Data glove based on fiber-optic bend sensor and pose reconstruction method

Abstract

This disclosure provides a data glove based on fiber optic bend sensors and a pose reconstruction method. The data glove includes: multiple fiber optic bend sensors, each disposed in the joint region of each finger of the glove, with each fiber optic bend sensor bending synchronously with the corresponding finger joint; a data acquisition module that acquires the electrical signals output by the fiber optic bend sensors and converts them into input samples; and a pose reconstruction module that reconstructs the fingertip pose based on the input samples. The pose reconstruction module includes a Mamba model and a rectified flow model. The Mamba model reconstructs a coarse prediction result of the fingertip pose based on the input samples, and the rectified flow model generates a residual estimate based on the coarse prediction result through numerical integration iteration. This achieves high-precision reconstruction of the six-degree-of-freedom pose of the operator's fingertips, making it particularly suitable for virtual reality, medical rehabilitation, and deep-sea exploration.

Description

Technical Field

[0001] This disclosure relates to the fields of wearable sensing, hand motion measurement and robot teleoperation technology, and specifically to a data glove based on an optical fiber bend sensor and a pose reconstruction method. Background Technology

[0002] In scenarios such as medical procedures, virtual reality interaction, teleoperated robots, aerospace operations, deep-sea exploration, and motion analysis, the spatial pose information of the operator's hand, especially the fingertips, is a crucial foundation for achieving precise interaction and stable control. Finger-end pose typically includes three-dimensional position and three-dimensional orientation; the measurement accuracy and stability of finger-end pose directly affect the system's control performance and interactive experience. Existing methods for acquiring hand pose mainly include vision-based approaches and wearable sensor-based approaches.

[0003] Vision-based methods typically acquire images using monocular, binocular, depth, or multi-view camera systems, and obtain hand pose parameters through hand keypoint detection, 3D reconstruction, or pose estimation algorithms. While vision-based methods do not require direct sensor attachment, they are often susceptible to factors such as changes in lighting, background interference, viewing angle changes, and hand self-occlusion, resulting in insufficient robustness and stability in complex environments.

[0004] Wearable sensor-based methods typically detect finger joint flexion or muscle activity by embedding inertial measurement units (IMUs), flexible resistive sensors, flexible capacitive sensors, pressure sensors, or electromyography (EMG) sensors on gloves. Wearable sensor-based methods are less dependent on the external environment and offer good portability and real-time performance. However, IMUs often suffer from integration drift and sensitivity to magnetic environments; flexible resistive or capacitive sensors may exhibit hysteresis, aging, and insufficient repeatability, affecting long-term stable use.

[0005] Fiber optic sensing technology boasts advantages such as strong resistance to electromagnetic interference, good flexibility, and suitability for wearable integration, and has been increasingly applied in deformation detection and attitude sensing in recent years. While fiber Bragg grating-based schemes can achieve high-precision strain measurement, their demodulation equipment is costly and the system structure is complex. Intensity-modulated bending detection schemes, although lower in cost, often exhibit strong nonlinear and temporal coupling between the sensing signal and the end-effector pose in complex dynamic motion scenarios of finger joints, making it difficult to obtain high-precision end-effector pose estimation results using simple calibration methods alone.

[0006] Therefore, there is an urgent need for a hand posture measurement device and method that combines high sensitivity, strong anti-interference ability, good wearability, and the ability to stably map multi-channel bending sensing signals into fingertip postures. Summary of the Invention

[0007] To address the aforementioned technical problems, this disclosure provides a fingertip pose reconstruction data glove and method based on flexible fiber optic sensing with active light loss characteristics, which at least partially solves the following problems existing in the prior art: visual solutions are susceptible to occlusion and lighting conditions, wearable inertial or resistive sensing solutions suffer from drift or hysteresis, and low-cost bending sensing signals are difficult to accurately reconstruct fingertip pose.

