light incident angle, and update the refraction correction parameters. Frame-by-frame dynamic pre-correction is performed on the pixel coordinates of human key points in each frame of the underwater acquisition channel.

[0034] A dynamic monitoring channel was selected by placing an underwater acquisition channel directly above the swimming lanes, with a complete field of view covering the lane lines and the water surface area. This channel acquires real-time global images of the swimming lanes, identifies the boundary between the lane lines and the water surface, and calculates the actual water surface height in the current frame image by combining the known standard spacing of the lane lines and global coordinate system parameters. This height is then compared to the initially calibrated still water surface height, and the system's vertical reference plane is updated in real-time to ensure consistency between the water surface height reference and the actual swimming scenario. Subsequently, the calculated actual water surface height is input into the previously constructed water-air dual-layer medium refraction propagation model, dynamically adjusting the water layer thickness parameters in the model. Simultaneously, the incident angle of light passing through the water-air interface is recalculated based on changes in water surface height, and the refraction correction parameters are updated synchronously to ensure the refraction propagation model always adapts to the current actual water surface state. Finally, based on the updated refraction correction parameters, dynamic pre-correction processing is performed frame-by-frame on the pixel coordinates of key human figures in each frame of images acquired by the underwater acquisition channel. This ensures that the pixel coordinates of key human figures in each frame eliminate refraction deviations caused by current water surface fluctuations, improving the coordinate positioning accuracy of the underwater acquisition channel and the system's environmental adaptability.

[0035] In some embodiments of the present invention, when identifying key points of the target human body, a human body key point detection algorithm is used to construct training sets for each swimming stroke of the target human body.

[0036] When identifying key points of a target human body, a portion of the collected multi-channel image data is manually annotated (e.g., approximately 1% of all samples) to form an annotated sample dataset. This dataset is then manually divided into a training set and a validation set (e.g., 9:1) according to a preset ratio. After dataset division, the human key point detection model is trained on the training set, and the validation set is used for performance evaluation and parameter tuning. Finally, the trained human key point detection model is used to detect key points of target human bodies in various swimming styles, including freestyle, breaststroke, backstroke, and butterfly. The training set is constructed around real-world swimming scenarios, covering complete motion sequences for different swimming strokes, including key movements such as start, stroke, turn, and touchdown. Each swimming stroke's training set contains a sufficient number of samples, covering key human features from different angles, under different lighting conditions, and in different water surface states, ensuring the diversity and representativeness of the training data. During training, the features of key points such as the head, torso, and limbs are extracted in a focused manner, taking into account the characteristics of swimming movements, to optimize the recognition accuracy of the algorithm model. The feature weights of the model are adjusted according to the differences in movements of different swimming strokes, enabling the model to accurately identify key points for each swimming stroke, avoiding recognition bias caused by differences in swimming strokes, improving the recognition accuracy of key human points for multiple swimming strokes, and ensuring the accuracy and reliability of motion trajectory data for different swimming strokes.

[0037] In some embodiments of the present invention, after extracting the coordinate data of key points and before calculating the spatial motion trajectory of the target human body, the method further includes: Based on predefined normal range of human skeleton length, skeleton proportion constraints, joint motion angle limits and swimming action timing constraints, abnormal coordinate data that do not conform to human anatomy and kinematic laws are eliminated and corrected. The detection confidence scores of each key point output by the human key point detection algorithm are obtained. Combined with the occlusion ratio of each acquisition channel, multiple sets of coordinate data of the same key point are dynamically weighted and fused in the overlapping observation area of ​​adjacent acquisition channels to obtain a unique valid coordinate.

