Noise reduction regression method and system based on human body motion data collected by depth camera
By performing support vector regression noise reduction processing on human motion data collected by depth cameras, the problems of insufficient data accuracy and high noise are solved, and higher accuracy and stability are achieved, and it is suitable for multi-group data processing.
Patent Information
- Application Number
- CN202411618605.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2019-08-24
- Publication Date
- 2025-02-28
AI Technical Summary
The existing human motion data collected based on depth cameras have the problem of insufficient accuracy, especially in millimeter-level accuracy and close-range measurement, which leads to high data noise and affects the accuracy of subsequent data analysis.
Support vector regression (SVR) algorithm is used to reduce noise regression processing on the human joint data collected by the depth camera to remove noise and obtain a three-dimensional motion trajectory of the joint node that conforms to human mechanics.
Through noise reduction regression processing, the accuracy and stability of the system are significantly improved, the impact of noise on the regression curve is removed, and the error is reduced to less than 1cm, which is suitable for multi-group data processing.
Smart Images
Figure CN119541048A_ABST
Abstract
Description
[0001] This application is a divisional application of a patent application entitled "Noise Reduction Regression Method and System for Collecting Human Motion Data Based on Depth Camera". The application date of the original application is August 24, 2019, and the application number is 201910786225.7. Technical Field
[0002] The present application relates to the technical field of motion data processing, and in particular to a denoising regression method and system for collecting human motion data based on a depth camera. Background Art
[0003] Now, human motion capture technology has been widely used in sports. This technology can capture the movements of athletes for quantitative analysis. Combined with the principles of human physiology and physics, it studies improvement methods to enable sports training to break away from the state of relying purely on experience and enter the era of theorization and digitization.
[0004] There are currently two existing and more mainstream methods for collecting human motion data:
[0005] 1. Contact collection method: Use sensors to mark each joint of the human body, collect data in space, and then use computer processing to identify, track, and calculate spatial coordinates of the marked points to realize the collection of human motion data.
[0006] 2. Non-contact acquisition method: Obtain human motion data through an optical depth camera. The optical depth camera - Kinect V2, which includes a Color Camera and a Deep Sensor (IR Camera + IR Project). The Deep Sensor uses the Time of Flight (TOF) method to calculate the distance of the object being measured from the camera by detecting the flight time after the projected light pulse contacts the object being measured and returns. Then, the camera's built-in hardware and software system establishes coordinates on the two-dimensional image plane, and with the distance information obtained previously and the position information of the skeletal joints identified and marked by its own SDK, the coordinates of each joint of the human body in three-dimensional space can be obtained to realize the collection of human motion data.
[0007] The disadvantages of existing methods for collecting human motion data are:
[0008] 1. Contact collection method: The system is expensive, the preliminary preparation work is complicated, the post-processing workload is large, and it is impossible to achieve real-time or short-term analysis, which is not suitable for civilian use;
[0009] 2. Non-contact collection method: It has high requirements for the accuracy of time measurement, and it is difficult to achieve millimeter-level accuracy. The accuracy of close-range measurement (within 1m) is low.
[0010] As mentioned above, the non-contact acquisition method based on the depth camera can only achieve centimeter-level accuracy, and it is difficult to reach millimeter-level accuracy. The reason is that the built-in hardware and inherent algorithms of the depth camera, including the marking, tracking and calculation of spatial coordinates of human joints, will cause certain errors in the collected data. Here are a few examples to illustrate:
[0011] 1. Due to clothing material, angle, obstruction, high-speed movement and other reasons, the SDK algorithm built into the depth camera has deviations in joint point recognition, and its built-in inference point algorithm also has errors with the actual joint points, resulting in inaccurate coordinates and even introducing large error information; this will cause greater errors in subsequent data analysis and calculations, which will result in the inability to obtain accurate angles in subsequent project calculations, making the function impossible to implement. If manual screening is performed to exclude coordinate points with large offsets, it will not only be time-consuming and labor-intensive, but also fail to achieve universality, which does not meet the original intention of designing the system;
[0012] 2. Human body movement is based on bones as levers, joints as pivots, and muscle contraction as power, which is consistent with the characteristics of human mechanics. When the camera cannot accurately collect joint point information, it draws inference points. Although the human skeleton information is consistent with the human body structure at a certain moment, it violates the human body's movement process. After all, the movement trajectory of each joint in the human body is linear during movement.
[0013] During the entire human motion process, most of the joint point coordinate information collected by the depth camera is accurate, but not completely linear; therefore, it is necessary to perform regression processing on these relatively accurate data to obtain a regression curve to reasonably fit the human motion trajectory. The two problems mentioned above are also inevitable, and they are called data noise, which affects the overall data representation and subsequent use of values. Therefore, it is necessary to reasonably process the existing data noise, eliminate the noise impact, and improve the accuracy and robustness of this system. Summary of the invention
[0014] In order to solve the problems mentioned above, the present invention provides a method and system for noise reduction and regression based on human motion data collected by a depth camera.
