A multi-jump rope pattern recognition counting method

Abstract

The present application relates to a kind of multi-hop rope mode identification counting method, including being set with rope no rope skipping mode, pedal skipping mode, two feet skipping mode and two hands cross skipping mode, and / or which mode in forward / reverse skipping mode, maximum continuous jump mode and skipping displacement mode as current skipping mode, and identification counting is carried out;The photographed picture is carried out to remove duplicate processing, and the highest confidence once skipping information is retained, and model arrangement processing is carried out;Physical skipping count is combined with AI rope dance amplitude identification to obtain physical counting confidence, AI comprehensive counting confidence is obtained by AI identification combined with sensor joint counting, the size of the confidence of two confidence is compared, and the counting of high confidence is used as final counting result.The present application adopts multiple identification data joint correction, carries out data arrangement, after the confidence of different data is compared, it is decided to use which data, deduct which data, so that the skipping data of identification is more accurate.

Description

Technical Field

[0001] This invention relates to the field of AI intelligent image recognition technology, and in particular to a method for recognizing and counting multiple jump rope patterns. Background Technology

[0002] Current AI methods for recognizing and counting jump ropes primarily employ the following approaches: 1. Target detection of key human body points through video frame extraction, obtaining multiple key human body points, training a parametric waveform diagram of human jump rope using historical data of these key human body points, and then recognizing new key human body points in the video based on the parametric waveform diagram to obtain the jump rope count; 2. Obtaining human posture information through video frame extraction using feature extraction methods, extracting jump rope counting feature sequences, filtering the jump rope count, calculating the second-order difference information after extracting features from the detected 33 key human body points, obtaining the temporal feature patterns of the feature sequence, finding the peaks and troughs in the jump rope process based on the filtered feature sequence, and finally obtaining an accurate jump rope count by comparing the human posture features with the jump rope peaks and troughs. 3. Acquire the jump rope video. Detect 33 key points of the human body in the first frame of the jump rope video using a human key point detection model, and output the corresponding skeletal joint diagram. Set the coordinates of the moving joints in each skeletal joint diagram as the coordinates of the first key point. Obtain the coordinates of several second key points for each key limb in multiple second frames of the jump rope video. The acquisition of second key point coordinates includes: obtaining them by tracking the previous frame image using a tracking algorithm or by using the human key point detection model. Correlate all first and second key point coordinates with the height changes caused by each jump rope movement to obtain jump rope counting data. 4. Determine the state data related to the jump rope behavior using the jump rope handle. If the state data effectively represents the jump rope behavior, determine the jump rope counting data. The jump rope handle includes a processor module and a counting module. The jump rope handle includes a Hall sensor and an accelerometer (or gyroscope). The Hall sensor is responsible for collecting counting data, and the accelerometer is responsible for detecting the validity of the data. 5. When the rope acceleration is zero, the recorded rope skipping behavior is invalid; 6. By filtering the six-axis gyroscope data of the rope test loop based on the Hall sensor signal, the sine curves of the rope test loop in the X, Y, and Z directions are fitted by numerical calculation, and the error between the sine curve of the rope test loop and the six-axis gyroscope data of the loop is calculated; the error is used to determine whether the rope test loop is a bad loop, avoiding the phenomenon of incorrect counting caused by the rotation of the rope handle head to count the number of loops in the existing technology.

[0003] However, in reality, the forward and backward displacement during rope skipping causes the captured data to fall outside the scope of model training. Moreover, existing methods have limited recognition points, making it easy to miss recognitions if a few frames are not captured, and there is no other data to correct the missed erroneous data. Some methods use multi-point detection to collect rope skipping data, but in the same rope skipping, some positions can detect the rope skipping while others cannot, resulting in inconsistent data. Some AI recognition solutions only recognize the number of objects that move up and down in the image or only recognize the number of head movements, which can lead to the algorithm counting fake jumps even if the dummy jumps. If only Hall sensors or mechanical gears are used to record the rotation data of the rope, the handle will spin freely, making it impossible to identify bad loops or illegal rope skipping such as swinging the rope. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a multi-jump rope pattern recognition and counting method, which solves the deficiencies in existing AI-based jump rope recognition and counting methods.