[0038] First, a multi-level constraint rule base is constructed based on human anatomy and swimming kinematics. Specifically, the human body can be simplified into n rigid segments, and n core key points are defined accordingly, including n key points above water such as the head, trunk center, left and right hips, left and right knees, left and right ankles, left and right feet, and left and right toes, as well as n key points underwater such as the head and trunk center. For fine-grained analysis of swimming strokes, key points such as left and right shoulders, left and right elbows, and left and right wrists are also added. Based on this, the normal range of skeletal lengths between adjacent key points and their proportional constraints are predefined; for example, the ratio of thigh length to forearm length should be within the normal physiological range. Simultaneously, the motion angle limits of each joint are predefined according to the characteristics of swimming movements, and temporal constraints for swimming movements are also constructed. For example, during the start and takeoff phase, the trajectory of the trunk center follows a parabolic shape, and during the swimming phase, the horizontal coordinates of each key point show a unidirectional increasing trend with no abnormal rebound.

[0039] After obtaining the coordinate data of each key point in each frame output by the human key point detection algorithm, the system performs the following checks in sequence: First, it calculates the Euclidean distance between each pair of adjacent key points. If the distance exceeds the normal range of bone length, it determines that the current key point coordinates are abnormal and performs linear interpolation correction on the current key point coordinates according to the bone ratio constraint. For example, when the distance from the center of the torso to the hip is abnormal, it uses the normal ratio relationship between the center of the torso and the hip to reversely calculate the reasonable position of the hip. It can also calculate the actual angle of each joint. If it exceeds the predefined joint motion angle limit, it performs smooth correction according to the angle change trend of adjacent frames.

[0040] Finally, the key point coordinate sequence is checked to see if it meets the timing constraints of swimming movements. For example, if the horizontal coordinate of the torso center in a certain frame is significantly smaller than that in the previous frame during the swimming phase, the point is identified as a timing anomaly. The coordinates of the previous and next frames are then used to perform cubic spline interpolation to effectively eliminate isolated noise points caused by underwater light refraction, water surface reflection, rapid limb swings, etc., making the trajectory data more consistent with the laws of real human movement.

[0041] After single-channel anomaly removal and correction, the system further performs multi-channel dynamic weighted fusion to obtain the detection confidence score of each keypoint output by the human keypoint detection algorithm. Simultaneously, for each acquisition channel, the system calculates the occlusion percentage of the keypoint's area in the current frame in real time. For example, when an athlete's limbs are obscured by lane lines, splashes, or their torso, the system uses image segmentation to calculate the proportion of the obscured pixel area to the total area of ​​the keypoint's detection box, and subtracts the occlusion percentage from 1 as the visibility factor. In overlapping observation areas of adjacent acquisition channels, such as areas where two adjacent cameras can clearly capture the athlete's hip joint, the same keypoint will simultaneously obtain multiple sets of coordinate data from different acquisition channels. The system calculates a fusion weight for each set of coordinate data: the weight is equal to the product of the detection confidence and the visibility factor, and then multiplied by the average detection stability coefficient of the key point in the historical frames of the current acquisition channel. Then, the coordinate data of all channels are weighted and averaged to obtain the unique valid coordinates of the key point in the current frame. This fully utilizes the redundant observation information from multiple perspectives, significantly improves the continuity and anti-interference ability of the key point coordinates, and provides a unique, accurate and conflict-free coordinate input for subsequent trajectory stitching.

[0042] In some embodiments of the present invention, such as Figure 6 As shown, calculating the spatial motion trajectory of the target human body includes: Convert the coordinate data of each key point into actual spatial coordinates in the global coordinate system; The actual spatial coordinates of each acquisition channel are spliced ​​together according to a unified time axis, and cubic spline interpolation is used to fill in the discontinuous segments to obtain the spliced ​​motion trajectory. The spliced ​​motion trajectory is smoothed by Kalman filtering to obtain a spatial motion trajectory that covers the entire swimming process of the target human body. Based on the spatial motion trajectory, swimming-specific technical indicators are quantitatively calculated.

