A hand pose calculation method and system based on electromagnetic positioning

Abstract

This invention discloses a hand pose calculation method and system based on electromagnetic positioning. The method includes: an electromagnetic transmitting module transmitting a continuous rotating excitation magnetic field through an orthogonal coil group to form a rotating magnetic dipole elliptical polarization field; magnetic induction receiving sensors arranged at multiple points on the hand continuously acquiring time-domain signals within a single excitation cycle; extracting elliptical polarization field parameters through phase-locked detection, and simultaneously calculating the three-dimensional position and attitude of each sensor within a single cycle by combining analytical mapping relationships; uniformly transforming the poses of each sensor to a hand back reference frame and, after extrinsic parameter compensation, inputting them into an inverse kinematics model to calculate the multi-degree-of-freedom joint angles of the hand. The system includes an electromagnetic transmitting module, multiple magnetic induction receiving sensors, a unified processing module, and a host computer processing module. The technical solution provided by this invention achieves single-cycle 6-DOF positioning with a position error <2mm, stability <0.2mm, real-time performance ≥70Hz, and supports continuous tracking of 22 degrees of freedom of the hand.

Description

Technical Field

[0001] This invention relates to the field of motion capture and human-computer interaction technology, specifically to a dynamic multi-degree-of-freedom hand pose calculation method and device based on near-field electromagnetic positioning, which is applicable to real-time tracking of hand poses in high-precision motion capture scenarios such as virtual reality / augmented reality, game / film production, precision medicine, and teaching. Background Technology

[0002] In high-precision motion capture scenarios such as VR / AR, gaming, film and television production, and precision medicine, wearable gesture capture devices, such as data gloves, are typically used to capture every minute movement of the hand in real time and without obstruction. The hand has extremely high degrees of freedom of movement, including multiple phalanges, metacarpals, and various complex joints.

[0003] Existing hand motion capture technologies each have their drawbacks. Specifically: Inertial navigation (IMU) solutions suffer from cumulative drift issues and experience accuracy degradation over long-term use, requiring frequent calibration and failing to meet the demands for high-precision, long-term continuous capture. Optical tracking solutions are primarily costly, prone to tracking failure when visually obstructed, and sensitive to ambient lighting, making them unsuitable for scenarios where hands frequently self-occlude. Exoskeleton solutions typically have complex mechanical structures, poor wearability, and hinder natural hand movements, making them unsuitable for VR / AR interactions requiring natural tactile feedback.

[0004] For electromagnetic positioning solutions, representative products in the current electromagnetic positioning motion capture field include: the Metagloves series from XsensTechnology Group BV (Netherlands), the 3D Guidance series from NDI (Northern Digital Inc.) (Canada), and the LIBERTY series from Polhemus (USA). These products share the following common technical bottlenecks:

[0005] 1) Poor noise resistance. The positioning accuracy fluctuates greatly, especially the Z-axis of the reference frame. In environments with dense metallic objects or strong electromagnetic interference (such as photography studios with steel structures or operating rooms with dense medical equipment), the positioning accuracy decreases significantly due to magnetic field distortion. The LIBERTY series shows a significant decrease in accuracy under strong interference environments.

[0006] 2) System errors accumulate and cannot be independently calibrated and compensated. The non-orthogonality between sensors, the inconsistency of sensitivity between channels, and the model's inherent errors between the sensor installation position and the theoretical model cannot be independently calibrated and compensated, causing errors to accumulate over time. Both NDI and Polhemus products have this problem and require periodic complex calibration.

[0007] 3) Insufficient dynamic accuracy and distorted sampling timing. Existing technologies mostly use MCU serial polling to read data from each sensor. When the hand makes complex movements, there are deviations in the sampling times between different sensors, resulting in the poses of each sensor in the same frame of data not being measured at the "same moment". This leads to timing distortion in the inverse kinematics (IK) solution input. Metaloves, 3DGuidance, and LIBERTY all use serial polling architecture.

[0008] 4) The synchronization mechanism is inefficient and lacks flexibility. Metagloves adopts a "listen-before-transmit (LBT)" distributed synchronization mechanism, which is strongly coupled to clock design and transmission hardware, and its bandwidth and flexibility are limited. NDI and Polhemus adopt a centralized control architecture, in which all sensors share the same clock domain and lack differential channel anti-interference capabilities. For example, Metagloves' LBT mechanism has high latency uncertainty in multi-device scenarios.

[0009] 5) Positioning results are ambiguous and lack robustness under symmetrical structures. When there is a symmetrical structure in the sensor or transmitting coil, existing algorithms struggle to eliminate positioning ambiguity, leading to jumps or errors in positioning results; currently, no existing product has effectively solved this problem.

[0010] In summary, existing electromagnetic positioning solutions have fundamental technical defects in four aspects: anti-interference capability, system error compensation, dynamic sampling synchronization, and synchronization mechanism efficiency. As a result, they cannot meet the requirements of real-time performance, stability, and accuracy in high-precision, high-degree-of-freedom hand motion capture scenarios. Summary of the Invention

[0011] To overcome the shortcomings of existing technologies, a hand pose calculation method and system based on electromagnetic positioning is proposed. The system solves the technical bottlenecks of existing electromagnetic positioning schemes from three levels: signal generation (transmitter), pose calculation (algorithm), and coordinate unification (host computer). It realizes single-cycle 6-DOF positioning, multi-sensor parallel synchronous sampling, and high-precision real-time reconstruction of 22-DOF hand pose.