[0005] The objective of this invention is achieved through the following technical solution: a multi-jump rope pattern recognition and counting method, wherein the recognition and calculation method includes:

[0006] The system identifies and counts which of the following jump rope modes—cordless jump rope, cycling jump rope, double-leg jump rope, and double-handed cross-over jump rope—and / or which of the following jump rope modes—forward / reverse jump rope, maximum consecutive jump, and jump rope displacement—is the current jump rope mode.

[0007] Multiple images captured by the front AI camera and / or the side AI camera are deduplicated, and the jump rope information with the highest confidence is retained and then processed for model arrangement.

[0008] The confidence level of the physical count is obtained by combining physical jump rope counting with AI rope swing amplitude recognition, and the confidence level of the AI ​​comprehensive count is obtained by combining AI recognition with sensor joint counting. The two confidence levels are compared, and the count with the higher confidence level is taken as the final recognition count result.

[0009] The front-facing AI camera identifies and counts the cycling and rope skipping patterns, including:

[0010] The AI ​​single-leg jump recognition model identifies whether each jump involves both feet landing simultaneously. If it is a single-leg jump, it counts as one jump. The jump time and confidence level are recorded after each valid single-leg jump recognition. If it is not a single-leg jump, it is a jump violation and the jump is not counted.

[0011] The AI-powered error jump detection model identifies incorrect jump rope postures. When both feet land simultaneously and when the rope is tripped, an error is recorded, the detection time is marked, and the confidence level of the data is retained.

[0012] The AI ​​rope-dancing vertex recognition model identifies the number of times the jump rope appears at the top. Each time the jump rope appears at the top, it is counted once.

[0013] The front-facing AI camera recognizes and counts jump rope patterns using a two-footed jump rope technique, including:

[0014] The AI-powered double-foot jump recognition model identifies whether the feet are together during each jump. If they are together, the jump is counted as one jump, and the jump time and confidence level are recorded. If not, the jump is not counted.

[0015] The AI-powered error jump detection model identifies incorrect jump rope postures. When the feet are not together or when the rope is tripped, an error is recorded, the detection time is marked, and the confidence level of the data is retained.

[0016] The AI ​​rope-dancing vertex recognition model identifies the number of times the jump rope appears at the top. Each time the jump rope appears at the top, it is counted once.

[0017] The side-mounted AI camera identifies and counts forward / backward jump rope patterns, including:

[0018] The AI ​​rope skipping sequence detection model sorts the rope skipping in time. When the rope appears from back to front, it satisfies the front skipping sequence and is a forward skipping mode. When the rope appears from front to back, it satisfies the reverse skipping sequence and is a reverse skipping mode.

[0019] When the forward jump rope mode is enabled, the AI ​​jump rope vertex recognition model identifies and counts the number of times the jump rope appears at the top. If the AI ​​jump rope sequence detection model detects that the time sequence of the jump rope in a certain dance meets the reverse dance sequence, then this count is invalid.

[0020] When the reverse jump rope mode is enabled, the AI ​​jump rope vertex recognition model directly identifies the number of times the jump rope appears at the top to complete the count.

[0021] The side-mounted AI camera identifies and counts the maximum consecutive jump patterns, including:

[0022] The AI ​​side foot jump recognition model is enabled to perform ground calibration of the jump position, find the relationship between the ground and the jump, and AI calibrates the two feet of the jump. The highest point of the jump is recorded once, and the confidence of skeleton recognition, jump recording time, jump height and jump accuracy for each jump is given.

[0023] The jump time is compared with the jump rope recognition time to obtain an accurate count of jumps.

[0024] The step of comparing the jump time with the jump rope recognition time to obtain an accurate jump count includes:

[0025] If two jump count points appear in a single jump rope count, it is determined whether the height of the multiple jumps is consistent. If the height is consistent and the time interval between the two jumps is consistent, it is determined that the jump rope technique point was missed. If the height is inconsistent, showing one high and one low, and the jump time is consistent, the jump rope count is determined to be a two-jump-one-dance rope pattern.