[0008] In one general aspect, a data glove based on an optical fiber bend sensor is provided. The data glove includes: multiple optical fiber bend sensors respectively disposed in each joint region of each finger of the data glove, each optical fiber bend sensor being synchronously bent with the corresponding finger joint; multiple data acquisition modules respectively acquiring the electrical signals output by the optical fiber bend sensors and converting them into input samples; and a pose reconstruction module reconstructing the fingertip pose based on the input samples. The pose reconstruction module includes a Mamba model and a rectified flow model. The Mamba model reconstructs a coarse prediction result of the fingertip pose based on the input samples, and the rectified flow model generates a residual estimate based on the coarse prediction result through numerical integration iteration. The coarse prediction result and the residual estimate constitute the fingertip pose.

[0009] According to an embodiment, each of the plurality of fiber optic bend sensors includes at least one flexible fiber with optical loss characteristics, the flexible fiber being arranged in the corresponding joint region along the finger extension direction.

[0010] According to an embodiment, the optical loss feature structure includes at least one of the following: a plurality of microcrack structures arranged axially side by side along the outer surface of the optical fiber, a deformation-based transmittance shielding material filling the interior of the optical fiber, and a rough texture on the surface of the optical fiber.

[0011] According to an embodiment, the rectified flow model generates residual estimates by numerical integration starting from the zero vector based on the coarse prediction results and their temporal characteristics.

[0012] In another general aspect, a pose reconstruction method for a data glove based on fiber optic bend sensors is provided. The pose reconstruction method includes: acquiring photoelectric signals through multiple fiber optic bend sensors respectively disposed in the joint region of each finger of the data glove, each fiber optic bend sensor being synchronously bent with the corresponding finger joint; acquiring the electrical signals output by the fiber optic bend sensors and converting them into input samples; reconstructing the fingertip pose based on the input samples through a pose reconstruction module, wherein the pose reconstruction module includes a Mamba model and a rectified flow model, the Mamba model reconstructing a coarse prediction result of the fingertip pose based on the input samples, and the rectified flow model generating a residual estimate based on the coarse prediction result through numerical integration iteration, the coarse prediction result and the residual estimate constituting the fingertip pose.

[0013] According to an embodiment, the rectified flow model generates residual estimates by numerical integration starting from the zero vector based on the coarse prediction results and their temporal characteristics.

[0014] According to an embodiment, the Mamba model is trained through the following steps: the input sample obtained by the data acquisition module from the electrical signal output of the fiber optic bending sensor is used as the sample data of the Mamba model; the pose data of the reflective points of the marking components set on the back of the hand and fingertips by optical capture is used as the label data of the Mamba model; the sample data and label data are input into the Mamba model, the loss value is determined by a predetermined loss function and the loss value is controlled by gradient clipping; and the parameters of the Mamba model are adjusted based on the loss value after gradient clipping to improve the detection accuracy.

[0015] According to an embodiment, the rectified flow model is trained through the following steps: with the parameters of the Mamba model fixed, the coarse prediction result calculated by the rectified flow model is used as the sample data of the rectified flow model, and the residual between the coarse prediction result of the Mamba model and the label data of the Mamba model is used as the label data of the rectified flow model; based on the sample data of the rectified flow model, the temporal characteristics of the coarse prediction result are used to guide the rectified flow model to learn the velocity field from a simple Gaussian distribution to the true residual distribution to generate a residual estimate, and based on the label data of the rectified flow model and the generated residual estimate, a loss value is determined through a predetermined loss function; and based on the loss value, the parameters of the rectified flow model are adjusted to improve the detection accuracy.

[0016] In another general aspect, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the pose reconstruction method for a data glove based on a fiber optic bend sensor as described above.

[0017] In another general aspect, a computer device is provided, the computer device comprising: a processor; and a memory storing a computer program that, when executed by the processor, implements the pose reconstruction method for a data glove based on a fiber optic bend sensor as described above.