[0012] A hand pose calculation method based on electromagnetic positioning includes:

[0013] S1. A continuous rotating excitation magnetic field is emitted into space through the orthogonal coil group of the electromagnetic emission module. The orthogonal coil group is supplied with excitation currents of orthogonal phase, forming a rotating magnetic dipole elliptical polarization field in space.

[0014] S2. The time-domain signal of the elliptic polarization field is continuously acquired within a complete excitation cycle by magnetic induction receiving sensors arranged at multiple positions on the hand.

[0015] S3. Perform phase-locked detection on the time-domain signal and extract the parameters of the elliptic polarization field, including: phase, ratio of major to minor axis, absolute amplitude, and elliptic surface normal vector;

[0016] S4. Based on the analytical mapping relationship between the ellipse parameters and the spatial position and attitude of the sensors, the three-dimensional position and three-dimensional attitude of each receiving sensor are simultaneously calculated within a single excitation cycle to obtain the pose data of each receiving sensor.

[0017] S5. Transform all the pose data of the receiving sensors into the same reference coordinate system, and correct the deviation between the actual installation position of the sensor and the preset joint point of the inverse kinematics model through external parameter compensation to obtain the compensated pose data of each sensor.

[0018] S6. Input the compensated pose data of each sensor into the inverse kinematics model to calculate the motion angles of multiple joints of the hand, thereby realizing multi-degree-of-freedom pose estimation of the hand.

[0019] A hand pose calculation system based on electromagnetic positioning, comprising:

[0020] An electromagnetic emission module, located on the back of the hand, includes an orthogonal coil group and a direct digital frequency synthesis circuit, used to generate and emit a continuous rotating excitation magnetic field.

[0021] Multiple magnetic induction receiving sensors are arranged at multiple locations on the hand to continuously acquire the time-domain signal of the elliptic polarization field within a single excitation cycle.

[0022] The unified processing module is connected to each magnetic induction receiving sensor via wired connection. It is used to centrally and synchronously trigger the acquisition of signals from all receiving sensors, and to amplify, filter, digitize, and perform pose calculation on the acquired signals.

[0023] The host computer processing module is used to receive the pose data from each sensor, perform reference coordinate system transformation, extrinsic parameter compensation and inverse kinematics calculation, and output the motion angles of multiple joints of the hand.

[0024] Within a complete excitation cycle, the unified processing module extracts elliptic polarization field parameters through phase-locked detection and calculates the three-dimensional position and three-dimensional attitude of each receiving sensor based on the analytical mapping relationship between the elliptic parameters and the spatial position and attitude of the sensor.

[0025] The hand pose calculation method and system based on electromagnetic positioning provided by this invention have the following technical advantages compared with the prior art:

[0026] 1) Achieve six-degree-of-freedom pose calculation in a single excitation cycle. Two orthogonal coils of the electromagnetic launch module are supplied with continuous sinusoidal currents of orthogonal phase, equivalent to a magnetic dipole rotating at a uniform angular velocity, forming an elliptical polarized composite field at any point in space. Elliptical parameters (phase, ratio of major and minor axes, absolute amplitude, and elliptical surface normal vector) are extracted through phase-locked loop detection. Utilizing the analytical mapping relationship between these parameters and the sensor's spatial position and attitude, the three-dimensional position and attitude can be calculated simultaneously within a single excitation cycle without iterative optimization. Compared to traditional three-axis time-division excitation schemes that require at least two to three excitation cycles to complete a single measurement, the theoretical sampling rate of this invention can be increased to three times that of traditional schemes. The continuous sinusoidal excitation mode is naturally suitable for lock-in amplifier detection, enabling signal extraction within an extremely narrow bandwidth, improving the equivalent signal-to-noise ratio by 40-60 dB, effectively suppressing out-of-band electromagnetic interference, and significantly improving the noise immunity of the positioning.

[0027] 2) Eliminating sampling timing distortion through a centralized synchronization architecture. The system employs an electromagnetic induction structure with back-of-hand transmission and fingertip reception. Each receiving sensor is wirelessly connected to a unified processing module, which uniformly executes synchronous trigger acquisition, magnetic induction signal conversion, and pose calculation. This centralized synchronization mechanism avoids the latency uncertainty caused by multiple devices competing for the listening window in existing "listen-before-send" distributed synchronization schemes. It ensures that the pose of each sensor in the same frame of data is the measured value at the same moment, avoiding the input timing distortion of inverse kinematics calculation caused by serial polling, thereby significantly improving dynamic tracking accuracy.

