Gesture recognition device testing method, apparatus, and system

Abstract

This application provides a method, apparatus, and system for testing gesture recognition devices, relating to three-dimensional interaction technology. The method includes: acquiring hand test data of the gesture recognition device under test and hand ground truth data of the gesture recognition ground truth device, based on a gesture recognition ground truth device and a gesture recognition device under test with relatively fixed positions; aligning the hand ground truth data and hand test data using a target transformation relationship; the target transformation relationship is determined through a preset motion process of a rigid body model of the hand, which is constructed from at least two first finger root nodes and at least one palm node, or at least two palm nodes and at least one first finger root node; the relative positions of the first finger root nodes and palm nodes are fixed when the hand is in different motion states; and determining the test result based on the aligned hand ground truth data and hand test data. This application enables efficient testing of the gesture recognition device under test.

Description

Technical Field

[0001] This application relates to three-dimensional interactive technology, and more particularly to a method, apparatus and system for testing gesture recognition devices. Background Technology

[0002] With the continuous development of human-computer interaction technology, gesture recognition devices have broad application prospects in various fields. In order to meet market demands, testing the accuracy and performance of gesture recognition devices has become particularly important.

[0003] Currently, in known technologies, for commercial gesture recognition devices, manufacturers can use the raw datasets collected by the gesture recognition devices to evaluate the recognition accuracy of the devices in order to determine their performance.

[0004] However, for users other than the equipment manufacturers, it is difficult to evaluate the recognition accuracy of gesture recognition devices because they cannot obtain test equipment with raw camera data. Summary of the Invention

[0005] Based on this, this application provides a method, apparatus and system for testing gesture recognition devices. The method for testing gesture recognition devices can evaluate the recognition accuracy of gesture recognition devices in the absence of raw data collected by the gesture recognition devices.

[0006] In a first aspect, this application provides a method for testing a gesture recognition device, the method comprising:

[0007] Based on a relatively fixed-position gesture recognition ground truth device and a gesture recognition device under test, hand test data of the gesture recognition device under test and hand ground truth data of the gesture recognition ground truth device are obtained.

[0008] The target transformation relationship is used to align the true hand data and the hand test data; the target transformation relationship is determined by a preset motion process of the rigid body model of the hand, which is constructed by at least two first finger root nodes and at least one palm node, or at least two palm nodes and at least one first finger root node; the relative positions of the first finger root nodes and the palm nodes are fixed when the hand is in different motion states.

[0009] The test results are determined based on the aligned ground truth data of the hand and the hand test data.

[0010] Secondly, this application provides a gesture recognition device testing apparatus, the apparatus comprising:

[0011] The acquisition module is used to acquire hand test data of the gesture recognition device under test and hand ground truth data of the gesture recognition device under test, based on the gesture recognition ground truth device and the gesture recognition device under test, which are in relatively fixed positions.

[0012] The processing module is used to align the true hand data and the hand test data using a target transformation relationship; the target transformation relationship is determined by a preset motion process of the rigid body model of the hand, which is constructed from at least two first finger root nodes and at least one palm node, or at least two palm nodes and at least one first finger root node; the relative positions of the first finger root nodes and the palm nodes are fixed when the hand is in different motion states.

[0013] The determination module is used to determine the test results based on the aligned hand ground truth data and hand test data.

[0014] Thirdly, this application provides an electronic device, including a processor and a memory communicatively connected to the processor;

[0015] The memory stores computer-executed instructions;

[0016] The processor executes computer execution instructions stored in the memory to implement the method as described in any of the first aspects.

[0017] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any of the first aspects.

[0018] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method as described in any of the first aspects.

[0019] In a sixth aspect, this application provides a gesture recognition device testing system, the system comprising an electronic device as described in the third aspect and a gesture recognition ground truth device, wherein the electronic device acquires hand ground truth data through the gesture recognition ground truth device.

[0020] The gesture recognition device testing method, apparatus, and system provided in this application are used to test gesture recognition devices. Specifically, when testing a gesture recognition device using the method of this application, hand ground truth data is acquired through a gesture recognition ground truth device with known internal parameters. The hand ground truth data and the hand test data of the gesture recognition device under test are then aligned based on a target transformation relationship. Finally, the test result is obtained based on the aligned hand ground truth data and hand test data. This allows for the evaluation of the recognition accuracy of the gesture recognition device even in the absence of the original data collected by the device, and it has low requirements for the evaluation environment. In this process, since the target transformation relationship is determined through a preset motion process of a rigid hand model, the rigid hand model involves a small number of nodes, which simplifies the algorithm for determining the target transformation relationship and improves the testing efficiency of the gesture recognition device under test. Furthermore, the nodes of the rigid hand model maintain a relatively fixed geometric relationship during the preset motion process, making the target transformation relationship more accurate. This results in a more accurate alignment of the hand ground truth data and hand test data, thereby improving the testing accuracy of the gesture recognition device. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram illustrating an application scenario of a gesture recognition device testing method provided in an embodiment of this application.

[0023] Figure 2 This is a schematic diagram of the structure of a gesture recognition device testing system provided in an embodiment of this application;

[0024] Figure 3A A flowchart illustrating a gesture recognition device testing method provided in this application embodiment. Figure 1 ;

[0025] Figure 3B A schematic diagram of a hand node provided in an embodiment of this application;

[0026] Figure 4 A flowchart illustrating a gesture recognition device testing method provided in this application embodiment. Figure 2 ;

[0027] Figure 5 A flowchart illustrating a testing method for a gesture recognition device provided in this application is shown in Figure 3.