[0018] The data glove according to the embodiments of this disclosure integrates microstructure fiber optic sensors and deep learning models to achieve high-precision reconstruction of the six-degree-of-freedom pose of the operator's fingertips. It has robustness in complex environments, long-term stability, optimized cost and system structure complexity, and can obtain high-precision end-effector pose estimation. It is especially suitable for scenarios that require high-precision and high-stability hand posture perception, such as virtual reality, medical rehabilitation, and aerospace deep-sea exploration. Attached Figure Description

[0019] The above and other aspects, features and advantages of this disclosure will become clearer from the following detailed embodiments, taken in conjunction with the accompanying drawings, in which: Figure 1 This is a schematic diagram of a data glove according to an embodiment of the present disclosure.

[0020] Figure 2 This is a schematic diagram of an optical fiber bend sensor according to an embodiment of the present disclosure.

[0021] Figure 3 This is a block diagram of a data glove according to an embodiment of the present disclosure.

[0022] Figure 4 This is a schematic diagram of a data glove for collecting tag data according to an embodiment of the present disclosure.

[0023] Figure 5 This is a flowchart of a pose reconstruction method according to an embodiment of the present disclosure.

[0024] Figure 6 This is a block diagram of a pose reconstruction module according to an embodiment of the present disclosure.

[0025] Explanation of reference numerals in the attached figures 100. Fiber optic bending sensor; 110. Flexible optical fiber; 111. Active optical loss characteristic structure; 120. Infrared light source; 130. Photodetector; 140. Substrate; 200+ data acquisition modules; 300. Pose Reconstruction Module; 400. Glove body; 410. Rigid body on the back of the hand; 420. Fingertip Rigid Body; 430. Reflective markers. Detailed Implementation

[0026] The following detailed embodiments are provided to aid the reader in gaining a comprehensive understanding of the methods, apparatus, and / or systems described herein. However, after understanding the disclosure of this application, various changes, modifications, and equivalents of the methods, apparatus, and / or systems described herein will be readily apparent. For example, the order of operations described herein is merely illustrative and is not limited to the order set forth herein; rather, changes that will be readily understood after understanding the disclosure of this application are possible, except for operations that must occur in a specific order. Furthermore, for clarity and brevity, descriptions of features known after understanding the disclosure of this application may be omitted.

[0027] The features described herein may be implemented in different forms and should not be construed as limited to the examples described herein. Rather, the examples described herein are provided only to illustrate some of the many feasible ways of implementing the methods, apparatus and / or systems described herein, many of which will become clear upon understanding the disclosure of this application.

[0028] As used herein, the term “and / or” includes any one of the associated listed items and any combination of any two or more.

[0029] Although terms such as “first,” “second,” and “third” may be used herein to describe various components, assemblies, regions, layers, or parts, these components, assemblies, regions, layers, or parts should not be limited by these terms. Rather, these terms are used only to distinguish one component, assembly, region, layer, or part from another. Thus, without departing from the teaching of the examples described herein, the first component, first assembly, first region, first layer, or first part referred to as the first component, first assembly, first region, first layer, or first part may also be referred to as the second component, second assembly, second region, second layer, or second part.

[0030] In the specification, when an element (such as a layer, region, or substrate) is described as being "on" another element, "connected to," or "bonded to" another element, the element may be directly "on" another element, directly "connected to," or "bonded to" the other element, or one or more other elements may be present in between. Conversely, when an element is described as being "directly on" another element, "directly connected to," or "directly bonded to" another element, no other elements may be present in between.

[0031] The terminology used herein is for the purpose of describing various examples only and is not intended to limit disclosure. Unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. The terms “comprising,” “including,” and “having” indicate the presence of the described features, quantities, operations, components, elements, and / or combinations thereof, but do not preclude the presence or addition of one or more other features, quantities, operations, components, elements, and / or combinations thereof.