[0028] 3) Eliminating systematic errors through reference frame transformation and extrinsic parameter compensation. The poses of each fingertip sensor in the transmitter reference frame are uniformly transformed to the back-of-hand reference coordinate system through the transpose of the quaternion-corresponding rotation matrix and Hamiltonian multiplication, eliminating the overall translation and rotation of the hand in space, and retaining only the local motion information of the fingers relative to the back of the hand. At the same time, the sensor measurements are rigidly transformed and corrected through an extrinsic parameter compensation matrix, eliminating the deviation between the actual sensor installation position and the preset joint points of the inverse kinematics model, and avoiding the accumulation of systematic errors such as sensor non-orthogonality and installation errors.

[0029] 4) Improve tracking stability and continuity through data normalization. Using the hand back sensor timestamp as a reference, linear interpolation is used for fingertip position and spherical linear interpolation is used for time alignment of fingertip pose quaternions; anti-jitter processing is performed on synchronized position data using a one-euro filter or a second-order Butterworth filter (cutoff frequency 10 to 15 Hz); data frames exceeding the preset reasonable range are discarded to ensure that the pose data input to the inverse kinematics model is continuous, smooth, and without jumps.

[0030] 5) The transmitting module uses direct digital frequency synthesis technology to generate the excitation signal. The waveform can be selected from sine waves, square waves, triangular waves, or modulation signals such as QPSK and QAM. The optimal waveform can be flexibly selected according to the channel characteristics and anti-interference requirements of the application scenario, improving the system's adaptability and anti-interference capability. The orthogonal coil group can be configured in two-axis or three-axis configurations: two-axis orthogonal coils generate a continuously rotating magnetic field to achieve single-cycle positioning; three-axis orthogonal coils are excited by time-division or frequency-division, and obtain an approximate coupling channel matrix based on the magnetic dipole approximation model, providing an additional spatial reference dimension for special application scenarios.

[0031] 6) Combining the above technical means, the system achieved performance indicators of position error less than 2mm, position stability less than 0.2mm, and real-time performance not less than 70Hz in the embodiment; it successfully restored the continuous motion trajectory of all 22 joints of the hand, with 100% coverage of the distal interphalangeal joints of the index finger, middle finger, and little finger, and more than 80% coverage of most flexion and extension joints. The maximum inter-frame angle change was less than 17 degrees per frame, and there were no outlier jumps. This verified the system's stable tracking performance in continuous motion scenarios and its ability to express complex gestures that combine multi-finger coordination and independent movement. Attached Figure Description

[0032] Figure 1 This is a diagram illustrating the overall architecture of the hand pose calculation system in an embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram of the electromagnetic transmission module and signal generation principle in an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of the magnetic induction receiver and the 6-DOF processing unit in an embodiment of the present invention;

[0035] Figure 4 This is a flowchart of the electromagnetic positioning algorithm in an embodiment of the present invention.

[0036] Figure 5 These are the motion angle trajectories of four representative joints in the time domain in the embodiments of the present invention. Detailed Implementation

[0037] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0038] like Figure 1As shown in the figure, the hand pose calculation system consists of four parts: an electromagnetic transmission module, multiple magnetic induction receiving sensors, a unified processing module, and a host computer processing module. The electromagnetic transmission module, located on the back of the hand, includes an orthogonal coil group and a direct digital frequency synthesis (DDS) circuit to generate and transmit a continuous rotating excitation magnetic field. In this embodiment, the orthogonal coil group uses two spatially orthogonal transmitting coils. Multiple magnetic induction receiving sensors are positioned at the fingertips of the five fingers, each with a built-in triaxial magnetometer to sense the spatial magnetic field and output triaxial time-domain signals. The unified processing module, connected to each magnetic induction receiving sensor via wires and mounted on the back of the hand or forearm, is responsible for centralized synchronous triggering, amplification, filtering, analog-to-digital conversion, and six-degree-of-freedom pose calculation of the signals from all receiving sensors. The host computer processing module receives the pose data from each sensor output by the unified processing module, performs reference coordinate system transformation, extrinsic parameter compensation, and inverse kinematics calculation, and outputs the motion angles of each joint of the hand.

[0039] like Figure 2 As shown in the figure, this diagram illustrates the structural composition of the electromagnetic launch module, including the signal synthesis unit, power drive unit, quadrature launch coil group, and synchronization module. It also illustrates the principle of generating rotating magnetic dipoles by passing quadrature sinusoidal currents through the two-axis quadrature coils. The electromagnetic launch module consists of the signal synthesis unit, power drive unit, quadrature launch coil group, and synchronization module. The signal synthesis unit uses DDS technology to generate a low-frequency excitation signal in the kilohertz range. The DDS, with a phase accumulator and waveform lookup table as its core, can flexibly switch the output waveform under the control of a field-programmable gate array or microcontroller. In this embodiment, the excitation signal uses a continuous sine wave in the typical frequency range of 20 kHz to 100 kHz. In other optional embodiments, square waves, triangular waves, or quadrature phase-shift keying (QPSK), or quadrature amplitude modulation (QAM) signals can be selected according to the application scenario to increase the signal information content or improve anti-interference capability. The power drive unit is located between the digital signal and the coil drive, used for analog modulation and power amplification of the signal, ensuring that the output power is sufficient to cover the entire range of finger movement and that the reflected power in the control space is within a safe range. The orthogonal transmitting coil group consists of two spatially orthogonal transmitting coils, each carrying a sinusoidal current with orthogonal phase; specifically, one coil carries an I0cos(ωt) current, and the other carries an I0sin(ωt) current. They are equivalent to a magnetic dipole with a magnetic moment of m rotating uniformly in a plane at an angular velocity ω. The synchronization module is used to establish a synchronization reference signal between the transmitter and the unified processing module, providing a phase reference for phase-locked detection at the receiver.