[0028] Figure 6 A flowchart illustrating a gesture recognition device testing method provided in this application embodiment. Figure 4 ;

[0029] Figure 7 This is a schematic diagram of the structure of a gesture recognition device testing apparatus provided in an embodiment of this application;

[0030] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0031] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. The described embodiments are only some, not all, of the embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0033] In known technologies, device manufacturers typically use raw data captured by the camera of a gesture recognition device to test and evaluate its performance. Understandably, testing gesture recognition devices presents greater complexity and uncertainty for users outside of device manufacturers. Specifically, users outside of device manufacturers cannot obtain the raw camera data from the gesture recognition device under test. They may not be able to use the same dataset for testing and need to collect data themselves and design complex test environments. For example, they may need to stain the fingers with infrared fluorescence and then use a multi-view motion capture camera to locate key points of the hand in real time. Furthermore, the data acquired by the multi-view motion capture camera and the gesture recognition device under test require complex processing, all of which increase testing complexity and lead to inefficiencies in the entire testing process.

[0034] Therefore, embodiments of this application provide a method, apparatus, and system for testing gesture recognition devices to solve the aforementioned problems. Specifically, in the method of this application, during the same application process, data obtained by a gesture recognition ground truth device with known internal parameters is used as hand ground truth data, and data obtained by the gesture recognition device under test is used as hand test data. Then, the hand ground truth data and hand test data are aligned based on a target transformation relationship. Finally, the test result is obtained based on the aligned hand ground truth data and hand test data.

[0035] The method in this application introduces a gesture recognition ground truth device and aligns the hand ground truth data and hand test data. This allows for the evaluation of the recognition accuracy of the gesture recognition device even in the absence of raw data collected by the device, with low requirements for the evaluation environment. Furthermore, this application determines the target transformation relationship using data from a rigid body model of the hand during a preset motion process. The rigid body model involves a small number of nodes, simplifying the algorithm for determining the target transformation relationship and improving the testing efficiency of the gesture recognition device under test. Moreover, the nodes within the rigid body model maintain relatively fixed geometric relationships during the preset motion process, making the target transformation relationship more accurate. This results in more accurate alignment between the hand ground truth data and hand test data, thereby improving the testing accuracy of the gesture recognition device.

[0036] It is understood that the gesture recognition device testing method of this application is applicable to scenarios involving accuracy testing of any gesture recognition device. For example, Figure 1 This is a schematic diagram illustrating an application scenario of a gesture recognition device testing method provided in an embodiment of this application.

[0037] like Figure 1 As shown, the gesture recognition device testing method of this application can be used in scenarios where accuracy testing is performed on head-mounted display devices that utilize gesture recognition technology, i.e., the gesture recognition device under test is a head-mounted display device that utilizes gesture recognition technology. Specifically, the method of this application is executed by any electronic device. When performing accuracy testing on the head-mounted display device, the electronic device obtains the ground truth data of the hand of the subject under test through a gesture recognition ground truth device, and obtains the hand test data of the subject under test through the head-mounted display device. Further, the electronic device performs alignment processing on the hand ground truth data and hand test data based on a target transformation relationship, and determines the test result based on the aligned hand ground truth data and hand test data.

[0038] It is understandable that electronic devices can connect to head-mounted displays and hand-based motion measurement devices via wired or wireless means. Furthermore, the subject of the measurement can be a human hand.

[0039] When testing head-mounted displays, determining the target conversion relationship between the head-mounted display and the gesture recognition ground truth device has the advantage of low processing complexity, which can effectively reduce the complexity of the entire testing process of the head-mounted display and thus help ensure testing efficiency.

[0040] This application provides a gesture recognition device testing system, specifically... Figure 2 This is a schematic diagram of the structure of a gesture recognition device testing system provided in an embodiment of this application, as shown below. Figure 2 As shown, the system includes an electronic device and a gesture recognition ground truth device. The electronic device is used to implement the gesture recognition device testing method of this application. The electronic device is communicatively connected to the gesture recognition ground truth device and is used to acquire hand ground truth data through the gesture recognition ground truth device.

[0041] In this embodiment, the gesture recognition ground truth device may include multiple RGB-D cameras with fixed positions and different shooting angles, used to collect hand data of the subject under test from different angles. It is understood that the multiple RGB-D cameras are located at different heights and / or different angles relative to the subject under test, used to achieve the collection of hand data of the subject under test from different angles. More specifically, such as... Figure 2 As shown, it can include six cameras: cam1, cam2, cam3, cam4, cam5, and cam6. The six cameras are fixedly connected to each other by structural components.

[0042] It is understood that the gesture recognition truth device can be any other device with known internal parameters, and this embodiment does not limit this.

[0043] It is understood that in practical applications, the electronic device and the gesture recognition ground truth device included in the gesture recognition device testing system can be independent devices, or the electronic device can be integrated into the gesture recognition ground truth device, or some functional modules of the electronic device can be integrated into the gesture recognition ground truth device. The gesture recognition ground truth device integrating the electronic device, or the electronic device and the gesture recognition ground truth device integrating some functional modules of the electronic device, can work together to implement the gesture recognition device testing method of this application. This embodiment does not limit this.

[0044] This application provides a method for testing a gesture recognition device. The content of this application will be described in detail below with reference to the accompanying drawings, so that those skilled in the art can have a clearer and more detailed understanding of the content of this application.

[0045] It should be noted that the following content is based on the foregoing embodiments. Figure 2 Taking the implementation of a gesture recognition device testing system as an example, this paper provides a detailed explanation of the method used in this application. It is understood that for... Figure 2The gesture recognition device testing system shown in this application is executed by an electronic device.

[0046] Figure 3A A flowchart illustrating a gesture recognition device testing method provided in this application embodiment. Figure 1 ,like Figure 3A As shown, the method provided in this application embodiment includes:

[0047] S301, based on a relatively fixed-position gesture recognition ground truth device and a gesture recognition device under test, acquires hand test data of the gesture recognition device under test and hand ground truth data of the gesture recognition ground truth device.

[0048] Understandably, the gesture recognition device under test and the gesture recognition ground truth device are used to identify the subject under test and capture the movements of the subject under test within its field of view. In this embodiment, the subject under test is specifically the hand. In this embodiment, the hand ground truth data is acquired by the gesture recognition ground truth device when the hand performs the test action, and includes multiple sets of pose data corresponding to the hand. The hand test data is acquired by the gesture recognition device under test when the hand performs the test action, and includes multiple sets of pose data corresponding to the hand.