[0026] If there is only one jump count point in a single jump rope count, the jump rope count is judged as a normal one-jump-one-dance rope pattern;

[0027] If only one jump count occurs in multiple jump rope counts, further analysis is performed. If the intervals between counts before and after the time are irregular, it is judged as a jump rope tripping error. If the time interval between multiple records is twice the average time interval, it is judged as a missed jump count. If the time intervals between multiple jump rope counts are regular, it is judged as a maximum consecutive jump pattern.

[0028] The model orchestration process includes a time interval determination step and a data integration step. The time interval determination step includes:

[0029] If the recording time rhythm is stable, the confidence level increases by m with each count;

[0030] If there are interruptions in the recorded time rhythm, the time interval between the interruptions is recorded and marked as uncounted points. The time of the interruption is compared with the time of the incorrect jump rope posture identified by the AI ​​error recognition model. When the time of the incorrect jump rope posture is within the time range of uncounted points, the confidence of the incorrect jump rope posture recognition is increased by m.

[0031] If the time interval between two records is twice the average time interval, it is recorded as a suspected missed count point. The two time intervals are compared with the incorrect rope skipping postures identified by the AI ​​error detection model. If the incorrect rope skipping posture is within the suspected missed count time, the confidence of the incorrect rope skipping posture identification is increased by m. If no incorrect rope skipping posture is found within the suspected missed count time range, the missed rope skipping data is automatically supplemented. The confidence of the supplemented data is n by default.

[0032] If the incorrect posture identified by AI falls on the correct AI recording point, it will further determine whether the pattern identified by AI is consistent with the current jump rope pattern. If they are consistent, the correct AI record is valid; if they are inconsistent, the correct AI record is invalid.

[0033] The data integration step includes: arranging the data identified by the three models—the recognition model for each pattern, the AI ​​error jump recognition model, and the AI ​​rope dancing vertex recognition model—to obtain the final recognition data.

[0034] The present invention has the following advantages: a multi-jump rope pattern recognition and counting method, which uses multiple recognition data for joint correction, performs data arrangement, and determines which data to use and which data to deduct after comprehensively comparing the confidence levels of different data, thereby achieving more accurate recognition of jump rope data and effectively avoiding recognition errors caused by various abnormal problems during the recognition process; it is compatible with more fancy jump rope recognition and can judge more cases of illegal jump rope. Attached Figure Description

[0035] Figure 1 This is a schematic flowchart of the method of the present invention;

[0036] Figure 2 This is a diagram illustrating forward rope skipping.

[0037] Figure 3 This is a diagram illustrating reverse rope skipping.

[0038] Figure 4 This is a diagram illustrating continuous jump rope.

[0039] Figure 5 A diagram showing the comparison between jump rope counting points and jump counting points;

[0040] Figure 6 This is a diagram illustrating the time interval determination. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application provided below with reference to the accompanying drawings is not intended to limit the scope of protection of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. The present invention will be further described below with reference to the accompanying drawings.

[0042] like Figure 1 As shown, the present invention specifically relates to a multi-jump rope pattern recognition and counting method, which includes the following:

[0043] S1. The current jump rope mode can be manually set via a touchscreen with a camera, or via a mobile app / official account / mini-program connected to the camera. This mode can be selected from among roped / cordless jump rope mode, cycling jump rope mode, two-foot jump rope mode, and cross-arm jump rope mode, and / or forward / reverse jump rope mode, maximum continuous jump mode, and jump rope displacement mode. The current jump rope mode is identified and counted by the front AI camera and / or the side AI camera. Multiple jump rope modes can coexist.

[0044] S2. Perform deduplication on multiple images taken by the front AI camera and / or the side AI camera, retain the jump rope information with the highest confidence, and perform model arrangement processing.

[0045] The deduplication process is necessary because the principle of AI image recognition is based on the number of times the camera takes pictures. A camera with 25-60 frames per second will capture many pictures in one effective count. Therefore, the algorithm needs to deduplicate the multiple pictures captured in one effective count and retain the jump rope information with the highest confidence.

[0046] S3. The confidence level of the physical count is obtained by combining the physical jump rope count with the AI ​​rope swing amplitude recognition. The confidence level of the AI ​​comprehensive count is obtained by combining the AI ​​recognition with the sensor joint count. The two confidence levels are compared and the count with the higher confidence level is taken as the final recognition count result.