[0032] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains upon understanding this disclosure. Unless expressly defined herein, terms (such as those defined in a general dictionary) shall be interpreted as having a meaning consistent with their meaning in the context of the relevant field and in this disclosure, and shall not be interpreted in an idealized or overly formalistic manner.

[0033] Furthermore, in the description of the examples, detailed descriptions of well-known related structures or functions will be omitted when it is believed that such detailed descriptions would lead to a vague interpretation of this disclosure.

[0034] To enable those skilled in the art to better understand this disclosure, specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings.

[0035] Figure 1 This is a schematic diagram of a data glove according to an embodiment of the present disclosure. Figure 2 This is a schematic diagram of an optical fiber bend sensor according to an embodiment of the present disclosure. Figure 3 This is a block diagram of a data glove according to an embodiment of the present disclosure.

[0036] Reference Figures 1 to 3 According to an embodiment of this disclosure, a data glove based on fiber optic bend sensors includes: a plurality of fiber optic bend sensors 100, respectively disposed in each joint region of each finger of the data glove, each fiber optic bend sensor bending synchronously with the corresponding finger joint; a plurality of data acquisition modules 200, which respectively acquire the electrical signals output by the plurality of fiber optic bend sensors and convert them into input samples; and a pose reconstruction module 300, which reconstructs the fingertip pose based on the input samples. The plurality of fiber optic bend sensors 100, the plurality of data acquisition modules 200, and the pose reconstruction module 300 are integrated on the glove body 400. The pose reconstruction module may include a Mamba model and a rectified flow model, wherein the Mamba model reconstructs a coarse prediction result of the fingertip pose based on the input samples, and the rectified flow model generates a residual estimate based on the coarse prediction result through numerical integration iteration, and the coarse prediction result and the residual estimate constitute the fingertip pose.

[0037] The fiber optic bend sensor 100 is used to sense the degree of bending of the finger joints. When the finger moves, the flexible fiber 110 deforms, and the active light loss feature structure 111 on the flexible fiber 110 causes optical signal leakage or changes in transmission efficiency, which are then captured as changes in voltage signals by the photodetector 130.

[0038] According to one embodiment, each of the plurality of fiber optic bend sensors 100 includes at least one flexible fiber with optical loss characteristics, the flexible fiber being arranged in the corresponding joint region along the finger extension direction.

[0039] According to one embodiment, the optical loss feature structure may include at least one of the following: a plurality of microcrack structures arranged axially side by side along the outer surface of the optical fiber, a deformation-based transmittance shielding material filling the interior of the optical fiber, and a rough texture on the surface of the optical fiber; and the optical loss feature structure may include a combination of two or more of the above structures.

[0040] Reference Figure 2 This disclosure uses a micro-crack structure as an example of an optical loss characteristic structure. A flexible optical fiber 110 (e.g., TPU material with a Shore A hardness of 80) is processed into a strip, and multiple micro-crack structures are cut into designated areas on its surface using a custom mold to form a bending-sensitive region. One end of the processed optical fiber is aligned and fixed to an infrared light source 120 (e.g., an infrared LED), and the other end is aligned and fixed to a photodetector 130 (e.g., a photodiode). Both the infrared light source 120 and the photodetector 130 are mounted on a substrate 140. Encapsulating adhesive is used to fix and protect both ends of the optical fiber. When the optical fiber is straight, light propagates through total internal reflection, resulting in higher light intensity at the end. When the optical fiber bends, the micro-cracks on the outside open, causing light leakage and a decrease in light intensity at the end, thereby modulating the bending angle information into a voltage signal. It is understood that if other active optical loss characteristic structures (such as internally filled obstructions) are used, their working principle is similar; bending causes changes in the distribution of obstructions or the interface characteristics between the optical fiber and the obstruction, thereby causing optical loss.

[0041] Multiple fabricated fiber optic bend sensors 100 are respectively fixed to each joint area of ​​each finger of the glove body 400. As an example, in... Figure 1 The middle finger is positioned on the dorsal side of the three joints to ensure that the fiber optic cable can effectively deform when the joint is bent.

