A kind of initialization of calibration method for M-IMU human motion capture system
Technical field
The present invention relates to the initialization of calibration method for M-IMU human motion capture system.
Background technology
Human motion capture is widely used in man-machine interaction, virtual reality, athletic training, medical diagnosis are controlled with rehabilitation
The fields such as treatment, robot control, Film Animation making.The most conventional human motion capture system has optical profile type, video image
Formula, mechanical type, M-IMU, electromagnetic type, infrared type, sound wave type, but these systems have shortcoming more or less, it is impossible to completely
Cover above-mentioned application.Along with the development and progress of science and technology, micro-electro-mechanical sensors (MEMS) is at miniaturization, precision and cost
On had a great breakthrough, human motion capture system attractive based on the MEMS research of numerous scholars.M-IMU sensor system
System is made up of a three axis accelerometer, a three-axis gyroscope and three axle magnetometers, and these sensors can be integrated
In the least a piece of chip, thus reduce size and the weight of system, be arranged on limbs and also just will not hinder human body
Often motion.
Human motion capture system based on M-IMU needs to install an IMU/M-IMU on each limbs of human body and passes
Sensor, by measuring and estimating that the attitude of limbs obtains the movable information of human body.But, the survey of sensor under normal conditions
Amount coordinate system is misaligned with corresponding human body coordinate system, and such motion measurement is nonsensical, it is necessary to by suitable
Body motion information could be converted into the biological parameters such as joint angle by initialization of calibration program.Initialization of calibration program is used for counting
Calculate the relative rotation matrices between sensor coordinate system and human body coordinate system, and then turning based on the measurement under sensor coordinate system
Change into based on the kinematics information under human body coordinate system, be the M-IMU requisite process of human motion capture system.
Initialization of calibration algorithm of the prior art is broadly divided into three classes:
First kind initialization of calibration method is to be installed by accurately sensor, makes measurement coordinate system and the human body of sensor
Coordinate system is completely superposed.Owing to the limbs surface of human body is irregular, the limited precision of this calibration steps.On the other hand,
From the point of view of user uses angle, it is a complicated task that accurate sensor is installed, and is not suitable in daily life.
Equations of The Second Kind initialization of calibration method is to be completed by aid, and this aid passes through skeleton index point
Measure the rotary shaft of human synovial, and then obtain the sensor coordinate system spin matrix relative to human body coordinate system.This side
Method needs certain anatomical knowledge, and process is loaded down with trivial details.
3rd class initialization of calibration method is to be completed by a series of calibration posture and action.People under calibration posture
One axle of body coordinate system is parallel with gravity, it is possible to use the acceleration of gravity of accelerometer measures obtains.In calibration actions
The joint of lower human body rotates around another axle, it is possible to use the angular velocity that gyroscope is measured obtains.Last axle can be in order to
Obtain with orthogonality principle.This calibration steps is simple, convenient, it is adaptable to daily life, but the calibration of this method
Precision depends on calibrating posture and the standard degree of calibration actions.
In sum, there is the defects such as operation is complicated, precision is low in existing initialization of calibration method.
Summary of the invention
The present invention is to solve the problems such as prior art existence operation is complicated, precision is low, and the one proposed is for M-
The initialization of calibration method of IMU human motion capture system.
A kind of initialization of calibration method for M-IMU human motion capture system realizes according to the following steps:
Step one: according to kinesiology's model and human anatomy, human body limb be divided into 17 parts and number, and
Each part of human body is set up human body coordinate systemWith corresponding sensor coordinate systemSet up reference frame simultaneouslyWherein i=1,2,3 ..., 17;
Step 2: M-IMU sensor is fixed on 17 corresponding limbs;
Step 3: set three kinds of calibration postures, i.e. calibration posture 1, calibration posture 2 and calibration posture 3;Record every kind of calibration
Sensor measurement data under postureWhereinInstitute
StateOn respectively 17 limbs, sensor coordinate system is relative to the spin matrix data of reference frame,The sensor coordinate system that respectively n time measurement obtains is relative to the spin matrix data of reference frame;
Step 4: the M-IMU sensor measurement data obtained step 3 under every kind of calibration posture accumulates average in doing,
Average measurement to M-IMU sensor under three kinds of calibration posturesWherein
DescribedIt is respectivelyAverage magnitude measured value;
Step 5: calculating the relative rotation matrices of shoulder joint under three kinds of calibration postures, the M-IMU that integrating step four obtains passes
The average magnitude measured value of sensor, utilizes what hand and eye calibrating algorithm obtained trunk sensor coordinate system and human body coordinate system to rotate against square
Battle array
Step 6: the trunk sensor coordinate system obtained according to step 5 obtains with the relative rotation matrices of human body coordinate system
Trunk human body coordinate system is relative to the spin matrix of reference frame
Step 7: calibrating posture 1 time, the human body coordinate system of trunk, basin bone, head, left upper extremity and left lower extremity is the most flat
OK, i.e.The human body coordinate system of right upper extremity and right lower extremity be parallel to each other and with the human body coordinate of trunk
System's one relative rotation matrices of differenceI.e.Wherein said
For trunk, basin bone, head, left upper extremity and left lower extremity limbs human body coordinate system relative to the spin matrix of reference frame,For the human body coordinate system of right upper extremity and right lower extremity relative to the spin matrix of reference frame;
Step 8: the sensor measurement under calibration posture 1 obtained according to step 4
And the human body coordinate system that obtains of step 7 is relative to the spin matrix of reference frameObtain the sensor of each limbs
Coordinate system is relative to the relative rotation matrices of human body coordinate system
Invention effect:
(1) relative rotation matrices obtained between sensor coordinate system and human body coordinate system can be calibrated, reduce installation position
Putting the inaccurate error brought, compared with not carrying out calibration actions, the precision of action capture is its 3 times.