[0043] First, the pixel coordinates of each key point in each frame of the image are converted into actual spatial coordinates in the global coordinate system. This can be achieved through the mapping relationship of visual calibration. The homography matrix is ​​used to map the pixel coordinates to the pool's global metric rectangular coordinate system with the starting platform's front edge as the horizontal zero point and the still water surface as the vertical zero point. For the underwater acquisition channel, the light refraction error caused by water surface ripples is first eliminated through a frame-by-frame dynamic refraction correction model. Then, the same mapping relationship is applied to convert it into global actual spatial coordinates. The actual spatial coordinates of each acquisition channel are then stitched together according to a unified time axis. Each frame of the image in each channel has the same timestamp. During the overlapping time period of adjacent channels, the same key point may have actual spatial coordinates from two channels at the same time. In this case, the unique valid coordinate after the aforementioned dynamic weighted fusion is used. In non-overlapping areas, the coordinate data of a single channel is used. For the coordinate data discontinuities caused by temporary occlusion, detection failure, etc., the system uses cubic spline interpolation to fill in the gaps. After stitching and interpolation, an original trajectory sequence covering the entire process of the target human body from starting into the water, swimming, turning, to touching the wall at the finish line is obtained.

[0044] Building upon the above embodiments, a Kalman filter can be used to smooth the stitched motion trajectory, eliminating high-frequency disturbances introduced by detection noise, slight jitter, or interpolation errors. The state vector of the Kalman filter is defined as the three-dimensional position, velocity, and acceleration of each keypoint in the global coordinate system. The system model is established based on the assumption of uniform acceleration in swimming, and the observation vector is the actual spatial coordinates after stitching. Through recursive calculations in two steps—prediction and update—the filter can adaptively fuse historical motion trends with current observations, outputting the optimal estimated trajectory. For example, during the swimmer's stroke phase, the original coordinate sequence of the wrist joint may exhibit random jitter of several centimeters per frame due to water reflection. After Kalman filtering, the jitter is effectively suppressed while preserving the rapid start and stop characteristics of the movement, avoiding over-smoothing. Ultimately, the system obtains a high-precision, high-smoothness spatial motion trajectory covering the entire swimming process of the target human.

[0045] Based on the above, the system automatically quantifies and calculates swimming-specific technical indicators according to spatial motion trajectories. For example, by analyzing the trajectory slope of the head and torso center during the start phase, it calculates the start reaction time and the distance to the entry point; by extracting the trajectory amplitude and frequency of the left and right wrist joints during the stroke cycle, it calculates stroke efficiency and symmetry; by tracking the motion arc of the hip and knee during the turn phase, it quantifies the turn-to-wall contact time and push-off speed; and by fitting the horizontal displacement curve of the torso center during the swimming phase, it calculates segmented swimming speed, acceleration fluctuations, and breathing timing offsets. All technical indicators can be output to the user interface in numerical and graphical form for coaches and athletes to conduct scientific training evaluations.

[0046] The present invention also provides a storage medium and computer instructions for causing a computer to execute a swimming trajectory analysis method.

[0047] Similarly, the storage medium described above in this invention can effectively enable a computer to execute the swimming trajectory analysis method, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0048] Although this application has been described in conjunction with specific features and embodiments, it is apparent that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and accompanying drawings are merely exemplary illustrations of the application as defined herein and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A motion data acquisition system, characterized in that, include: A multi-channel image acquisition unit includes multiple image acquisition devices and a synchronization triggering device. The synchronization triggering device is connected to the circuit of each of the image acquisition devices and is used to control each of the image acquisition devices to synchronously acquire image data of the target human body during the movement process according to the synchronization acquisition command. The control unit includes a controller, which is circuitically connected to the synchronization triggering device and is used to send the synchronization acquisition command and receive the raw image data transmitted by each of the image acquisition devices. The processing unit includes a processor, which is connected to the control circuit and is used to perform spatial coordinate unification, human body key point extraction and spatial motion trajectory calculation on the original image data. The processor has the following built-in features: The data fusion module is used to perform visual calibration on the acquisition channels corresponding to each of the image acquisition devices, establish spatial coordinate relationships between the acquisition channels, and obtain multi-channel fused data. The data extraction module is used to identify key points of the target human body and extract the coordinate data of the key points; The trajectory calculation module is used to calculate the spatial motion trajectory of the target human body based on the coordinate data of the key points and the multi-channel fused data.