[0042] In practical use, the pose reconstruction module 300 calculates the precise pose of the fingertip based on real-time acquired sensor signals. The pose reconstruction module 300 includes a deep learning model. According to embodiments of this disclosure, the deep learning model includes a coarse prediction unit based on the Mamba model and a residual correction unit based on flow matching.

[0043] Figure 4 This is a schematic diagram of a data glove for collecting tag data according to an embodiment of the present disclosure. Figure 5 This is a flowchart of a pose reconstruction method according to an embodiment of the present disclosure. Figure 6 This is a block diagram of a pose reconstruction module according to an embodiment of the present disclosure.

[0044] Reference Figures 4 to 6 The pose reconstruction method for a data glove based on a fiber optic bend sensor according to embodiments of this disclosure includes: In step 101, photoelectric signals are acquired by multiple fiber optic bend sensors respectively installed in the joint region of each finger of the data glove, with each fiber optic bend sensor bending synchronously with the corresponding finger joint. In step 102, the electrical signals output by the fiber optic bend sensors are acquired and converted into input samples. In step 103, the fingertip pose is reconstructed based on the input samples using a pose reconstruction module. The pose reconstruction module may include a Mamba model and a rectified flow model, wherein the Mamba model reconstructs a coarse prediction result of the fingertip pose based on the input samples, and the rectified flow model generates a residual estimate based on the coarse prediction result through numerical integration iteration; the coarse prediction result and the residual estimate constitute the fingertip pose.

[0045] The embodiments of this disclosure can perform data preprocessing on the raw data, including timestamp alignment, outlier removal, smoothing filtering, and feature normalization. Next, sample data X for model input is constructed. Assuming a time window length of T (e.g., T = 96), each sample data X consists of observations from T consecutive frames, with each frame of sample data X being input... It is composed of three parts: the 3D position of the rigid body on the back of the hand. 6-dimensional continuous rotation representation of the rigid body on the back of the hand (Obtained by expanding the first and second columns of the rotation matrix), and the 4-channel normalized fiber optic sensor signals. .Right now The supervision label y represents the 9-dimensional pose of the fingertip rigid body in the last frame of the time window (3-dimensional position + 6-dimensional rotation representation), i.e. .

[0046] Reference Figure 6 In the model inference phase, the input sequence of sample data X first passes through a coarse predictor. The core of the coarse predictor is a state-space model based on Mamba, used to learn the main mapping relationship from historical observation sequences to fingertip poses. This model efficiently models long-term time-series data and captures the dynamic features of hand movements. The Mamba model outputs a global temporal feature vector h, which is then passed through a multilayer perceptron regression head to obtain the coarse prediction result of the fingertip pose. 0.

[0047] The input sequence of the sample data can include 3D position of the back of the hand, 6D pose of the back of the hand, and 4D sensor signal, for a total of 13 dimensions. In the coarse predictor of the Mamba model, the input sequence is transmitted to the input encoder, and after passing through 6 Mamba blocks (32×6), a 9-dimensional coarse prediction vector of the fingertip's 3D position and 6D pose is obtained.

[0048] Although coarse predictors based on the Mamba model can learn the main temporal mapping relationships, they are still affected by sensor noise, nonlinear coupling errors, and local dynamic disturbances under complex motion conditions, resulting in residuals. To further improve prediction accuracy, embodiments of this disclosure introduce a conditional rectified flow model on the basis of coarse prediction to model the residual distribution.

[0049] In the generative residual correction block of the Rectified Flow model, the coarse prediction vector is input to the Flow Matching module, where it is sinusoidally position-encoded in chronological order. Through six radial coupling blocks, the output is the residual prediction, which includes a 3D position and a 6D pose vector of the fingertip (9-dimensional vector). The residual vector is then added to the coarse prediction vector to obtain the final prediction of the fingertip pose.