(2) calibrating mode is simple, the shortest, and only then sensor need to be arranged on each dismemberent of health does three standard appearances
Gesture can realize calibration, it is to avoid loaded down with trivial details action calibration, and because the nonstandard error brought of action.
(3) cheap, it is not necessary to by other equipment, utilize IMU sensor can realize calibration.According to sides such as optics
Method is calibrated, and needs to increase extra optical device, is greatly improved calibration cost.This method can be substantially reduced and be calibrated to
This.
(4) accumulate in have employed average algorithm to sensor acquisition to data carried out pretreatment, ensure follow-up calibration
The reliability of algorithm.
Accompanying drawing explanation
Fig. 1 human body limb is numbered
Fig. 2 reference frame and human body coordinate system schematic diagram;
Fig. 3 calibrates posture 1 illustraton of model;
Fig. 4 calibrates posture 2 illustraton of model;
Fig. 5 calibrates posture 3 illustraton of model.
Detailed description of the invention
Detailed description of the invention one: a kind of initialization of calibration method following steps for M-IMU human motion capture system:
Hinged multi-rigid model is widely used in human motion analysis and biomechanics Research, and this model is by human body
Each limbs regard rigid body as, ignore the translational motion of human synovial, only consider its rotary motion.Simplify anthropometric dummy by head,
Trunk, basin bone, upper limb and lower limb composition.
Step one: according to kinesiology's model and human anatomy, human body limb be divided into 17 parts and number, as
Shown in Fig. 1, and each part of human body is set up human body coordinate systemWith corresponding sensor coordinate systemSet up ginseng simultaneously
Examine coordinate systemWherein i=1,2,3 ..., 17;
Step 2: M-IMU sensor is fixed on 17 corresponding limbs;
Step 3: set three kinds of calibration postures, i.e. calibration posture 1, calibration posture 2 and calibration posture 3;Record every kind of calibration
Sensor measurement data under postureWhereinInstitute
StateOn respectively 17 limbs, sensor coordinate system is relative to the spin matrix data of reference frame,The sensor coordinate system that respectively n time measurement obtains is relative to the spin matrix data of reference frame;
Step 4: the M-IMU sensor measurement data obtained step 3 under every kind of calibration posture accumulates average in doing,
Average measurement to M-IMU sensor under three kinds of calibration posturesWherein
DescribedIt is respectivelyAverage magnitude measured value;
Step 5: calculating the relative rotation matrices of shoulder joint under three kinds of calibration postures, the M-IMU that integrating step four obtains passes
The average measurement of sensor, utilizes hand and eye calibrating algorithm to obtain the relative rotation matrices of trunk sensor coordinate system and human body coordinate system
Step 6: the trunk sensor coordinate system obtained according to step 5 obtains with the relative rotation matrices of human body coordinate system
Trunk human body coordinate system is relative to the spin matrix of reference frame
Step 7: calibrating posture 1 time, the human body coordinate system of trunk, basin bone, head, left upper extremity and left lower extremity is the most flat
OK, i.e.The human body coordinate system of right upper extremity and right lower extremity be parallel to each other and with the human body coordinate of trunk
System's one relative rotation matrices of differenceI.e.Wherein said
For trunk, basin bone, head, left upper extremity and left lower extremity human body coordinate system relative to the spin matrix of reference frame,For the human body coordinate system of right upper extremity and right lower extremity relative to the spin matrix of reference frame;
Step 8: the sensor measurement under calibration posture 1 obtained according to step 4With
And the human body coordinate system that obtains of step 7 is relative to the spin matrix of reference frameThe sensor obtaining each limbs is sat
Mark system is relative to the relative rotation matrices of human body coordinate system
Detailed description of the invention two: present embodiment is unlike detailed description of the invention one: to human body in described step one
Each part set up human body coordinate systemWith corresponding sensor coordinate systemSet up reference frame simultaneouslySpecifically
For:
According to kinesiology's model, human body it is divided into 17 parts and numbers;Motion morphology in conjunction with human body defines human body
Coordinate system, in body erect, both hands are flattened when health both sides, set up people on trunk, basin bone, head, left upper extremity, left lower extremity
Body coordinate system, x-axis sensing health front, y-axis sensing left side of body, z-axis is vertically upward;The human body coordinate of right upper extremity and right lower extremity
System be parallel to each other but with the human body coordinate system specular of left limb, the x-axis of the human body coordinate system of right upper extremity and right lower extremity
Pointing to health front, y-axis points to right side of body, and z-axis is vertically downward;Reference frame is terrestrial coordinate system, and X* axle points to earth magnetism
North, Y* axle points to east, and Z* axle is vertically downward.The schematic diagram of reference frame and human body coordinate system schematic diagram is as shown in Figure 2.Sensing
Device coordinate systemBeing determined by the installation site of sensor internal gyroscope, M-IMU sensor is arbitrarily installed in each part.