[0047] Furthermore, the front-facing AI camera also supports anti-counterfeiting measures during jump rope recognition (anti-counterfeiting refers to the ability to detect inaccurate jump rope counts by using non-human movements, such as a paper doll shaking up and down, to exploit vulnerabilities in AI image recognition). When this feature is enabled, the camera will perform N face recognition attempts during a one-minute jump rope session. The timing and number of recognition attempts can be set manually. The default settings are: first, approximately 3 seconds after the start of the jump rope; second, approximately 30 seconds after the start; and third, approximately 58 seconds after the start. At least one valid recognition attempt is required for the jump rope data to be valid; otherwise, the jump rope data is invalid.

[0048] Furthermore, after manually activating the rope skipping mode via a mobile phone or large screen, the AI ​​primarily identifies the rope, with the identification of the jumping human body as a secondary function. It uses both a rope apex recognition model and a foot jump recognition model for identification. Only after the rope mode is manually activated can the continuous jump mode, forward / backward rope skipping mode, and displacement mode be activated. The combined judgment of these two models can prevent violations such as skipping without actually jumping.

[0049] When the cordless jump rope mode is enabled, the AI ​​does not recognize the rope, but only the jumping human body. During the jump, the user does not need to constantly face the camera; as long as they jump normally within the designated area, it will be detected. With cordless mode enabled, continuous jumps, forward / backward rope dancing, and displacement modes are not supported; cycling mode is also not supported. This mode can automatically detect instances of fake jumps by real people.

[0050] Furthermore, the fancy rope skipping method involves skipping rope with arms crossed. When the arm-crossing rope skipping mode is enabled, only the data at the top of the rope is recorded, and no other data is recorded. When the arm-crossing rope skipping mode is disabled, the system will automatically prompt that skipping rope with arms crossed is a foul. At this time, an arm-crossing rope skipping posture recognition model is retrieved, and the rope skipping image captured by the front camera is identified. The model of crossed hands is also identified. When the system detects that the hands are crossed during the rope skipping process, a foul is indicated.

[0051] Furthermore, the cycling jump rope mode is a jump rope method that alternates between both feet. After turning on the cycling jump rope mode, it can not only record the cycling jump rope method, but also the single-foot jump rope method and the foot-alternating jump rope method; the cycling jump rope mode does not support the cordless jump rope mode.

[0052] Among them, the front-facing AI camera identifies and counts the cycling and rope skipping patterns, including:

[0053] A1. Use the AI ​​single-leg jump recognition model to identify whether each jump involves both feet landing simultaneously. If not, count the jump as one jump and record the jump time and confidence level after each valid single-leg jump recognition. If it is a jump, the jump is a violation and the jump is not counted.

[0054] A2. Identify incorrect jump rope postures using an AI error jump recognition model. Record each error when both feet land simultaneously and when tripping over the rope, mark the recognition time, and retain the confidence level of the data.

[0055] A3. The number of times the jump rope appears at the top is identified by the AI ​​rope dancing vertex recognition model. Each time the jump rope appears at the top, it is counted once.

[0056] Furthermore, the double-foot jump rope mode refers to jumping up and down with both feet together. When the double-foot jump rope mode is turned on, only jump rope with both feet together can be recorded as valid jump rope; jump rope with other postures is invalid.

[0057] The front-facing AI camera's recognition and counting of the two-foot jump rope pattern includes:

[0058] B1. Use an AI two-foot jump recognition model to identify whether the feet are together during each jump. If they are together, count the jump and record the time and confidence level. If not, do not count the jump.

[0059] If a jump rope is twirled once and then jumps twice, the AI ​​two-foot jump recognition model only needs to record all jumping actions and the time of each jump, and then compare it with the rope data obtained by the AI ​​rope twirling vertex recognition model.

[0060] B2. Identify incorrect jump rope postures using an AI error jump recognition model. Record each error when the feet are not together or when the rope trips, mark the recognition time, and retain the confidence level of the data.

[0061] B3. The number of times the jump rope appears at the top is identified by the AI ​​rope dancing vertex recognition model. Each time the jump rope appears at the top, it is counted once.