[0049] Specifically, in this embodiment, the gesture recognition device under test and the gesture recognition truth value device are rigidly connected to achieve a relatively fixed position for the gesture recognition device under test and the gesture recognition truth value device. In practical applications, the gesture recognition device under test and the gesture recognition truth value device can also achieve a relatively fixed position through any of the following methods: welding, adhesive connection, snap-fit, and fasteners, etc., which are not limited in this embodiment. For example, the gesture recognition device under test and the gesture recognition truth value device can also be installed on two relatively fixed mounting bases, such as the gesture recognition device under test being installed on a tooling table and the gesture recognition truth value device being installed on a bracket.

[0050] During the test, the subject performs test actions such as pinch, click, and move within the field of view of the gesture recognition device and the gesture recognition ground truth device. The gesture recognition device and the gesture recognition ground truth device collect data on the subject during the test and transmit the collected results to the electronic device, so that the electronic device can obtain hand test data and real hand data.

[0051] In this embodiment, the electronic device and the gesture recognition truth-based device are communicatively connected, and the gesture recognition truth-based device is controlled by the electronic device. The gesture recognition truth-based device and the electronic device share the same time source. The electronic device can obtain hand truth data through the gesture recognition truth-based device by sending corresponding control signals to it.

[0052] In this embodiment, the electronic device obtains hand test data from the gesture recognition device under test by calling relevant functions in the SDK. It is understood that the hand test data obtained by the gesture recognition device under test is processed data, not the raw data captured by the device's camera.

[0053] Specifically, in this embodiment, both hand test data and actual hand data include pose data of at least one position of the hand during the test action. For example, Figure 3B This is a schematic diagram of a hand node provided in an embodiment of the present application. If the hand is divided into 21 hand nodes, the gesture recognition ground truth device is the same as the gesture recognition device under test, and has the ability to recognize 21 hand nodes. On this basis, the hand test data and the real hand data specifically include the pose data of at least one hand node during the test action, preferably six-degree-of-freedom pose data.

[0054] It is understood that hand test data and hand ground truth data specifically include six-DOF pose data corresponding to several hand nodes, determined by the test requirements. When testing the gesture recognition device under test, the electronic device controls the gesture recognition ground truth device to collect six-DOF pose data of several hand nodes according to the test requirements. After receiving the data collected by the gesture recognition device under test, the electronic device selects the six-DOF pose data of the corresponding hand nodes as the hand test data according to the test requirements. In practical applications, the user can also configure the gesture recognition device under test based on the test requirements, causing the device to collect the six-DOF pose data of the hand nodes indicated by the test requirements; this embodiment does not limit this.

[0055] S302 uses a target transformation relationship to align the true hand data and the hand test data.

[0056] The target transformation relationship is determined by the preset motion process of the rigid body model of the hand. The rigid body model of the hand is constructed by at least two first finger root nodes and at least one palm node, or at least two palm nodes and at least one first finger root node. The relative positions of the first finger root nodes and the palm nodes are fixed when the hand is in different motion states.

[0057] In this embodiment, hand nodes are categorized into four types: first finger root nodes, second finger root nodes, palm nodes, and other finger nodes. The relative positions of the first finger root nodes and palm nodes used to construct the rigid body model of the hand are fixed. Specifically, when determining the hand nodes used to construct the rigid body model of the hand, the electronic device can use vector and matrix operations to verify whether the relative positions between hand nodes are fixed, and determine the hand nodes used to construct the rigid body model of the hand based on the verification results.

[0058] Furthermore, in practical applications, electronic devices can also pre-verify whether the relative positional relationships between each hand node are fixed and store the verification result for each hand node. It is understood that the verification result indicates the connection relationship between that hand node and other hand nodes, including connections where the relative position remains unchanged and connections where the relative position changes. Based on this, when determining the hand nodes used to construct the rigid body model of the hand, the electronic device identifies the hand nodes by looking up the verification result of each hand node, thereby saving time in constructing the rigid body model of the hand and thus improving testing efficiency.

[0059] It is understandable that the second finger root node and other finger nodes are finger root nodes that cannot maintain a relatively fixed geometric relationship with the first finger root node and the palm node. If this embodiment divides the hand into 21 hand nodes, specifically, as follows... Figure 3B As shown, it can be understood that the first finger root node includes four hand nodes: index finger root node 5, middle finger root node 9, ring finger root node 13, and little finger root node 17. The second finger root node includes thumb root node 1. The palm node includes palm root node 0. The remaining hand nodes are other finger nodes.

[0060] As can be seen from the foregoing, when the hand makes any gesture, the relative positional relationship between the first finger root node and the palm root node 0 will not change, and they can be calculated using the same coordinate system. Therefore, the first finger root node and the palm root node 0 can be used as a rigid body to determine the target transformation relationship between the gesture recognition device under test and the gesture recognition ground truth device.

[0061] It should be understood that in this embodiment, the hand is set to 21 hand nodes. In practical applications, more finger nodes and palm nodes may be set. Therefore, in this embodiment, it is further proposed to use at least two first finger root nodes and at least one palm node, or at least two palm nodes and at least one first finger root node, to construct the rigid body model of the hand, so as to further reduce the processing complexity.

[0062] It is worth noting that the aforementioned palm nodes should be palm nodes that can maintain a relatively fixed geometric relationship with the first finger nodes, such as... Figure 3B The joint point 0 shown can be either the palm root node, the palm center node, or other palm nodes that can maintain a relatively fixed geometric relationship with the first finger node. This embodiment does not limit this.

[0063] Based on this, the process of determining the target transformation relationship based on the rigid body model of the hand in this embodiment is as follows: The rigid body model of the hand (i.e., the hand) performs a preset motion process. During this process, the gesture recognition device under test collects the motion data of the hand (or the rigid body model of the hand), and the electronic device controls the gesture recognition ground truth device to collect the motion data of the rigid body model of the hand. Based on the operation data obtained by the gesture recognition device under test and the gesture recognition device, the target transformation relationship is determined. This target transformation relationship is the transformation relationship obtained through the entire hand. The preset motion process includes translational motion and rotational motion. Translational motion refers to movement in space along the three directions of XYZ, and rotational motion refers to flipping in space along the pitch, yaw, and roll directions.