Other step and parameter are identical with detailed description of the invention one.
Detailed description of the invention three: present embodiment is unlike detailed description of the invention one or two: described step 3 lieutenant colonel
Quasi-posture 1, calibration posture 2 and calibration posture 3 particularly as follows:
Calibration posture 1 is for keeping health and lower limb upright, and both feet close up, and both arms naturally droop, and the centre of the palm is towards inside health;
Calibration posture 2 is for keeping health and lower limb upright, and both feet close up, and both hands are flattened to health both sides, and the centre of the palm is downward;
Calibration posture 3 is for keeping health and lower limb upright, and both feet close up, and both arms stretch forward flat with shoulder, and the centre of the palm is downward.
User performs three kinds of calibration postures successively, and every kind of calibration posture keeps ten seconds, at every kind of posture lower sensor
Data are gathered with the sample frequencys of 10 hertz.Three kinds of calibration postures are as shown in Fig. 3, Fig. 4 and Fig. 5.
Other step and parameter are identical with detailed description of the invention one or two.
Detailed description of the invention four: present embodiment is unlike one of detailed description of the invention one to three: described step 4
In step 3 is accumulate under every kind of calibration posture, the M-IMU sensor measurement data that obtains does average, obtain three kinds of calibrations
The average measurement of M-IMU sensor under postureParticularly as follows:
User inevitably shakes when performing calibration posture, accumulates average algorithm and survey sensor in utilization
Amount is averaged.
As a example by the measurement of one of them sensor, n group is measuredDo average:
Wherein R is spin matrix, RtFor measure every time spin matrix.
This interior core accumulateing average algorithm is will to utilize logarithmic mapping that from Lie group SO (3), spin matrix is transformed into Lie algebra
In so (3), then doing averagely at so (3), be finally transformed into SO (3), algorithm is as follows:
Calculate the error matrix on SO (3)Error matrix is converted to so (3) from SO (3):
Wherein Δ θtFor the anglec of rotation of each limbs sensor, Δ r in so (3)tPass for each limbs in so (3)
The skew symmetric matrix form of the rotary shaft vector of sensor.
It is averaging error matrix at so (3):
Mean error matrix is transformed into SO (3) from so (3):
WhereinAverage for Δ θ.
SO (3) calculates mean matrix:
Other step and parameter are identical with one of detailed description of the invention one to three.
Detailed description of the invention five: present embodiment is unlike one of detailed description of the invention one to four: described step 5
The relative rotation matrices of shoulder joint under middle calculating three kinds calibration posture, the average survey of the M-IMU sensor that integrating step four obtains
Amount, utilizes hand and eye calibrating algorithm to obtain the relative rotation matrices of trunk sensor coordinate system and human body coordinate systemConcrete mistake
Cheng Wei:
When calibrating posture 1, the human body coordinate system of left upper extremity is counterclockwise around X-axis relative to trunk human body coordinate system
90-degree rotation, now the spin matrix of shoulder joint is:
When calibrating posture 2, the human body coordinate system of left upper extremity is parallel with trunk human body coordinate system, now shoulder joint
Spin matrix be:
When calibrating posture 3, the human body coordinate system of left upper extremity is the most counterclockwise relative to trunk human body coordinate system
90-degree rotation, now the spin matrix of shoulder joint is:
Under three kinds of calibration postures, the kinematical equation of left shoulder joint is as follows:
It is hand and eye calibrating equation by formula (2) abbreviation:
WhereinFor trunk human body coordinate system relative to the spin matrix of basin bone human body coordinate system;
The relative rotation matrices of trunkBy using method of least square to obtain in Lie group:
Wherein vector αj,βj, j=1,2 is matrix Aj,Bj, j=1, the normalized index coordinates of 2, by the following method
Calculate and obtain:
The relative rotation matrices of trunk sensor coordinate system and human body coordinate system is:
Other step and parameter are identical with one of detailed description of the invention one to four.