[0015] The present application discloses a denoising regression method based on collecting human motion data with a depth camera, characterized in that the denoising regression method based on collecting human motion data with a depth camera comprises:
[0016] Collect human joint data based on depth camera;
[0017] The human joint point data is subjected to denoising regression processing to obtain a three-dimensional motion trajectory of the joint point that conforms to human body mechanics and remove noise; the denoising regression processing method is a support vector regression algorithm; the denoising regression processing can remove the broken line or point jump phenomenon caused by camera acquisition errors or the introduction of inferred point data;
[0018] Perform subsequent data analysis on the three-dimensional motion trajectory of the joint points.
[0019] As a further improvement of the present application, the human joint point data is subjected to denoising regression processing, including:
[0020] Inputting the human body joint point data into the corresponding support vector regression algorithm;
[0021] Adjust the relevant parameters of the support vector regression algorithm;
[0022] The support vector regression algorithm with adjusted parameters is used to perform noise reduction regression processing on the human joint point data.
[0023] As a further improvement of the present application, the support vector regression algorithm is encapsulated in the third-party module sklearn of Python and is directly called when used.
[0024] The present application also discloses a denoising regression system based on collecting human motion data with a depth camera, and the denoising regression system based on collecting human motion data with a depth camera comprises:
[0025] Depth camera, used to collect data on human joints;
[0026] A denoising regression processing module is used to perform denoising regression processing on the human joint point data to obtain a three-dimensional motion trajectory of the joint point that conforms to human body mechanics and remove noise; the denoising regression processing method is a support vector regression algorithm; the denoising regression processing can remove the broken line or point jump phenomenon caused by camera acquisition errors or the introduction of inferred point data;
[0027] The data analysis module is used to perform subsequent data analysis on the three-dimensional motion trajectory of the joint points.
[0028] As a further improvement of the present application, the denoising regression processing module is used to:
[0029] Inputting the human body joint point data into the corresponding support vector regression algorithm;
[0030] Adjust the relevant parameters of the support vector regression algorithm;
[0031] The support vector regression algorithm with adjusted parameters is used to perform noise reduction regression processing on the human joint point data.
[0032] As a further improvement of the present application, the support vector regression algorithm is encapsulated in the third-party module sklearn of Python and is directly called when used.
[0033] Compared with the prior art, the beneficial effects of this application are:
[0034] The present application can regress the motion trajectory of the joints that conforms to human body mechanics through the human joint data collected by the existing depth camera, while excluding the noise value, removing the influence of noise on the regression curve, and improving the accuracy and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 This is a flow chart of a denoising regression method based on collecting human motion data with a depth camera, disclosed in an embodiment of the present application;
[0036] Figure 2 A framework diagram of a denoising regression system based on collecting human motion data with a depth camera according to an embodiment of the present application;
[0037] Figure 3 A human joint point data graph collected by a depth camera disclosed in an embodiment of the present application;
[0038] Figure 4 A data comparison diagram of a denoising regression process and a data comparison diagram of a non-denoising regression process disclosed in an embodiment of the present application;
[0039] Figure 5 A plurality of groups of human joint point data graphs collected by a depth camera disclosed in an embodiment of the present application;
[0040] Figure 6 A comparison diagram of multiple groups of data after denoising regression processing and without denoising regression processing disclosed in an embodiment of the present application. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0042] The problem to be solved in this application is: how to regress the motion trajectory of the joint points that conforms to human ergonomics through the existing coordinate points; at the same time, exclude the noise value, remove the influence of noise on the regression curve, and improve the accuracy and stability of the system.
[0043] To this end, the present application discloses a denoising regression method and system for collecting human motion data based on a depth camera, including: collecting human joint point data based on a depth camera; performing denoising regression processing on the human joint point data, the denoising regression processing adopts a support vector regression algorithm for processing; and performing subsequent data analysis on the denoising regression processed data.
[0044] The present application is further described in detail below with reference to the accompanying drawings:
[0045] like Figure 1 As shown, the present application provides a denoising regression method based on collecting human motion data with a depth camera, comprising:
[0046] S1. Collecting human joint data based on a depth camera; wherein:
[0047] The depth camera can use the existing Kinect camera;
[0048] The human joint data collected by the depth camera is as follows: Figure 3 As shown, Figure 3 Specifically, it is a trajectory diagram of the wrist joint of the human body in three-dimensional space when performing a complete forehand attack; the two-dimensional plane diagrams named "==X==", "==Y==", and "==Z==" are curve diagrams of the wrist joint in three-dimensional space changing with time along different coordinate axes (X, Y, and Z axes). Figure 3 The data analysis shows that during a set of forehand attacking movements, broken lines and point jumps occurred, which do not conform to the laws of human movement. The cause of this problem is that some of the points are the result of point data collected by the camera due to errors or inferred due to insufficient data.