[0064] In this embodiment, the target transformation relationship includes a time transformation relationship and a spatial transformation relationship. The electronic device aligns the hand test data and the hand ground truth data based on the time transformation relationship and the spatial transformation relationship.

[0065] Specifically, electronic devices can transform hand test data based on time and space transformation relationships to obtain hand test data in time and space corresponding to the true hand data.

[0066] In practical applications, electronic devices can also convert hand truth data based on time conversion relationships and spatial conversion relationships, or convert hand test data based on time conversion relationships and hand truth data based on spatial conversion relationships, etc. This embodiment does not limit this, as long as the converted hand test data and hand truth data are consistent in time and space.

[0067] S303, based on the aligned hand ground truth data and hand test data, determines the test result.

[0068] Specifically, for at least one hand node, the electronic device can determine the test trajectory corresponding to the gesture recognition device under test based on its corresponding aligned hand test data, and determine the ground truth trajectory corresponding to the gesture recognition device under test based on its corresponding aligned hand ground truth data. Then, the test result is determined by comparing the test trajectory and the ground truth trajectory. The process of determining the corresponding test trajectory based on the hand test data and the corresponding ground truth trajectory based on the hand ground truth data can be implemented using an interpolation algorithm, specifically, using B-spline interpolation.

[0069] More specifically, electronic devices can determine test results by calculating metrics such as spatial accuracy and / or jitter based on the true value trajectory and the test trajectory. For example, the electronic device calculates spatial accuracy and jitter metrics based on the true value trajectory and the test trajectory. If the spatial accuracy and jitter metrics are within acceptable ranges, the test result indicates that the test has passed; otherwise, the test result indicates that the test has failed.

[0070] In practical applications, for at least one hand node, the electronic device can also directly compare the aligned hand test data and the true hand data to obtain the test results of the gesture recognition device for the hand node. This embodiment does not limit this.

[0071] In this embodiment, the electronic device aligns the hand test data and the hand ground truth data based on time and space transformation relationships, thereby making the hand test data and the hand ground truth data comparable and enabling more accurate alignment and comparison.

[0072] In the method provided in this application embodiment, the electronic device determines the target transformation relationship between the gesture recognition device under test and the gesture recognition ground truth device based on the preset motion process of the rigid body model of the hand. The acquired hand test data and hand ground truth data are aligned according to the target transformation relationship. Then, the test result is obtained through the aligned hand test data and hand ground truth data, thereby realizing the testing of the gesture recognition device under test. This solves the problem in the known technology that it is difficult to evaluate the recognition accuracy of the gesture recognition device because the test device cannot obtain the original data of the camera.

[0073] Furthermore, in this embodiment, since the target transformation relationship is determined by the rigid body model of the hand, the process of determining the target transformation relationship is simplified, thereby reducing the complexity of the entire testing process and thus improving testing efficiency.

[0074] This application also provides a method embodiment for illustrating how to obtain the target transformation relationship between the gesture recognition device under test and the gesture recognition ground truth device. Figure 4 A flowchart illustrating a gesture recognition device testing method provided in this application embodiment. Figure 2 ,like Figure 4 As shown, the method for determining the time conversion relationship between the gesture recognition device under test and the gesture recognition truth device in this embodiment includes:

[0075] S401, obtains the true value calibration data of the hand rigid body model when it performs a preset motion process through a gesture recognition truth value device.

[0076] The true calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand.

[0077] In this embodiment, the pose data specifically refers to six-degree-of-freedom pose data, including translation data and rotation data. The rotation data indicates rotation information about the pitch, yaw, and roll axes, and can be expressed in different forms such as Euler angles, quaternions, or rotation matrices. The translation information indicates movement information along the x, y, and z axes, typically represented by a translation vector (x, y, z). The six-degree-of-freedom pose data is generated by the rigid body model of the hand during a preset motion; that is, a set of six-degree-of-freedom pose data corresponds to a specific moment.

[0078] Specifically, in this embodiment, the truth value calibration data is determined through the following process:

[0079] Step 1: Obtain the ground truth data of the rigid body model of the hand when it performs a preset motion process through a gesture recognition ground truth device.

[0080] Specifically, the ground truth data in step 1 includes spatial position data of at least two first finger root nodes and at least one palm node. The spatial position data is used to indicate the position information of at least two first finger root nodes and at least one palm node in three-dimensional space.

[0081] Step 2: Determine the target spatial position data of the geometric center of the rigid body model of the hand based on at least two first finger root nodes and at least one palm node.

[0082] In this embodiment, the electronic device determines the geometric center of the rigid body model of the hand by using the coordinates of at least two first finger root nodes and at least one palm node in the hand coordinate system. Specifically, the average value of the coordinates of at least two first finger root nodes and at least one palm node is calculated, and the point corresponding to the corresponding average value is taken as the geometric center.

[0083] In practical applications, the geometric center can also be determined by other methods, such as weighted average or centroid calculation. This embodiment does not limit this method.

[0084] Furthermore, the spatial location data of the geometric center is determined using the spatial location data of at least two first finger root nodes and at least one palm node.

[0085] Step 3: Based on the target spatial location data and the rotation data of any hand node in the rigid body model of the hand, obtain the true value calibration data corresponding to the geometric center.

[0086] In this embodiment, the electronic device uses the rotation data of any hand node in the rigid body model of the hand, along with the target spatial position data obtained in the aforementioned steps, as ground truth calibration data. Specifically, the electronic device only needs to be able to acquire the rotation data of that hand node.

[0087] It is understandable that if the rotation data of all the hand nodes in the rigid body model can be obtained, the obtained rotation data can be averaged to obtain the average rotation data, and this average rotation data can be used as the rotation data of the rigid body model of the hand.