Detailed description of the invention six: present embodiment is unlike one of detailed description of the invention one to five: described according to step
The trunk human body coordinate system spin matrix relative to reference frame is obtained in rapid sixParticularly as follows:
Other step and parameter are identical with one of detailed description of the invention one to five.
Detailed description of the invention seven: present embodiment is unlike one of detailed description of the invention one to six: described step 7
The human body coordinate system of middle right upper extremity and right lower extremity is parallel to each other and differs one with the human body coordinate system of trunk and rotates against square
Battle arrayI.e.Particularly as follows:
The human body coordinate system of right upper extremity and right lower extremity is parallel to each other and differs a rotation with the human body coordinate system of trunk
Matrix
The human body coordinate system of right upper extremity and right lower extremity relative to the spin matrix of reference frame is:
Other step and parameter are identical with one of detailed description of the invention one to six.
Detailed description of the invention eight: present embodiment is unlike one of detailed description of the invention one to seven: described step 8
In obtain the sensor coordinate system relative rotation matrices relative to human body coordinate system of each limbsDetailed process be:
The sensor coordinate system of trunk and left limb relative to the relative rotation matrices of human body coordinate system is:
The sensor coordinate system of trunk and right side limbs relative to the relative rotation matrices of human body coordinate system is:
Other step and parameter are identical with one of detailed description of the invention one to seven.
Embodiment one:
Because the reason of mounting means, on each limbs of human body, sensor coordinate system and corresponding human body coordinate system are not weigh
Close.In order to improve the precision of capturing movement, so needing that it is carried out initialization of calibration to obtain sensor coordinate system and human body
The relative rotation matrices of coordinate system.In order to ask for this relative rotation matrices we firstly the need of each by sensor measurement human body
The attitude data of individual limbs.Sensor uses adding of 3 axis accelerometers, 3 axle gyroscopes and the 3 axle each limbs of magnetometer measures human body
Speed, angular speed and Geomagnetism Information.According to human skeleton model, only sensor is placed at skeleton and just can be compared
More accurate data, but it is because the restriction of mounting means, the position fewer with human body skin can only be installed, as far as possible
Place with skeleton laminating.Research shows that laying of sensor should reduce the impact that soft tissue brings as far as possible.In consideration
State principle, when person upright's both arms naturally droop, the sensor of large arm is placed in the flesh of triceps brachii and elbow joint junction
Tendon position, little arm sensor is placed in forearm back, and the sensor of palm is placed in the back of the hand center.Leg sensor also in compliance with
Same principle so that it is fit with skeleton.
Three standard gestures i.e. standard gestures 1, standard gestures 2, standard gestures 3 is taken turns doing after installing sensor.Sensing
Device measures many group acceleration of each limbs, angular velocity and Geomagnetism Information.Then by Kalman filtering by sensor measurement
Data carry out merging the spin matrix obtaining each limbs sensor coordinate system relative to reference frame
Because human body can produce shake unavoidably when doing standard gestures, so we accumulate flat in asking it by the multi-group data measuring it
Equal method obtains a meansigma methodsThis algorithm utilizes logarithmic mapping by spin matrix
It is transformed into Lie algebra so (3) from Lie group SO (3), then does averagely at so (3), reconvert to SO (3).Finally obtain sensor
Coordinate system is relative to the meansigma methods of the spin matrix of reference frame.Under three standard gestures, we can obtain shoulder joint phase
Spin matrix for reference frameThen in conjunction with kinematical equationBy its abbreviation it is
Method of least square is being used to obtain trunk sensor coordinate system relative to human body coordinate system in Lie group after hand and eye calibrating equation
Spin matrixThen the trunk human body coordinate system spin matrix relative to reference frame is tried to achieveAt standard gestures 1
The human body coordinate system of head portion, trunk, basin bone and left limb is parallel to each other, and the rotating against of known right-hand limbs and trunk
Quaternary number.So finally can obtain each in the hope of the human body coordinate system of each limbs relative to the spin matrix of reference frame
The sensor coordinate system of individual limbs is relative to the spin matrix of human body coordinate system