[0040] Rotating magnetic dipole elliptic polarization field model

[0041] The core positioning principle of this invention is based on the near-field elliptic polarization characteristics of a rotating magnetic dipole. When two orthogonal coils are supplied with excitation currents of I0cos(ωt) and I0sin(ωt) respectively, the equivalent magnetic moment vector m continuously rotates in the plane with an angular velocity ω. According to the near-field theory of magnetic dipoles, the instantaneous magnetic field B generated by this rotating magnetic dipole at any point P in space can be decomposed into a radial component B_r and a tangential component B_θ. Because the magnetic dipole rotates continuously, the angle between the observation direction and the axis of the magnetic dipole changes synchronously with time, causing the trajectory of the composite field vector at point P within a complete excitation cycle to be depicted as a spatial ellipse. This ellipse has the following characteristic parameters: phase (reflecting the position of the ellipse's starting point within the cycle), the ratio of its major and minor axes a / b (reflecting the flatness of the ellipse), absolute amplitude (reflecting the magnitude of the magnetic field strength), and the elliptical surface normal vector (reflecting the orientation of the ellipse in space). There is a definite analytical mapping relationship between the above ellipse parameters and the spatial position and attitude of the sensor: the phase encodes the azimuth angle ψ (ψ=2φ, where φ is the direction angle of the observation point in the dipole rotation plane), the ratio of the major and minor axes a / b encodes the elevation angle θ, the absolute amplitude encodes the distance r, and the ellipse surface normal vector encodes the sensor attitude (i.e., the three-axis direction cosines of the sensor).

[0042] Signal Acquisition and Ellipse Parameter Extraction. The triaxial magnetometers of each magnetic induction receiving sensor continuously acquire time-domain signals within a complete excitation period T = 2π / ω. The time-domain signal output by the k-th axis (k = 1, 2, 3) is denoted as B_k(t). The unified processing module uses the excitation current signal as a reference to perform digital phase-locked loop detection on the signals of each axis. Specifically, this includes extracting the in-phase component (I-path) and quadrature component (Q-path) of each axis signal, and obtaining the amplitude A_k and phase φ_k of each axis after low-pass filtering. Simultaneously, the squared field strength signal |B(t)|² = B1²(t) + B2²(t) + B3²(t) is calculated. Due to the continuous rotation of the magnetic dipole, the squared field strength signal is a sine wave with a frequency of 2ω. By extracting the phase, peak value |B|²_max, and valley value |B|²_min of this signal, the rotation phase of the ellipse, the ratio of the major and minor axes a / b, and the absolute amplitude |B| can be determined. The normal vector of the elliptical surface is obtained by solving the direction cosine relationship between the amplitude A_k and phase φ_k of each axis.

[0043] like Figure 3 As shown, the signal processing flow of the magnetic induction receiver and the 6-DOF processing unit is as follows: After the magnetic induction receiver collects the time domain signals of the triaxial magnetometers of each sensor, the signal processing unit amplifies / gains and filters them and digitizes them. Then, the 6-DOF pose is calculated in real time, and the data is transmitted to the host computer for further detailed inverse kinematics calculation and restoration of the hand position and movement.

[0044] Six-DOF pose calculation. Position calculation is based on the analytical mapping relationship between elliptical parameters and spherical coordinates, and is directly completed in a closed loop within a unified processing module without iterative optimization. Specifically: the azimuth angle ψ is directly obtained from the elliptical rotation phase (ψ=2φ); the elevation angle θ is obtained from the ratio of the major and minor axes a / b through the analytical relationship θ=arccos(a / b) (when a / b is in the range [0,1]); the distance r is obtained by inverting the absolute amplitude through the magnetic field attenuation formula r∝(1 / |B|)^(1 / 3). Transforming the spherical coordinates (ψ, θ, r) to rectangular coordinates yields the three-dimensional position (x, y, z) of the sensor in the transmitter coordinate system.

[0045] Attitude calculation utilizes the amplitude A_k and phase φ_k of each axis output by phase-locked loop detection, along with the previously extracted elliptical surface normal vector, to solve for the sensor's three-dimensional orientation relative to the transmitter through direction cosine relationships. For the k-th axis of the sensor, let the angle between this axis and the direction of the major axis of the ellipse be δ_k, and the angle with the direction of the direction of the minor axis of the ellipse be ε_k. The direction cosine pairs (cos δ_k, cos ε_k) of the three axes constitute two columns in the rotation matrix. Combined with the elliptical surface normal vector, the direction of the third column is determined. After normalization and orthogonalization, the complete three-dimensional rotation matrix of the sensor relative to the transmitter is obtained, or equivalently represented as quaternions (q_w, q_x, q_y, q_z). Thus, the six-degree-of-freedom (three-dimensional position plus three-dimensional attitude) pose calculation of the sensor is completed within a single excitation cycle T. All fingertip sensors are sampled and calculated in parallel under the control of the same synchronous trigger signal, and the six-degree-of-freedom pose data of each sensor belong to the measurement values ​​at the same time.