[0088] It is understandable that, in the above process, the rigid body model of the hand includes at least two first finger root nodes and at least one palm node. When the rigid body model of the hand includes at least one first finger root node and at least two palm nodes, the process of determining the true value calibration data is as follows:

[0089] Step 1: Obtain the ground truth data of the rigid body model of the hand when it performs a preset motion process through a gesture recognition ground truth device.

[0090] The truth data in step 1 includes the spatial location data of at least one first finger root node and at least two palm nodes.

[0091] Step 2: Determine the target spatial position data of the geometric center of the rigid body model of the hand based on at least one first finger root node and at least two palm nodes.

[0092] Step 3: Based on the target spatial location data and the rotation data of any hand node in the rigid body model of the hand, obtain the true value calibration data corresponding to the geometric center.

[0093] As a preferred example, two finger root nodes and one palm node can be selected to form a rigid body model of the hand, which can effectively reduce the computational complexity of determining the first true data, thereby helping to ensure testing efficiency.

[0094] S402, acquire test calibration data of the rigid body model of the hand when it performs a preset motion process through the gesture recognition device under test.

[0095] The test calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand.

[0096] In this embodiment, the pose data specifically refers to six-degree-of-freedom pose data. When acquiring test calibration data, the electronic device first obtains the test data included in the rigid body model of the hand through the gesture recognition device under test. Analogous to the aforementioned ground truth data, the test data includes the spatial position data of at least two first finger root nodes and at least one palm node, or the spatial position data of at least one first finger root node and at least two palm nodes. Specifically, the test data can be obtained by the electronic device after acquiring the test data of all hand nodes of the gesture recognition device under test and filtering it according to the rigid body model of the hand, or it can be obtained directly after the user configures the gesture recognition device under test to collect only the hand nodes included in the rigid body model of the hand. This embodiment does not limit this.

[0097] Secondly, the electronic device determines the target spatial data of the geometric center of the rigid body model of the hand based on the spatial position data of each hand node contained in the rigid body model of the hand, and further combines the rotation data of any hand node in the rigid body model of the hand to obtain the test calibration data corresponding to the geometric center.

[0098] For details on how to determine the target spatial data of the geometric center, please refer to the aforementioned embodiments, which will not be repeated here.

[0099] It is understandable that in practical applications, the true value data of any hand node in the rigid body model of the hand can be directly used as the true value calibration data and the test data as the test calibration data, as long as both the true value calibration data and the test calibration data contain six degrees of freedom pose data. This embodiment does not limit this.

[0100] S403 determines the true trajectory information based on the true calibration data and the test trajectory information based on the test calibration data.

[0101] The true trajectory information and the test trajectory information both include the pose data of the rigid body model of the hand at each time point during the preset motion process.

[0102] In this embodiment, both the truth calibration data and the test calibration data obtained by the electronic device are discrete data. The electronic device further obtains the truth trajectory information and the test trajectory information respectively through an interpolation algorithm.

[0103] S404, based on the true trajectory information and the test trajectory information, calculates the rotational modulus sequence.

[0104] S405, based on the rotated modulus sequence and cross-correlation algorithm, determines the maximum cross-correlation value and uses the time difference corresponding to the maximum cross-correlation value as the time transformation relationship.

[0105] In this embodiment, there are two rotation modulus sequences. Specifically, the electronic device first represents the true trajectory information and the test trajectory information in the same format, such as quaternion or rotation matrix format. Secondly, based on the true trajectory information and the test trajectory information, a rotation modulus sequence reflecting the rotation error of the test trajectory relative to the true trajectory, and a rotation modulus sequence reflecting the rotation error of the true trajectory relative to the test trajectory are generated. Further, using a cross-correlation algorithm, the two rotation modulus sequences are compared to find the optimal time difference, and this value is used as the time transformation relationship.

[0106] Understandably, to ensure the accuracy of the final time conversion relationship, S401 and S402 above need to be performed simultaneously.

[0107] Furthermore, in this embodiment, the process by which the electronic device determines the spatial transformation relationship between the gesture recognition device under test and the gesture recognition ground truth device includes:

[0108] The ground truth calibration data of the rigid body model of the hand is obtained by a gesture recognition ground truth device when the model performs a preset motion. The ground truth calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand, specifically six-degree-of-freedom pose data.

[0109] The test calibration data of the rigid body model of the hand is obtained by the gesture recognition device under test when the hand performs a preset motion process; the test calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand.

[0110] Based on the true value calibration data and the test calibration data, determine the spatial transformation relationship.

[0111] Specifically, the true calibration data for the rigid body model of the hand consists of its six-degree-of-freedom pose data at its geometric center. The determination process can be referred to in the aforementioned embodiments, and will not be repeated here. Similarly, the process for obtaining the test calibration data can also be referred to in the aforementioned embodiments, and will not be repeated here.

[0112] Furthermore, in this embodiment, the electronic device aligns the ground truth calibration data and the test calibration data to ensure that each ground truth point is correctly paired with its corresponding test point. Then, the spatial transformation relationship between the gesture recognition device under test and the ground truth gesture recognition device is determined according to the spatial transformation model. For example, the spatial transformation relationship between the gesture recognition device under test and the ground truth gesture recognition device is determined according to the Umeyama algorithm. Specifically, the spatial transformation relationship includes rotation matrices and translation vectors.

[0113] In the method provided in this embodiment, the electronic device determines the time and space transformation relationships between the gesture recognition device under test and the gesture recognition ground truth device based on the ground truth data and the true data of the hand nodes contained in the rigid body model of the hand. This effectively reduces the amount of data computation, thereby improving testing efficiency. At the same time, it reduces the computing power requirements of the electronic device.

[0114] This application also provides a method embodiment for illustrating how to determine test results based on hand truth data and hand test data. Figure 5 A flowchart illustrating a testing method for a gesture recognition device provided in this application is shown in Figure 3. Figure 5 As shown, the method in this embodiment includes:

[0115] S501 synchronizes hand test data to the time standard corresponding to the true hand data based on the time conversion relationship.