[0050] According to one embodiment, the rectified flow model generates residual estimates by numerical integration starting from the zero vector based on the coarse prediction results and their temporal characteristics.

[0051] Specifically, the generative residual correction block can be a conditional rectified flow model. It provides coarse predictions. 0 and temporal features h are used as conditional information c to calculate the residual r = y - y between the coarse prediction and the true label y. Modeling begins with zero. During inference, starting from the zero vector and guided by the conditional information c, the residual estimate r is generated iteratively through numerical integration (such as the fourth-order Runge-Kutta method). Finally, the coarse prediction and the residual estimate are added together to obtain the final high-precision fingertip pose prediction. = 0+α·r , where α is a coefficient for adjusting the residual strength.

[0052] According to one embodiment, the Mamba model is trained through the following steps: the input sample obtained by the data acquisition module from the electrical signal output of the fiber optic bending sensor is used as the sample data of the Mamba model; the pose data of the reflective points of the marking components set on the back of the hand and fingertips by optical capture is used as the label data of the Mamba model; the sample data and label data are input into the Mamba model, the loss value is determined by a predetermined loss function and the loss value is controlled by gradient clipping; and the parameters of the Mamba model are adjusted based on the loss value after gradient clipping to improve the detection accuracy.

[0053] According to one embodiment, the rectified flow model is trained through the following steps: with the parameters of the Mamba model fixed, the coarse prediction branch is stabilized first; the coarse prediction result calculated by the rectified flow model is used as the sample data of the rectified flow model; and the residual between the coarse prediction result of the Mamba model and the label data of the Mamba model is used as the label data of the rectified flow model. Based on the sample data of the rectified flow model, the temporal characteristics of the coarse prediction result guide the rectified flow model to learn the velocity field from a simple Gaussian distribution to the true residual distribution to generate a residual estimate; and based on the label data of the rectified flow model and the generated residual estimate, a loss value is determined through a predetermined loss function. Finally, based on the loss value, the parameters of the rectified flow model are adjusted to improve detection accuracy.

[0054] Specifically, during the training phase, to obtain high-precision label data for model training, rigid bodies 410 on the back of the hand and 420 on the fingertips are set at the back of the hand and fingertips, respectively. Each rigid body has four non-collinear optical reflective markers 430 at its four corners. The actual hand pose is acquired as label data through an optical motion capture subsystem (including a motion capture camera, the rigid bodies 410 and 420 fixed to the back of the hand and the reflective markers 430). By calculating the spatial position of these four points, the precise six-DOF pose of the rigid body center can be obtained. Voltage signals from fiber optic sensors are simultaneously acquired and converted into sample data. The sample data and label data are strictly aligned in time and together constitute the training dataset. During the training of the rectified flow model, the rectified flow model learns a velocity field that transitions from a simple Gaussian distribution to the true residual distribution.

[0055] By employing the two-stage strategy of coarse prediction combined with fine residual correction, the data glove according to the embodiments of this disclosure can stably provide high-precision end-effector pose output during dynamic motion.

[0056] The technical solutions of this disclosure have been described with reference to the preferred embodiments shown in the accompanying drawings. It will be readily understood by those skilled in the art that the scope of protection of this disclosure is obviously not limited to these specific embodiments. Without departing from the principles of this disclosure, those skilled in the art can make equivalent changes or substitutions to the relevant technical features. For example, replacing the Mamba model with other advanced timing models, or replacing the rectified flow model with other generation models, will all fall within the scope of protection of this disclosure.

[0057] The data glove and pose reconstruction method based on fiber optic bend sensor according to the embodiments of this disclosure achieves high-sensitivity and electromagnetic interference-resistant bend measurement through microstructured optical fiber; combined with multimodal data and advanced algorithms, the end-effector position estimation accuracy reaches the millimeter level; providing a low-constraint and high-stability hand posture measurement solution for fields such as virtual reality, medical rehabilitation and special robot teleoperation.