[0046] like Figure 4 The diagram shown is a flowchart of the overall calculation process of the electromagnetic positioning algorithm of this invention. It illustrates the data flow and control relationships between each processing node, from signal acquisition, ellipse parameter extraction, position calculation, attitude calculation to outputting 6-DOF pose data.

[0047] A triaxial orthogonal coil implementation is also available. In another implementation, the orthogonal coil group of the electromagnetic transmitting module employs a triaxial orthogonal coil. In this case, the three axial coils are excited in a time-division or frequency-division manner, creating an equivalent rotating magnetic dipole field within each excitation time slot or frequency band. Based on the coupling model approximating the magnetic dipole, the coupling channel matrix between the transmitting coil and the receiving sensor can be estimated at the receiving end. Since the elements of the channel matrix contain distance and direction information, the position and attitude of each sensor can also be calculated within a single period (the total period of the three sub-time slots in time-division mode) by inverting or pseudo-inverseing the channel matrix. This implementation provides an additional spatial reference dimension, suitable for applications requiring higher redundancy or special geometric configurations.

[0048] Input data format

[0049] The host computer processing module receives the pose data from each sensor output by the unified processing module. The input data format is as follows:

[0050] 6DoF pose data mainly includes the following:

[0051] Fields illustrate Timestamp The acquisition time of each frame of data unit Unified data units for calculation coordinate system Unified coordinate system for calculation Quaternion conventions Quaternion notation rules Hand position and posture The three-dimensional position of the back-of-hand emission source or the back-of-hand reference coordinate system and attitude quaternions fingertip position and posture The three-dimensional position and orientation quaternion of the back of the hand relative to the guide wire in its own reference frame.

[0052] Host computer data processing flow. After the unified processing module packages and uploads the six-DOF pose data from each sensor to the host computer processing module, the host computer completes data processing and hand pose restoration step by step according to the following process.

[0053] (a) Data Normalization. The received raw data is first normalized: the unit is unified to millimeters, the coordinate system is a right-handed coordinate system, and all quaternions are normalized to ensure that the modulus is in units of length. Independent attitude estimation is performed on each frame of data, and the uniqueness of the attitude represented by the quaternion is ensured according to the agreed rules.

[0054] (II) Timestamp Synchronization. Since the data from each fingertip sensor is transmitted to the host computer through different channels, there is a slight deviation between its timestamp and that of the back-of-hand sensor. This step uses the timestamp of the back-of-hand sensor as a reference to align the data from the five fingertip sensors to the same moment. Linear interpolation is used for the position data: Let the times of two adjacent frames of the back-of-hand sensor be t1 and t2, and the positions be p(t1) and p(t2) respectively. The fingertip position at the target time t is obtained by interpolation using the following formula: p(t) = p(t1) + (p(t2) - p(t1)) × (t - t1) / (t2 - t1). Spherical linear interpolation is used for attitude quaternions: q(t)=q(t_1)(q^{-1}(t_1)q(t_2))^{(t-t_1) / (t_2-t_1)}, ensuring that the interpolation result is always a unit quaternion and the interpolation path is along the shortest arc on the four-dimensional unit sphere. Linear interpolation of quaternion components is not allowed.

[0055] (III) Filtering and Anti-shake. The synchronized position and attitude data are filtered separately. Position filtering uses a one-euro filter or a second-order Butterworth low-pass filter with a cutoff frequency of 10 Hz to 15 Hz to suppress high-frequency noise and hand tremors. Attitude filtering uses the attitude sliding window mean as a reference for low-pass smoothing. Furthermore, each frame of data undergoes a rationality check: if the position coordinates in a frame exceed the preset spatial range accessible to the wearer's arm, or the attitude angle exceeds the physiological limits of the human hand joints, it is determined to be an abnormal frame and discarded directly, not participating in subsequent calculations.

[0056] (iv) Reference Frame Transformation. The pose data of each fingertip sensor is initially defined in its own local sensor coordinate system. The poses of all fingertip sensors are uniformly transformed to the back-of-hand reference coordinate system. Let the position of the i-th fingertip sensor in the transmitter reference system B be pᵢB, and its attitude quaternion be qᵢB; let the position of the back-of-hand reference coordinate system H in the transmitter reference system B be p_HB, and its attitude quaternion be q_HB. Then the position p_iH and attitude q_iH of the fingertip sensor in the back-of-hand reference system H are obtained by the following formulas: p_iH=R(q_HB)^T(p_iB-p_HB); q_iH=q_HB^{-1}⊗q_iB. Where R(q) is the rotation matrix corresponding to the quaternion q, (·)^T represents the matrix transpose (i.e., inverse rotation), and ⊗ represents the Hamiltonian multiplication of the quaternion. This transformation eliminates the translation and rotation of the hand as a whole in space, retaining only the local motion information of each finger relative to the back of the hand.