[0116] S502 synchronizes the true hand data to the spatial standard corresponding to the hand test data according to the spatial transformation relationship.

[0117] In this embodiment, after the electronic device obtains hand test data through the gesture recognition device under test and hand ground truth data through the gesture recognition ground truth device, it synchronizes the hand test data to the time standard corresponding to the hand ground truth data according to the time conversion relationship. Simultaneously, it synchronizes the hand ground truth data to the spatial standard corresponding to the hand test data according to the spatial conversion relationship. Based on this, temporal and spatial consistency between the hand ground truth data and the hand test data is achieved.

[0118] This setting enables electronic devices to perform time alignment and spatial alignment simultaneously, thereby further improving testing efficiency.

[0119] S503, for at least one hand node, the aligned ground truth data of the hand and the hand test data are compared to obtain the test results of the gesture recognition device under test for the hand node.

[0120] It is understandable that ground truth hand data includes the ground truth pose data of at least one hand node, and hand test data includes the test pose data of at least one hand node. Specifically, the electronic device receives the user's test request and acquires the ground truth pose data and test pose data of at least one hand node based on the test request, for use in testing at least one hand node. In actual testing, the number of hand nodes tested can be adjusted according to the accuracy evaluation requirements. It is understood that the more hand nodes tested, the higher the test evaluation accuracy of the gesture recognition device under test.

[0121] As one possible approach, for any hand node, the electronic device determines the ground truth trajectory corresponding to that hand node based on the aligned ground truth pose data. Then, based on the aligned test pose data, it determines the test trajectory corresponding to that hand node. Finally, based on the ground truth trajectory and the test trajectory of that hand node, the test result for that hand node is obtained.

[0122] In this embodiment, the hand node is any one of the aforementioned first finger root node, second finger root node, palm node, and other finger nodes. It is understood that since the target transformation relationship is determined based on the rigid body model of the hand, the test results obtained when testing the accuracy of the gesture recognition device under test for the hand nodes used to construct the rigid body model of the hand have the smallest error and higher reference value.

[0123] Specifically, as mentioned above, the number of hand nodes is related to user needs. Each hand node corresponds to one ground truth trajectory and one test trajectory, and also corresponds to one test result. For example, if a user inputs information into the electronic device to instruct on testing the recognition accuracy of a certain hand node of the gesture recognition device under test, then one test result corresponding to that hand node should be output. If a user inputs information into the electronic device to instruct on testing the recognition accuracy of two certain hand nodes of the gesture recognition device under test, then two test results corresponding to those two hand nodes should be output.

[0124] It is understandable that for each hand node, the electronic device can determine the test result by calculating the similarity between its corresponding ground truth trajectory and test trajectory. This embodiment does not limit this.

[0125] The method described in this embodiment enables the testing of the accuracy of the hand joint recognition device in recognizing each hand joint.

[0126] This application also provides a method embodiment for illustrating how to... Figure 2 The gesture recognition ground truth device shown acquires hand ground truth data. Figure 6 A flowchart illustrating a gesture recognition device testing method provided in this application embodiment. Figure 4 ,like Figure 6 As shown, the method in this embodiment includes:

[0127] S601 acquires RGB image data and depth data obtained from multiple RGB-D cameras capturing images of the same calibration board.

[0128] S602 performs camera calibration on multiple RGB-D cameras based on each RGB image and its corresponding depth data, obtaining multiple RGB-D data conversion relationships.

[0129] S603, based on the data conversion relationship, obtains RGBD data after RGB-D camera distortion correction and depth alignment.

[0130] S604 determines the alignment effect of RGBD data from each RGB-D camera. If the alignment effect does not meet the preset conditions, it optimizes the data conversion relationship.

[0131] Specifically, in this embodiment, for multiple RGB-D cameras, firstly, any one RGB-D camera is designated as the target RGB-D camera. Then, a calibration board is used for joint calibration to obtain the data conversion relationship between the other RGB-D cameras and the target RGB-D camera. Specifically, firstly, RGB image data and depth data obtained by capturing images of the calibration board from corresponding angles using multiple RGB-D cameras are acquired. Then, camera calibration is performed on the multiple RGB-D cameras to obtain the data conversion relationship between the other RGB-D cameras and the target RGB-D camera. Further, based on the data conversion relationship, RGBD data after distortion correction and depth alignment of the RGB-D cameras is obtained. The alignment effect of the RGBD data of each RGB-D camera is judged. If the alignment effect does not meet the standard, the data conversion relationship is optimized.

[0132] The data transformation relationships include both the transformation relationships between RGB and Depth cameras within each RGB-D camera, specifically involving intrinsic parameters and distortion coefficients, and the transformation relationships between different RGB-D cameras, specifically involving extrinsic parameters such as rotation matrices and translation vectors. Optimizing these data transformation relationships can employ techniques commonly used in the field, such as optimizing the quality of the acquired calibration data.

[0133] When the S605 receives a test request, it acquires RGB image data and depth data collected by multiple RGB-D cameras during the execution of a preset motion process on the rigid body model of the hand or during the hand's execution of test actions.

[0134] Understandably, the multiple RGB-D cameras at this point have already undergone the calibration process described above using S601-S603.

[0135] S606 processes RGB image data and depth data from multiple RGB-D cameras based on the data conversion relationship of multiple RGB-D cameras, and then fuses the processed RGB image data and depth data from multiple RGB-D cameras to obtain multiple sets of pose data including ground truth calibration data or hand ground truth data.

[0136] Specifically, the electronic device generates 2D hand ground truth data for each RGB-D camera based on RGB-D image data from multiple RGB-D cameras. Further, it optimizes the 3D hand ground truth data captured by each RGB-D camera based on the corresponding depth data. Finally, it processes the 3D hand ground truth data from different angles corresponding to multiple RGB-D cameras according to data transformation relationships. The processed 3D hand ground truth data is then optimized through multi-view fusion to obtain the final accurate hand ground truth data.