[0058] The embodiments of this disclosure convert the deformation caused by finger joint bending into detectable changes in light intensity by setting active light loss feature structures on flexible optical fibers, thereby achieving sensitive detection of joint bending states. Compared to traditional micro-slit solutions, various light loss feature structures, including obstruction filling and surface textures, provide more optimization space for sensor design and manufacturing to adapt to different sensitivity requirements and cost control. Furthermore, flexible optical fiber sensing has strong anti-electromagnetic interference capabilities, which is beneficial to improving the stability of use in complex environments.

[0059] By simultaneously acquiring photoelectric signals and reference pose data, calibration samples of "sensor signals - end-effector pose" can be constructed, improving the reliability of pose reconstruction model training. Through a combination of temporal feature extraction, coarse prediction, and residual correction, it exhibits robustness in complex environments and long-term stability, optimizing cost and system structure complexity, and achieving high-precision end-effector pose estimation. It is suitable for single-finger end-effector pose measurement and can also be applied to multi-finger or full-hand pose reconstruction scenarios by adding sensor channels and expanding the model structure.

[0060] An exemplary embodiment of the present disclosure also provides a computer-readable storage medium storing a computer program. The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to execute the pose reconstruction method for a data glove based on a fiber optic bend sensor according to the present disclosure. The computer-readable recording medium is any data storage device capable of storing data readable by a computer system. Examples of computer-readable recording media include: read-only memory, random access memory, read-only optical disk, magnetic tape, floppy disk, optical data storage device, and carrier waves (such as data transmission via the Internet through wired or wireless transmission paths).

[0061] An exemplary embodiment of the present disclosure also provides a computer device. The computer device includes a processor and a memory. The memory stores a computer program. The computer program is executed by the processor, causing the processor to execute the computer program of the pose reconstruction method for a data glove based on a fiber optic bend sensor according to the present disclosure.

[0062] Therefore, the exemplary embodiments of this disclosure can be implemented as methods in a computer or a non-transitory computer-readable medium storing computer-executable instructions. In the exemplary embodiments, when executed by a processor, the computer-readable instructions can perform a method according to at least one aspect of this disclosure.

[0063] Furthermore, the methods according to exemplary embodiments of this disclosure can be implemented in the form of program instructions that can be executed by various computer devices and recorded on a computer-readable medium.

[0064] Computer-readable media may include program instructions, data files, data structures, etc., individually or in combination. Program instructions recorded on a computer-readable medium may be specifically designed and configured for the exemplary embodiments of this disclosure, or may be known and available to those skilled in the art of computer software. Computer-readable recording media may include hardware devices configured to store and execute program instructions. For example, computer-readable recording media may be or include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical media such as CD-ROMs and DVDs; magneto-optical media such as optical-magnetic disks; ROM; RAM; flash memory; etc. Program instructions may include not only machine language code generated by a compiler, but also high-level language code executable by a computer through an interpreter, etc.

[0065] While this disclosure includes specific examples, it will be apparent to those skilled in the art that various changes in form and detail may be made to these examples without departing from the spirit and scope of the claims and their equivalents. The examples described herein are to be considered descriptive only and not for limiting purposes. The description of features or aspects in each example is to be considered applicable to similar features or aspects in other examples. Suitable results may be obtained if the described techniques are performed in a different order, and / or if components in the described system, architecture, apparatus, or circuit are combined in a different manner and / or if components in the described system, architecture, apparatus, or circuit are replaced or supplemented by other components or their equivalents. Therefore, the scope of this disclosure is not limited by the specific embodiments but by the claims and their equivalents, and all variations within the scope of the claims and their equivalents shall be construed as included in this disclosure.