[0062] Furthermore, forward jump rope and reverse jump rope, such as Figure 2 and Figure 3 As shown, when the system detects that the side camera is turned on, the data recognized by the side AI camera can be manually switched between two modes: forward jump rope detection and reverse jump rope detection. The side AI camera's recognition and counting of forward / reverse jump rope modes includes:

[0063] C1. The jump rope sequence detection model is used to sort the jump ropes in time. When the order of the jumping ropes is from back to front, it satisfies the front jumping sequence and is the forward jumping rope mode. When the order of the jumping ropes is from front to back, it satisfies the reverse jumping sequence and is the reverse jumping rope mode.

[0064] C2. When the forward jump rope mode is enabled, the AI ​​jump rope vertex recognition model identifies and counts the number of times the jump rope appears at the top. If the AI ​​jump rope sequence detection model detects that the time sequence of the jump rope in a certain dance meets the reverse dance sequence, then this count is invalid.

[0065] C3. When the reverse jump rope mode is enabled, the AI ​​jump rope vertex recognition model directly identifies the number of times the jump rope appears at the top to complete the counting.

[0066] Furthermore, the skipping rope pattern, such as Figure 4 As shown, when the system detects that the side AI camera is on, it can switch between the maximum consecutive jump mode and the 1-2 jumps-1 rope dance mode. Normally, there are 1-2 jumps and one rope dance. If the maximum consecutive jump occurs, it will display a foul. After manually turning on the maximum consecutive jump mode, it can recognize both the 1-2 jumps-1 rope dance mode and the maximum consecutive jump mode.

[0067] Among them, the side AI camera identifies and counts the maximum consecutive jump patterns, including:

[0068] D1. Enable the AI ​​side foot jump recognition model to perform ground calibration on the jump position, find the relationship between the ground and the jump, and AI calibrate the two feet of the jump. Record the highest point of the jump once, and give the skeleton recognition, the time of the jump recording, the height of the jump, and the confidence of the accuracy of the jump once for each jump.

[0069] D2. Compare the jump time with the jump rope recognition time to obtain an accurate count of jumps.

[0070] Specifically, such as Figure 5 As shown, if two jump count points appear in a single jump rope count, it is determined whether the height of the multiple jumps is consistent. If the height is consistent and the time interval between the two jumps is consistent, it is determined that the jump rope technique point was missed. If the height is inconsistent, showing one high and one low, and the jump time is consistent, the jump rope count is determined to be a two-jump-one-dance rope pattern.

[0071] If there is only one jump count point in a single jump rope count, the jump rope count is judged as a normal one-jump-one-dance rope pattern;

[0072] If only one jump count occurs in multiple jump rope counts, further analysis is performed. If the intervals between counts before and after the time are irregular, it is judged as a jump rope tripping error. If the time interval between multiple records is twice the average time interval, it is judged as a missed jump count. If the time intervals between multiple jump rope counts are regular, it is judged as a maximum consecutive jump pattern.

[0073] Furthermore, the maximum consecutive jump mode recognizes jumping 2-6 times in one rope swing. If the maximum consecutive jump mode is turned off, the system defaults to the maximum consecutive jump being a violation, and the model will automatically determine that the action is a violation when it recognizes the maximum consecutive jump action. If the maximum consecutive jump mode is turned on, the system defaults to the maximum consecutive jump being a non-violation, and the model will automatically determine that the action is a non-violation when it recognizes the maximum consecutive jump action, and record the number of double jumps, triple jumps, quadruple jumps, etc.

[0074] Furthermore, rope displacement recognition refers to situations where the jumper experiences significant displacement or directional rotation during rope skipping, which severely impacts the normal recognition capabilities of the front AI camera. Activating the side AI camera enables the rope displacement mode. In other words, when the human detection model detects a significant shift in the human body, it can switch the human body recognition and rope tip recognition models from data from the front AI camera to data from the side AI camera. The front and side AI cameras can freely switch models, thus better supporting the normal recognition of jumper movements. If the side AI camera is disabled, it will not support jumper movements during skipping. The jumper can only perform rope skipping movements within a designated area, maintaining the front AI camera's view. When significant displacement occurs during skipping, the front AI camera will issue a warning indicating a violation.

[0075] Therefore, both frontal and side AI cameras require an electronic coil and a model to recognize the front and side of the human body. Based on the body's movement, the system can effectively determine whether the jumper is violating regulations within the electronic coil, or whether the recognition model at the top of the jump rope needs to switch between frontal and side AI camera data to identify the jump rope.