[0137] In the method provided in this embodiment, multiple RGB-D cameras are calibrated before being used to obtain the true hand data required during testing, which effectively ensures the accuracy of the true hand data.

[0138] The above embodiments describe a method for testing a gesture recognition device from the perspective of process flow. The following embodiments describe a device for testing a gesture recognition device from the perspective of virtual modules or virtual units. For details, please refer to the following embodiments.

[0139] This application provides a gesture recognition device testing apparatus, specifically... Figure 7 This is a schematic diagram of the structure of a gesture recognition device testing apparatus provided in an embodiment of this application. Figure 7 As shown, the device includes:

[0140] The acquisition module 71 is used to acquire hand test data of the gesture recognition device under test and hand ground truth data of the gesture recognition device based on the gesture recognition ground truth device and the gesture recognition device under test, which are in relatively fixed positions.

[0141] Processing module 72 is used to align the true hand data and the hand test data using a target transformation relationship. The target transformation relationship is determined by a preset motion process of the rigid body model of the hand. The rigid body model of the hand is constructed by at least two first finger root nodes and at least one palm node, or at least two palm nodes and at least one first finger root node. The relative positions of the first finger root nodes and the palm nodes are fixed when the hand is in different motion states.

[0142] The determination module 73 is used to determine the test results based on the aligned hand ground truth data and hand test data.

[0143] In one possible implementation, the processing module 72 is further configured to:

[0144] The ground truth calibration data of the rigid body model of the hand is obtained by a gesture recognition ground truth device when the model performs a preset motion. The ground truth calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand.

[0145] The test calibration data of the rigid body model of the hand is obtained by the gesture recognition device under test when the hand performs a preset motion process; the test calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand.

[0146] Based on the true value calibration data and the test calibration data, determine the spatial transformation relationship.

[0147] In one possible implementation, the processing module 72 is further configured to:

[0148] The ground truth calibration data of the rigid body model of the hand is obtained by a gesture recognition ground truth device when the model performs a preset motion. The ground truth calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand.

[0149] The test calibration data of the rigid body model of the hand is obtained by the gesture recognition device under test when the hand performs a preset motion process; the test calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand.

[0150] Based on the true value calibration data and the test calibration data, determine the spatial transformation relationship.

[0151] In one possible implementation, the processing module 72 is further configured to:

[0152] The true trajectory information is determined based on the true calibration data, and the test trajectory information is determined based on the test calibration data; both the true trajectory information and the test trajectory information include the pose data of the rigid body model of the hand at each time point in the preset motion process;

[0153] Based on the true trajectory information and the test trajectory information, the time transition relationship is determined.

[0154] In one possible implementation, both the truth calibration data and the test calibration data include rotation data; the processing module 72 is specifically used for:

[0155] Calculate the rotation modulus sequence based on the true trajectory information and the test trajectory information;

[0156] Based on the rotated modulus sequence and cross-correlation algorithm, the maximum cross-correlation value is determined, and the time difference corresponding to the maximum cross-correlation value is used as the time transformation relationship.

[0157] In one possible implementation, processing module 72 is specifically used for:

[0158] Determine the true calibration data and test calibration data for the geometric center of the rigid body model of the hand;

[0159] The true trajectory information is determined based on the true calibration data of the geometric center of the rigid body model of the hand, and the test trajectory information is determined based on the test calibration data of the geometric center of the rigid body model of the hand.

[0160] In one possible implementation, the hand ground truth data is acquired by a gesture recognition ground truth device when the hand performs a test action, and includes multiple sets of pose data corresponding to the hand.

[0161] Hand test data is acquired by the hand gesture recognition device under test when the hand performs test actions, and includes multiple sets of pose data corresponding to the hand.

[0162] In one possible implementation, the determining module 73 is specifically used for:

[0163] For at least one hand node, the aligned ground truth data of the hand and the hand test data are compared to obtain the test results of the gesture recognition device under test for the hand node.

[0164] In one possible implementation, the gesture recognition ground truth device includes multiple RGB-D cameras with fixed positions and different shooting angles; multiple sets of pose data are determined in the following manner:

[0165] Acquire RGB image data and depth data from multiple RGB-D cameras during the execution of a preset motion process on a rigid body model of a hand or when the hand is performing a test action;

[0166] Based on the data conversion relationship of multiple RGB-D cameras, the RGB image data and depth data of multiple RGB-D cameras are processed, and the processed RGB image data and depth data of multiple RGB-D cameras are fused to obtain ground truth calibration data or multiple sets of pose data including hand ground truth data.

[0167] In one possible implementation, the acquisition module 71 is also used for:

[0168] Acquire RGB image data and depth data obtained from multiple RGB-D cameras capturing the same calibration board;

[0169] Based on each RGB image and its corresponding depth data, camera calibration is performed on multiple RGB-D cameras to obtain multiple RGB-D data conversion relationships;

[0170] Based on the data conversion relationship, we obtain the RGBD data after RGB-D camera distortion correction and depth alignment;

[0171] Determine the alignment effect of the RGBD data of each RGB-D camera. If the alignment effect does not meet the preset conditions, optimize the data conversion relationship.

[0172] In one possible implementation, the first finger root node includes the index finger root node, the middle finger root node, the ring finger root node, and the little finger root node; the palm node includes the palm center node and the palm root node.

[0173] The gesture recognition device testing apparatus provided in this application embodiment is applicable to the above-described method embodiment, and will not be described again here.

[0174] This application provides an electronic device. Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 8As shown, the electronic device includes a processor 81 and a memory 82. The processor 81 and the memory 82 are connected, for example, via a bus 83. Optionally, the electronic device may also include a transceiver 84. It should be noted that in practical applications, the transceiver 84 is not limited to one type, and the structure of this electronic device does not constitute a limitation on the embodiments of this application.

[0175] Processor 81 may be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 81 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0176] Bus 83 may include a pathway for transmitting information between the aforementioned components. Bus 83 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Bus 83 may be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus 83 or one type of bus 83.