[0057] (V) Extrinsic Parameter Compensation. In actual use, the magnetic induction receiving sensor is usually mounted on the surface of the fingernail or fingertip. There is a rigid offset between its origin and the positions of the joints (such as metacarpophalangeal joints, proximal interphalangeal joints, distal interphalangeal joints, and fingertips) preset in the inverse kinematics model. This step uses a pre-calibrated extrinsic parameter compensation matrix T_ext (containing translation vectors and rotation matrices) to rigidly transform and correct the sensor pose data after reference frame transformation: p_iH'=R_ext·p_iH+t_ext; q_iH'=q_ext⊗q_iH. Here, R_ext and t_ext are the rotation and translation components of the extrinsic parameter compensation matrix T_ext, and q_ext is the quaternion corresponding to R_ext. After extrinsic parameter compensation, the sensor's measurement pose is corrected to the joint positions preset in the inverse kinematics model, eliminating the system error introduced by the sensor installation position deviation.

[0058] (vi) Inverse kinematics calculation. After the above steps are completed, all sensor data have been transformed to the same back-of-hand reference frame and extrinsic parameter compensation has been completed. The spatial coordinates correspond one-to-one with each preset joint point of the inverse kinematics model. At this time, the inverse kinematics model is called for calculation. This embodiment uses a 22-DOF Revolute joint model, covering five degrees of freedom for the thumb (wrist-metacarpophalangeal joint flexion-extension and lateral movement, metacarpophalangeal joint flexion-extension and lateral movement, interphalangeal joint flexion-extension), four degrees of freedom for the index finger (metacarpophalangeal joint flexion-extension and lateral movement, proximal interphalangeal joint flexion-extension, distal interphalangeal joint flexion-extension), four degrees of freedom for the middle finger, four degrees of freedom for the ring finger, and five degrees of freedom for the little finger (wrist-metacarpophalangeal joint curling, metacarpophalangeal joint flexion-extension and lateral movement, proximal interphalangeal joint flexion-extension, distal interphalangeal joint flexion-extension). Since the transmitter is located on the back of the hand, the sensor pose calculation results are naturally in the back of the hand reference frame. The inverse kinematics model can directly obtain the global position and attitude of all joints of the hand in the back of the hand reference frame, and calculate the rotation angle of each joint through the inverse kinematic chain, and finally output the motion angle data of all 22 degrees of freedom of the hand.

[0059] Example: Motion capture glove drives a 22-DOF manipulator

[0060] The following experimental data will be used to verify and explain the pose estimation effect of the present invention.

[0061] This embodiment uses a motion capture glove (single-handed) equipped with the positioning algorithm described in this invention to collect the operator's hand movements and reproduce these movements in real time on a 22-DOF robotic hand (modeled based on a physics simulation engine). The motion capture glove employs a structure with a two-axis orthogonal transmitting coil group mounted on the back of the hand and a three-axis magnetic induction receiving sensor mounted on each of the five fingertips. The sampling rate is set to 66Hz. The operator continuously performs basic hand gestures such as clenching and opening the fist, and independently flexing and extending each finger, recording approximately 22 seconds of data.

[0062] After the experiment, the actual range of motion of each joint was compared with the theoretical range of motion, and the coverage rate (the percentage of the measured range to the theoretical range) was calculated. The specific results are as follows:

[0063] finger joint Types of sports Theoretical Scope Measured range Coverage thumb CMC lateral movement -10.0°~110.0° 16.0°~110.0° 78.4% thumb CMC flexion and extension -20.0°~20.0° -2.8°~20.0° 57.0% thumb MCP flexion and extension -30.0°~80.0° 3.6°~28.2° 22.4% thumb MCP lateral movement -20.0°~20.0° -20.0°~19.6° 99.0% thumb IP flexion and extension 0°~100.0° 0°~48.1° 48.1% index finger MCP flexion and extension -10.0°~90.0° -10.0°~69.1° 79.1% index finger MCP lateral movement -20.0°~20.0° -10.5°~10.5° 52.3% index finger PIP flexion and extension 0°~100.0° 0°~86.1° 86.1% index finger DIP flexion and extension 0°~80.0° 0°~80.0° 100.0% middle finger MCP flexion and extension -10.0°~90.0° -10.0°~79.0° 89.0% middle finger MCP lateral movement -20.0°~20.0° -5.0°~4.6° 24.2% middle finger PIP flexion and extension 0°~100.0° 0°~95.0° 95.0% middle finger DIP flexion and extension 0°~80.0° 0°~80.0° 100.0% ring finger MCP flexion and extension -10.0°~90.0° -10.0°~82.7° 92.7% ring finger MCP lateral movement -20.0°~20.0° -4.9°~8.1° 32.5% ring finger PIP flexion and extension 0°~100.0° 0°~94.6° 94.6% ring finger DIP flexion and extension 0°~80.0° 0°~78.0° 97.5% little finger CMC curly 0°~15.0° 0°~15.0° 99.9% little finger MCP flexion and extension -10.0°~90.0° -10.0°~82.6° 92.6% little finger MCP lateral movement -20.0°~20.0° 1.2°~19.8° 46.5% little finger PIP flexion and extension 0°~100.0° 0°~96.2° 96.2% little finger DIP flexion and extension 0°~80.0° 0°~80.0° 100.0%