[0076] Furthermore, such as Figure 6 As shown, the model orchestration process includes a time interval determination step and a data integration step. The time interval determination step includes:

[0077] Step 1: If the recording time rhythm is stable, the confidence level is increased by 20% for each count;

[0078] Step 2: If there are interruptions in the recording of the time rhythm, record the time interval between the interruptions and mark it as a non-counting point. Compare the time of the interruption with the time of the incorrect jump rope posture identified by the AI ​​error recognition model. When the time of the incorrect jump rope posture is within the time range of the non-counting point, the confidence of the incorrect jump rope posture recognition is increased by 20%.

[0079] Step 3: If the time interval between two records is twice the average time interval, it is recorded as a suspected missed count point. The two time intervals are compared with the incorrect rope skipping postures identified by the AI ​​error recognition model. If the incorrect rope skipping posture is within the suspected missed count time, the confidence of the incorrect rope skipping posture recognition is increased by 20%. If no incorrect rope skipping posture is found within the suspected missed count time range, the missed rope skipping data is automatically supplemented. The confidence of the supplemented data is 60% by default.

[0080] Step 4: If the incorrect posture identified by AI falls on the correct AI recording point, then it is further determined whether the pattern identified by AI is consistent with the current jump rope pattern. If they are consistent, the correct AI record is valid; if they are inconsistent, the correct AI record is invalid.

[0081] Furthermore, the data integration step includes: arranging the data identified by the three models—the recognition model for each pattern, the AI ​​error jump recognition model, and the AI ​​rope dancing vertex recognition model—to obtain the final recognition data; specifically:

[0082] A. When the data for a valid jump rope is too close to the data for an invalid jump rope, the valid information is marked as potentially erroneous. Similarly, when the data for a valid jump is too close to the data for an invalid jump rope, the valid jump rope data is marked as potentially erroneous.

[0083] B. Comparing the error times of suspected jump rope data and suspected jump data with the data from the AI ​​error jump recognition model, the following situation occurs:

[0084] Scenario 1: The AI ​​skipping rope incorrectly identified data, and two other models also marked it as potentially incorrect within a similar time frame. In this case, the skipping rope data was deducted.

[0085] Scenario 2: The AI ​​skipping rope incorrectly identified data, while only one of the other two models was suspected of being marked within a similar time frame. In this case, the skipping rope data was deducted.

[0086] Scenario 3: The AI ​​skipping rope error was identified, but the other two models did not mark it as a possible error within a similar time frame. In this case, the skipping rope data was retained.

[0087] Scenario 4: The AI ​​jump rope incorrectly identified and no data was available. The other two models also marked it as potentially suspicious within a similar time frame. In this case, the jump rope data will be deducted.

[0088] Scenario 5: The AI ​​jump rope incorrectly identified no data, and only one of the other two models had a suspected label within a similar time frame. In this case, the jump rope data will be retained.

[0089] The data integration time can be set according to the computing power available, for example, once every 1 second, 3 seconds, 5 seconds, 10 seconds, 20 seconds, 30 seconds, etc.

[0090] Furthermore, the confidence level of the physical jump rope count, obtained by combining physical jump rope counting with AI rope swing amplitude recognition, includes the following:

[0091] The principle of physical counting jump rope is that the counting module of the jump rope records one jump when it detects that the jump rope has swung 360 degrees in one rotation. When the jump rope has swung less than 360 degrees in one rotation, it is not counted. This can lead to inaccurate counting by the physical counting system because some people's arms are spread out more, resulting in a jump rope amplitude of less than 360 degrees each time. When the AI ​​camera and the counting rope are used simultaneously, the front AI camera automatically starts a model to recognize the rope amplitude to identify the angle between the rope and the handle, specifically including the following:

[0092] 1. Separate recognition of left and right arms. When the angle data of the left and right arms are recognized is less than 5° (which can be adjusted according to the actual situation), it means that the test subject is facing the camera and the recognition data is valid.