[0049] S2. Performing denoising and regression processing on the human joint point data; wherein:
[0050] In order to eliminate the broken lines, point jumps, etc. analyzed in S1, this application uses a vector regression algorithm to perform noise reduction regression processing on the human joint point data after trying methods such as linear regression and ellipsoid regression to obtain cross-sectional curves.
[0051] Specific:
[0052] The support vector regression used in this application is referred to as SVR; the implementation of this mathematical method has been implemented and encapsulated in the third-party Python module sklearn commonly used in machine learning. There is no need to implement the algorithm and can be called directly. In addition to the need to pass in data after calling SVR, other parameters that need to be adjusted need to be adjusted accordingly according to different situations, and there is no fixed pattern.
[0053] The data comparison between the denoising regression process and the non-denoising regression process is as follows: Figure 4 As shown, Figure 4 The curve in Figure D represents: the “distance” between the acquisition point and the regression point in three-dimensional space at the same time point; Figure 4 The straight line in the D graph represents the average value obtained through all "distances" during this period of time. This graph shows the actual error value reduced by the regression curve at each time point and the average error reduced during the entire period of time. This shows that the denoising regression algorithm has a significant impact on the actual situation and improves accuracy.
[0054] pass Figure 4 It can be seen that:
[0055] 1. The curve passes through most of the collected data points;
[0056] 2. Remove points with large deviations, discounts or jumps;
[0057] 3. The curve is smoother and conforms to the actual motion trajectory;
[0058] 4. The error is reduced to less than 1cm.
[0059] Similarly, this method is also effective for multiple sets of data, without the need for manual screening, and good data can be obtained after automatic execution. Figure 5 , Figure 6 shown; through Figure 5 , Figure 6 It can be seen that:
[0060] Figure 5 , Figure 6 After performing denoising and regression processing on two more sets of data, we can see that good results are achieved in the connection part between the two sets of actions, which is also the place where data collection is most inaccurate.
[0061] S3. Perform subsequent data analysis on the data processed by denoising regression.
[0062] like Figure 2 As shown, the present application provides a denoising regression system for collecting human motion data based on a depth camera, comprising:
[0063] A depth camera, used to implement the above S1;
[0064] A denoising regression processing module, used to implement the above S2;
[0065] The data analysis module is used to implement the above S3.
[0066] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A denoising regression method based on human motion data collected by a depth camera, characterized in that: The denoising regression method based on collecting human motion data with a depth camera includes: Collect human joint data based on depth camera; The human joint point data is subjected to denoising regression processing to obtain a three-dimensional motion trajectory of the joint point that conforms to human body mechanics and remove noise; the denoising regression processing method is a support vector regression algorithm; the denoising regression processing can remove the broken line or point jump phenomenon caused by camera acquisition errors or the introduction of inferred point data; Perform subsequent data analysis on the three-dimensional motion trajectory of the joint points.
2. The denoising regression method according to claim 1, characterized in that: The human joint point data is subjected to denoising and regression processing, including: Inputting the human body joint point data into the corresponding support vector regression algorithm; Adjust the relevant parameters of the support vector regression algorithm; The support vector regression algorithm with adjusted parameters is used to perform noise reduction regression processing on the human joint point data.
3. The denoising regression method according to claim 1 or 2, characterized in that: The support vector regression algorithm is encapsulated in the third-party module sklearn of Python and is directly called when used.
4. A denoising regression system based on human motion data collected by a depth camera, characterized in that: The denoising regression system based on collecting human motion data with a depth camera comprises: Depth camera, used to collect data on human joints; A denoising regression processing module is used to perform denoising regression processing on the human joint point data to obtain a three-dimensional motion trajectory of the joint point that conforms to human body mechanics and remove noise; the denoising regression processing method is a support vector regression algorithm; the denoising regression processing can remove the broken line or point jump phenomenon caused by camera acquisition errors or the introduction of inferred point data; The data analysis module is used to perform subsequent data analysis on the three-dimensional motion trajectory of the joint points.
5. The denoising regression system according to claim 4, characterized in that: The denoising regression processing module is used to: Inputting the human body joint point data into the corresponding support vector regression algorithm; Adjust the relevant parameters of the support vector regression algorithm; The support vector regression algorithm with adjusted parameters is used to perform noise reduction regression processing on the human joint point data.
6. The denoising regression system according to claim 4 or 5, characterized in that: The support vector regression algorithm is encapsulated in the third-party module sklearn of Python and is directly called when used.