[0064] The data in the table above shows that the coverage rate of the distal interphalangeal joints of the index, middle, and little fingers all reached 100%, indicating that the system can completely capture the flexion and extension movements of the distal phalanges. The coverage rate of metacarpophalangeal joint flexion and extension and proximal interphalangeal joint flexion and extension of most fingers exceeded 80%, with the middle finger's proximal interphalangeal joint coverage reaching 95%, the ring finger's reaching 94.6%, and the little finger's reaching 96.2%. The coverage rate of thumb carpometacarpophalangeal joint lateral flexion was 78.4%, and the little finger's carpometacarpophalangeal joint curling coverage was close to 100%, indicating that both thumb opposition and little finger independent movements can be accurately reproduced. The measured range of lateral flexion (adduction / abduction) of the metacarpophalangeal joints of each finger was relatively small (coverage rate between 24% and 52%), which is consistent with the physiological law of limited changes in lateral flexion angle during normal fist clenching and opening movements, and is not due to insufficient system tracking ability.

[0065] like Figure 5 As shown, the time-domain angular trajectories of four representative joints in this experiment are presented. The selected joints are thumb CMC lateral oscillation, index finger PIP flexion and extension, middle finger PIP flexion and extension, and little finger PIP flexion and extension, representing the thumb's large-range palmar opposition movement, fist clenching / opening dominant joint, multi-finger coordinated movement, and distal finger tracking ability, respectively.

[0066] In the approximately 22 seconds of continuous motion data recorded, the trajectories of each joint angle were smooth and continuous, with no abnormal jumps between frames. Statistically, the maximum inter-frame angle change was less than 17 degrees per frame, and there were no outlier jump points. The temporal trajectories of four representative joints (thumb carpometacarpal joint lateral oscillation, index finger proximal interphalangeal joint flexion and extension, middle finger proximal interphalangeal joint flexion and extension, and little finger proximal interphalangeal joint flexion and extension) showed that the proximal interphalangeal joints of the index, middle, and little fingers exhibited a clear synchronous flexion and extension trend, consistent with the physiological characteristics of coordinated movement of the four fingers in a fist-clenching action; the lateral oscillation trajectory of the thumb carpometacarpal joint showed a phase difference with the movement of the four fingers, reflecting the palmar opposition movement characteristics of the thumb independent of the four fingers.

[0067] The experimental results above show that the method and system of the present invention can stably complete the 6-DOF pose calculation of each sensor within a single excitation cycle. After reference frame transformation, external parameter compensation and inverse kinematics calculation, the continuous motion trajectory of all 22 joints of the hand can be successfully restored in real time. It has stable tracking performance and the ability to express complex gestures in continuous motion scenarios.

[0068] The above is a description of the present invention to help understand it; however, the implementation of the present invention is not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the principle of the present invention should be considered equivalent substitutions and are included within the protection scope of the present invention.

Claims

1. A hand pose calculation method based on electromagnetic positioning, characterized in that, include: S1. A continuous rotating excitation magnetic field is emitted into space through the orthogonal coil group of the electromagnetic emission module. The orthogonal coil group is supplied with excitation currents of orthogonal phase, forming a rotating magnetic dipole elliptical polarization field in space. S2. The time-domain signal of the elliptic polarization field is continuously acquired within a complete excitation cycle by magnetic induction receiving sensors arranged at multiple positions on the hand. S3. Perform phase-locked detection on the time-domain signal and extract the parameters of the elliptic polarization field, including: phase, ratio of major to minor axis, absolute amplitude, and elliptic surface normal vector; S4. Based on the analytical mapping relationship between the ellipse parameters and the spatial position and attitude of the sensors, the three-dimensional position and three-dimensional attitude of each receiving sensor are simultaneously calculated within a single excitation cycle to obtain the pose data of each receiving sensor. S5. Transform all the pose data of the receiving sensors into the same reference coordinate system, and correct the deviation between the actual installation position of the sensor and the preset joint point of the inverse kinematics model through external parameter compensation to obtain the compensated pose data of each sensor. S6. Input the compensated pose data of each sensor into the inverse kinematics model to calculate the motion angles of multiple joints of the hand, thereby realizing multi-degree-of-freedom pose estimation of the hand.

2. The hand pose calculation method based on electromagnetic positioning as described in claim 1, characterized in that, In S1, the orthogonal coil group is a two-axis orthogonal coil or a three-axis orthogonal coil; When using two-axis orthogonal coils, sinusoidal currents with orthogonal phases are passed through them, which is equivalent to a magnetic dipole rotating in a plane. When using a triaxial orthogonal coil, time-division or frequency-division excitation is performed. An approximate coupling channel matrix is ​​obtained based on the magnetic dipole approximation model. Distance and direction are calculated in a single cycle by inverting the channel matrix.