[0093] 2. When the recognition data is valid, the recognition angle is determined by the following:

[0094] (1) When the arm angle reaches 90 degrees or more, the confidence level of the rope data is 90%;

[0095] (2) When the arm angle reaches 80-90 degrees, the confidence level of the rope data is 80%;

[0096] (3) When the arm angle reaches 70-80 degrees, the confidence level of the rope data is 70%;

[0097] (4) When the arm angle reaches 60-70 degrees, the confidence level of the rope data is 60%;

[0098] (5) When the arm angle is below 60 degrees, the confidence level of the rope data is 50%;

[0099] The parameters relating the arm angle to the confidence level of the rope data can be adjusted according to the actual situation.

[0100] The data obtained by physically counting jump ropes is combined with the AI ​​rope swing amplitude recognition model, and the resulting data will have a confidence level.

[0101] Furthermore, the confidence level of the AI ​​comprehensive count obtained by combining AI image recognition with sensor joint counting specifically includes the following:

[0102] Step 1: Gear or Hall sensor provides normal jump rope data, which is transmitted to the AI ​​recognition camera in real time via Bluetooth or WiFi.

[0103] Step 2: Capture jump rope video images through a camera, extract frames from the video images according to specified rules, and assign a timing data accurate to 0.01 seconds to each image.

[0104] Step 3: Find a range of time intervals that are not counted due to their relatively long intervals, record the stop time and start time of the current jump rope session, and determine the time range that is not counted.

[0105] Step 4: Identify the human body's tripping posture and the posture of re-dancing the jump rope, and record the time range in which the posture occurs.

[0106] Step 5: Compare the number of abnormal jump rope states within the uncounted time range, and identify any tripping postures. If any are found, deduct from the count; otherwise, do not deduct from the count.

[0107] Step 6: Subtract the number of times the rope handle sensor detects abnormal jump rope states from the number of times the AI ​​image recognition detects them, and the result will be the accurate data.

[0108] Step 7: Using the AI ​​+ sensor combined counting method, the accuracy threshold for the obtained jump rope data is set at 80%.

[0109] This confidence level will be compared with other confidence levels identified by the AI. If the confidence level of the physical jump rope count is greater than the confidence level of the AI ​​count, the physical count will be used as the final result; if the confidence level of the physical count is less than the confidence level of the AI's overall count, the AI ​​count will be used as the final result.

[0110] Finally, the confidence level of the physical count obtained by combining physical jump rope counting with AI rope swing amplitude recognition will be compared with the confidence level of the AI ​​comprehensive count obtained by combining AI recognition with sensor joint counting. The count with higher confidence level will be taken as the final recognition count result.

[0111] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for recognizing and counting multiple jump rope patterns, characterized in that: The identification and counting method includes: The system identifies and counts which of the following jump rope modes—cordless jump rope, cycling jump rope, double-leg jump rope, and double-handed cross-over jump rope—and / or which of the following jump rope modes—forward / reverse jump rope, maximum consecutive jump, and jump rope displacement—is the current jump rope mode. Multiple images captured by the front AI camera and / or the side AI camera are deduplicated, and the jump rope information with the highest confidence is retained and then processed for model arrangement. The confidence level of the physical count is obtained by combining physical jump rope counting with AI rope swing amplitude recognition, and the confidence level of the AI ​​comprehensive count is obtained by combining AI recognition with sensor joint counting. The two confidence levels are compared, and the count with the higher confidence level is taken as the final recognition count result. The model orchestration process includes a time interval determination step and a data integration step. The time interval determination step includes: If the recording time rhythm is stable, the confidence level increases by m with each count; If there are interruptions in the recorded time rhythm, the time interval between the interruptions is recorded and marked as uncounted points. The time of the interruption is compared with the time of the incorrect jump rope posture identified by the AI ​​error recognition model. When the time of the incorrect jump rope posture is within the time range of uncounted points, the confidence of the incorrect jump rope posture recognition is increased by m. If the time interval between two records is twice the average time interval, it is recorded as a suspected missed count point. The two time intervals are compared with the incorrect rope skipping postures identified by the AI ​​error detection model. If the incorrect rope skipping posture is within the suspected missed count time, the confidence of the incorrect rope skipping posture identification is increased by m. If no incorrect rope skipping posture is found within the suspected missed count time range, the missed rope skipping data is automatically supplemented. The confidence of the supplemented data is n by default. If the incorrect posture identified by AI falls on the correct AI recording point, it will further determine whether the pattern identified by AI is consistent with the current jump rope pattern. If they are consistent, the correct AI record is valid; if they are inconsistent, the correct AI record is invalid. The data integration step includes: arranging the data identified by the three models—the recognition model for each pattern, the AI ​​error jump recognition model, and the AI ​​rope dancing vertex recognition model—to obtain the final recognition data.