2. A method for analyzing swimming trajectories, characterized in that, include: Simultaneously acquire multi-channel image data of the target human body during its movement; Visual calibration is performed on each acquisition channel of the multi-channel image data to establish spatial coordinate relationships between each acquisition channel, thereby obtaining multi-channel fused data. Identify key points of the target human body and extract the coordinate data of the key points; The spatial motion trajectory of the target human body is calculated based on the coordinate data of the key points and the multi-channel fused data.

3. The swimming trajectory analysis method according to claim 2, characterized in that, When performing visual calibration on each of the acquisition channels, a reference point with known actual spatial coordinates is selected. Based on the reference object with a known spatial position, the mapping relationship between pixel coordinates and actual spatial coordinates is calibrated, and the mapping relationship between pixel coordinates and actual spatial coordinates is established.

4. The swimming trajectory analysis method according to claim 3, characterized in that, Visual calibration is performed on each of the acquisition channels to obtain multi-channel fused data, including: constructing a water-air dual-layer medium refraction propagation model adapted to underwater swimming pool shooting scenarios based on Snell's law of refraction; The pixel coordinates of the underwater acquisition channel are pre-corrected using the refraction propagation model. Based on the corrected pixel coordinates, a spatial coordinate association is established between each acquisition channel to achieve multi-channel image data fusion and obtain the multi-channel fused data.

5. The swimming trajectory analysis method according to claim 4, characterized in that, After performing refraction pre-correction on the pixel coordinates of the underwater acquisition channel, water surface correction is also included: The boundary between the lane line and the water surface is identified in real time through the acquisition channel located on the water, the actual water surface height of the current frame is calculated and the vertical reference plane is updated. Substitute the actual water surface height into the refraction propagation model, dynamically adjust the calculation logic of water layer thickness and light incident angle, and update the refraction correction parameters. Frame-by-frame dynamic pre-correction is performed on the pixel coordinates of human key points in each frame of the underwater acquisition channel.

6. The swimming trajectory analysis method according to claim 2, characterized in that, Key points for identifying the target human body include: Collect sample data containing multiple swimming strokes, and manually annotate the sample data to form an annotated sample dataset; The labeled sample dataset is divided into a training set and a validation set according to a preset ratio; The human keypoint detection model is trained based on the training set, and the parameters can be verified and optimized using the validation set. The trained human key point detection model is used to detect key points on target human bodies in the various swimming strokes.

7. The swimming trajectory analysis method according to claim 6, characterized in that, After extracting the coordinate data of the key points, and before calculating the spatial motion trajectory of the target human body, the process further includes: Based on predefined normal range of human skeleton length, skeleton proportion constraints, joint motion angle limits and swimming motion timing constraints, abnormal coordinate data that does not conform to human anatomy and kinematic laws are eliminated and corrected. The detection confidence scores of each key point output by the human key point detection algorithm are obtained, and combined with the occlusion ratio of each acquisition channel, multiple sets of coordinate data of the same key point are dynamically weighted and fused in the overlapping observation area of ​​adjacent acquisition channels to obtain a unique valid coordinate.

8. The swimming trajectory analysis method according to claim 2, characterized in that, Calculating the spatial motion trajectory of the target human body includes: The coordinate data of each key point are converted into actual spatial coordinates in the global coordinate system; The actual spatial coordinates of each acquisition channel are spliced ​​together according to a unified time axis, and the discontinuous segments are filled by cubic spline interpolation to obtain the spliced ​​motion trajectory. Based on the described spatial motion trajectory, swimming-specific technical indicators are quantitatively calculated.

9. A storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the swimming trajectory analysis method according to any one of claims 2-8.

Images