3. The hand pose calculation method based on electromagnetic positioning as described in claim 1, characterized in that, In step S3, extracting elliptic parameters through phase-locked detection specifically includes: Using the excitation current as a reference signal, the in-phase and quadrature components of the signals of each axis are extracted to obtain the amplitude and phase of each axis. Calculate the time-domain signal of the squared field strength, and extract the phase, peak value, and valley value of the time-domain signal; Among them, the phase encodes the azimuth angle, the ratio of the major and minor axes encodes the elevation angle, the absolute amplitude encodes the distance, and the elliptical surface normal vector encodes the sensor attitude.

4. The hand pose calculation method based on electromagnetic positioning as described in claim 1, characterized in that, In S4, the position calculation is completed in a closed loop in the arm-mounted processing module through analytical formulas, without the need for iterative optimization. Attitude calculation is performed by solving the sensor’s three-dimensional orientation using the direction cosine relationship. The direction cosine pairs of each axis are used to form a rotation matrix. Combined with the elliptical surface normal vector, the three-axis orientation is transformed to the transmitter coordinate system, resulting in the complete three-dimensional rotation matrix or equivalent quaternion of the sensor relative to the transmitter.

5. The hand pose calculation method based on electromagnetic positioning as described in claim 1, characterized in that, In step S5, the reference coordinate system transformation specifically includes: Let the position of the i-th fingertip receiving sensor in the transmitter reference frame be piB, and its attitude quaternion be qiB; The position of the back-of-hand reference coordinate system in the source reference system is pHB, and the attitude quaternion is qHB; The position piH and orientation qiH of the fingertip sensor in the back-of-hand reference frame H are obtained through the quaternion rotation formula and Hamiltonian multiplication, thus unifying the orientation of all fingertip sensors to the back-of-hand reference frame and eliminating overall hand translation and rotation.

6. The hand pose calculation method based on electromagnetic positioning as described in claim 1, characterized in that, It also includes data synchronization and filtering steps: Based on the timestamp of the back-of-hand sensor, a linear interpolation method is used to interpolate the fingertip position between two adjacent frames of the back-of-hand sensor. The spherical linear interpolation method is used to interpolate the fingertip posture quaternion between two adjacent frames of the back-of-hand sensor. The synchronized position and attitude data are filtered separately. The position filter uses a 1-euro filter or a second-order Butterworth filter with a cutoff frequency of 10-15Hz. If the position or attitude angle of the received data frame exceeds the preset reasonable range, the data frame is discarded.

7. The hand pose calculation method based on electromagnetic positioning as described in claim 1, characterized in that, The excitation signal of the electromagnetic transmission module is generated using direct digital frequency synthesis technology. The signal waveform can be selected as sine wave, square wave, triangle wave, or QPSK or QAM modulated signal. The magnetic induction receiving sensor is wirelessly connected to a unified processing module, which performs synchronous trigger acquisition, magnetic induction signal conversion, and pose calculation.

8. A hand pose calculation system based on electromagnetic positioning, characterized in that, include: An electromagnetic emission module, located on the back of the hand, includes an orthogonal coil group and a direct digital frequency synthesis circuit, used to generate and emit a continuous rotating excitation magnetic field. Multiple magnetic induction receiving sensors are arranged at multiple locations on the hand to continuously acquire the time-domain signal of the elliptic polarization field within a single excitation cycle. The unified processing module is connected to each magnetic induction receiving sensor via wired connection. It is used to centrally and synchronously trigger the acquisition of signals from all receiving sensors, and to amplify, filter, digitize, and perform pose calculation on the acquired signals. The host computer processing module is used to receive the pose data from each sensor, perform reference coordinate system transformation, extrinsic parameter compensation and inverse kinematics calculation, and output the motion angles of multiple joints of the hand. Within a complete excitation cycle, the unified processing module extracts elliptic polarization field parameters through phase-locked detection and calculates the three-dimensional position and three-dimensional attitude of each receiving sensor based on the analytical mapping relationship between the elliptic parameters and the spatial position and attitude of the sensor.

9. The hand pose calculation system based on electromagnetic positioning as described in claim 8, characterized in that, The electromagnetic transmission module also includes a synchronization signal output interface for providing a synchronization reference signal to the unified processing module; The unified processing module includes a multi-channel synchronous sampling circuit. The signal channels of all magnetic induction receiving sensors are sampled in parallel under the control of the same synchronous trigger signal, ensuring that the pose of each sensor in the same frame of data is the measured value at the same time.

10. The hand pose calculation system based on electromagnetic positioning as described in claim 8, characterized in that, The host computer processing module is specifically used for: The pose data of each magnetic induction receiving sensor are uniformly transformed to the back of the hand reference coordinate system using a quaternion rotation formula; wherein, the pose data includes 6-DOF pose data; The sensor measurements are rigidly transformed and corrected by using an external parameter compensation matrix to eliminate the deviation between the actual installation position of the sensor and the preset joint points of the inverse kinematics model. The compensated pose data is input into an inverse kinematics model containing 22 degrees of freedom to calculate the motion angles of each joint of the hand. The 22 degrees of freedom include: 5 degrees of freedom for the thumb, 4 degrees of freedom for the index finger, 4 degrees of freedom for the middle finger, 4 degrees of freedom for the ring finger, and 5 degrees of freedom for the little finger.

Images