2. The multi-jump rope pattern recognition and counting method according to claim 1, characterized in that: The front-facing AI camera identifies and counts the cycling and rope skipping patterns, including: The AI ​​single-leg jump recognition model identifies whether each jump involves both feet landing simultaneously. If it is a single-leg jump, it counts as one jump. The jump time and confidence level are recorded after each valid single-leg jump recognition. If it is not a single-leg jump, it is a jump violation and the jump is not counted. The AI-powered error jump detection model identifies incorrect jump rope postures. When both feet land simultaneously and when the rope is tripped, an error is recorded, the detection time is marked, and the confidence level of the data is retained. The AI ​​rope-dancing vertex recognition model identifies the number of times the jump rope appears at the top. Each time the jump rope appears at the top, it is counted once.

3. The multi-jump rope pattern recognition and counting method according to claim 1, characterized in that: The front-facing AI camera recognizes and counts jump rope patterns using a two-footed jump rope technique, including: The AI-powered double-foot jump recognition model identifies whether the feet are together during each jump. If they are together, the jump is counted as one jump, and the jump time and confidence level are recorded. If not, the jump is not counted. The AI-powered error jump detection model identifies incorrect jump rope postures. When the feet are not together or when the rope is tripped, an error is recorded, the detection time is marked, and the confidence level of the data is retained. The AI ​​rope-dancing vertex recognition model identifies the number of times the jump rope appears at the top. Each time the jump rope appears at the top, it is counted once.

4. The multi-jump rope pattern recognition and counting method according to claim 1, characterized in that: The side-mounted AI camera identifies and counts forward / backward jump rope patterns, including: The AI ​​rope skipping sequence detection model sorts the rope skipping in time. When the rope appears from back to front, it satisfies the front skipping sequence and is a forward skipping mode. When the rope appears from front to back, it satisfies the reverse skipping sequence and is a reverse skipping mode. When the forward jump rope mode is enabled, the AI ​​jump rope vertex recognition model identifies and counts the number of times the jump rope appears at the top. If the AI ​​jump rope sequence detection model detects that the time sequence of the jump rope in a certain dance meets the reverse dance sequence, then this count is invalid. When the reverse jump rope mode is enabled, the AI ​​jump rope vertex recognition model directly identifies the number of times the jump rope appears at the top to complete the count.

5. The multi-jump rope pattern recognition and counting method according to claim 1, characterized in that: The side-mounted AI camera identifies and counts the maximum consecutive jump patterns, including: The AI ​​side foot jump recognition model is enabled to perform ground calibration of the jump position, find the relationship between the ground and the jump, and AI calibrates the two feet of the jump. The highest point of the jump is recorded once, and the confidence of skeleton recognition, jump recording time, jump height and jump accuracy for each jump is given. The jump time is compared with the jump rope recognition time to obtain an accurate count of jumps.

6. The multi-jump rope pattern recognition and counting method according to claim 5, characterized in that: The step of comparing the jump time with the jump rope recognition time to obtain an accurate jump count includes: If two jump count points appear in a single jump rope count, it is determined whether the height of the multiple jumps is consistent. If the height is consistent and the time interval between the two jumps is consistent, it is determined that the jump rope technique point was missed. If the height is inconsistent, showing one high and one low, and the jump time is consistent, the jump rope count is determined to be a two-jump-one-dance rope pattern. If there is only one jump count point in a single jump rope count, the jump rope count is judged as a normal one-jump-one-dance rope pattern; If only one jump count occurs in multiple jump rope counts, further analysis is performed. If the intervals between counts before and after the time are irregular, it is judged as a jump rope tripping error. If the time interval between multiple records is twice the average time interval, it is judged as a missed jump count. If the time intervals between multiple jump rope counts are regular, it is judged as a maximum consecutive jump pattern.

Images