Claims

1. A data glove based on fiber-optic bend sensor, characterized in that, The data gloves include: Multiple fiber optic bend sensors are respectively installed in each joint area of ​​each finger of the data glove, and each fiber optic bend sensor bends synchronously with the corresponding finger joint; Multiple data acquisition modules respectively acquire the electrical signals output by the multiple fiber optic bending sensors and convert them into input samples; The pose reconstruction module reconstructs the fingertip pose based on the input samples. The pose reconstruction module includes a Mamba model and a rectified flow model. The Mamba model reconstructs a coarse prediction of the fingertip pose based on the input samples. The rectified flow model generates a residual estimate based on the coarse prediction result through numerical integration iteration. The coarse prediction result and the residual estimate together constitute the fingertip pose.

2. The data glove according to claim 1, characterized in that, Each of the plurality of fiber optic bend sensors includes at least one flexible fiber with optical loss characteristics, the flexible fiber being arranged in the corresponding joint region along the finger extension direction.

3. The data glove according to claim 2, characterized in that, The optical loss feature structure includes at least one of the following: multiple microcrack structures arranged axially side by side along the outer surface of the optical fiber, a deformation-based transmittance shielding material filling the interior of the optical fiber, and a rough texture on the surface of the optical fiber.

4. The data glove according to claim 1, characterized in that, The rectified flow model generates residual estimates by numerical integration starting from the zero vector, based on the coarse prediction results and their temporal characteristics.

5. A method for pose reconstruction of a data glove based on an optical fiber bend sensor, characterized in that, The pose reconstruction method includes: Photoelectric signals are acquired by multiple fiber optic bend sensors respectively set in each joint area of ​​each finger of the data glove, with each fiber optic bend sensor bending synchronously with the corresponding finger joint; The electrical signal output from the fiber optic bend sensor is acquired and converted into an input sample; The fingertip pose is reconstructed using a pose reconstruction module based on the input samples. The pose reconstruction module includes a Mamba model and a rectified flow model. The Mamba model reconstructs a coarse prediction of the fingertip pose based on the input samples. The rectified flow model generates a residual estimate based on the coarse prediction result through numerical integration iteration. The coarse prediction result and the residual estimate together constitute the fingertip pose.

6. The pose reconstruction method according to claim 5, characterized in that, The rectified flow model generates residual estimates by numerical integration starting from the zero vector, based on the coarse prediction results and their temporal characteristics.

7. The pose reconstruction method according to claim 6, characterized in that, The Mamba model is trained through the following steps: The input sample obtained by the data acquisition module from the electrical signal output of the fiber optic bending sensor is used as the sample data of the Mamba model, and the pose data of the reflective points of the marking components set on the back of the hand and fingertips by optical capture is used as the label data of the Mamba model. The sample data and label data are input into the Mamba model, the loss value is determined by a predetermined loss function, and the loss value is controlled by gradient clipping; and Based on the loss value after gradient clipping, the parameters of the Mamba model are adjusted to improve detection accuracy.

8. The pose reconstruction method according to claim 7, characterized in that, The rectified flow model is trained through the following steps: With the parameters of the Mamba model fixed, the coarse prediction result calculated by the rectified flow model is used as the sample data of the rectified flow model, and the residual between the coarse prediction result of the Mamba model and the label data of the Mamba model is used as the label data of the rectified flow model. Based on the sample data of the rectified flow model, the temporal characteristics of the coarse prediction results guide the rectified flow model to learn the velocity field from a simple Gaussian distribution to the true residual distribution to generate residual estimates. Based on the labeled data of the rectified flow model and the generated residual estimates, the loss value is determined by a predetermined loss function. as well as Based on the loss value, the parameters of the rectified flow model are adjusted to improve detection accuracy.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the pose reconstruction method for a data glove based on a fiber optic bend sensor as described in any one of claims 5 to 8.

10. A computer device, characterized in that, The computer device includes: processor; The memory stores a computer program that, when executed by a processor, implements the pose reconstruction method for a data glove based on a fiber optic bend sensor as described in any one of claims 5 to 8.

Images