[0177] The memory 82 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0178] The memory 82 is used to store application code that executes the solution of this application, and its execution is controlled by the processor 81. The processor 81 is used to execute the application code stored in the memory 82 to implement the content shown in the foregoing method embodiments.

[0179] This application also provides a computer-readable storage medium, which may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. Specifically, the computer-readable storage medium stores program instructions, which are used to implement the methods in the above embodiments.

[0180] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the technical solution of the above method embodiments. Its implementation principle and technical effects are similar, and will not be repeated here.

[0181] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0182] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for testing a gesture recognition device, characterized in that, The method includes: Based on a relatively fixed-position gesture recognition ground truth device and a gesture recognition device under test, hand test data of the gesture recognition device under test and hand ground truth data of the gesture recognition ground truth device are obtained. The target transformation relationship is used to align the true hand data and the hand test data; the target transformation relationship is determined by a preset motion process of the rigid body model of the hand, which is constructed by at least two first finger root nodes and at least one palm node, or at least two palm nodes and at least one first finger root node; the relative positions of the first finger root nodes and the palm nodes are fixed when the hand is in different motion states. The test results are determined based on the aligned ground truth data of the hand and the hand test data.

2. The method according to claim 1, characterized in that, The method further includes: The gesture recognition ground truth device acquires ground truth calibration data of the rigid body model of the hand during the execution of a preset motion process; the ground truth calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand. The hand gesture recognition device under test acquires test calibration data of the rigid body model of the hand when it performs a preset motion process; the test calibration data includes multiple sets of pose data corresponding to the rigid body model of the hand. Based on the true value calibration data and the test calibration data, the spatial transformation relationship is determined.

3. The method according to claim 2, characterized in that, The method further includes: The true trajectory information is determined based on the true calibration data, and the test trajectory information is determined based on the test calibration data; both the true trajectory information and the test trajectory information include the pose data of the rigid body model of the hand at each time point during the preset motion process. Based on the true trajectory information and the test trajectory information, the time transformation relationship is determined.

4. The method according to claim 3, characterized in that, Both the truth calibration data and the test calibration data include rotation data; determining the time transformation relationship based on the truth trajectory information and the test trajectory information includes: Based on the true trajectory information and the test trajectory information, calculate the rotational modulus sequence; Based on the rotated modulus sequence and cross-correlation algorithm, the maximum cross-correlation value is determined, and the time difference corresponding to the maximum cross-correlation value is used as the time transformation relationship.

5. The method according to claim 3, characterized in that, The process of determining the true trajectory information based on the true calibration data and determining the test trajectory information based on the test calibration data includes: Determine the true calibration data and test calibration data of the geometric center of the rigid body model of the hand; The true trajectory information is determined based on the true calibration data of the geometric center of the rigid body model of the hand, and the test trajectory information is determined based on the test calibration data of the geometric center of the rigid body model of the hand.

6. The method according to claim 1, characterized in that, The hand truth data is acquired by the gesture recognition truth device when the hand performs a test action, and includes multiple sets of pose data corresponding to the hand. The hand test data is acquired by the hand gesture recognition device when the hand performs a test action, and includes multiple sets of pose data corresponding to the hand.

7. The method according to any one of claims 1-6, characterized in that, The determination of test results based on the aligned hand ground truth data and hand test data includes: For at least one hand node, the aligned ground truth data of the hand and the hand test data are compared to obtain the test result of the gesture recognition device under test for the hand node.

8. The method according to claim 2 or 6, characterized in that, The gesture recognition ground truth device includes multiple RGB-D cameras with fixed positions and different shooting angles; the multiple sets of pose data are determined in the following manner: Acquire RGB image data and depth data collected by the multiple RGB-D cameras during the execution of a preset motion process on the rigid body model of the hand or when the hand performs a test action; Based on the data conversion relationship of the multiple RGB-D cameras, the RGB image data and depth data of the multiple RGB-D cameras are processed, and the processed RGB image data and depth data of the multiple RGB-D cameras are fused to obtain the true value calibration data or the multiple sets of pose data included in the hand true value data.

9. The method according to claim 8, characterized in that, Before processing the RGB image data and depth data of the multiple RGB-D cameras based on the data conversion relationship of the multiple RGB-D cameras, the method further includes: Acquire RGB image data and depth data obtained by capturing images of the same calibration board using the multiple RGB-D cameras; Based on each RGB image and its corresponding depth data, camera calibration is performed on the plurality of RGB-D cameras to obtain multiple RGB-D data conversion relationships; Based on the data conversion relationship, the RGBD data after RGB-D camera distortion correction and depth alignment is obtained; Determine the alignment effect of the RGBD data of each RGB-D camera. If the alignment effect does not meet the preset conditions, optimize the data conversion relationship.

10. The method according to any one of claims 1-6, characterized in that, The first finger root node includes the index finger root node, the middle finger root node, the ring finger root node, and the little finger root node; the palm node includes the palm center node and the palm root node.

11. A testing device for a gesture recognition equipment, characterized in that, The device includes: The acquisition module is used to acquire hand test data of the gesture recognition device under test and hand ground truth data of the gesture recognition device under test, based on the gesture recognition ground truth device and the gesture recognition device under test, which are in relatively fixed positions. The processing module is used to align the true hand data and the hand test data using a target transformation relationship; the target transformation relationship is determined by a preset motion process of the rigid body model of the hand, which is constructed from at least two first finger root nodes and at least one palm node, or at least two palm nodes and at least one first finger root node; the relative positions of the first finger root nodes and the palm nodes are fixed when the hand is in different motion states. The determination module is used to determine the test results based on the aligned hand ground truth data and hand test data.

12. An electronic device, characterized in that, Includes a processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-10.

14. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1-10.

15. A gesture recognition device testing system, characterized in that, The system includes the electronic device as described in claim 12 and the gesture recognition truth value device, wherein the electronic device acquires hand truth value data through the gesture recognition truth value device.

16. The system according to claim 15, characterized in that, The gesture recognition ground truth device includes multiple RGB-D cameras with fixed positions and different shooting angles.

Images