CN104535077A - Pedestrian step length estimation method based on intelligent mobile terminal equipment - Google Patents

Abstract

The invention provides a pedestrian step length estimation method based on intelligent mobile terminal equipment. According to the method, a sensor signal acquisition device (111), an information acquirer (112), a step length estimation core processing unit (113) and an output controller (114) which are sequentially connected are included. After a step length estimation model and an estimated step length are initialized, the pedestrian step length estimation method includes the steps that S1, the step length estimation core processing unit (113) obtains and updates the estimated step length and the step length estimation model according to sensing information acquired by an acceleration sensor, a gyroscope, a magnetometer and a camera used for obtaining images; S2, the output controller judges whether the step estimation model is convergent or not, if yes, only the sensing information of the acceleration sensor needs to be acquired, the estimated step length is obtained according to the sensing information of the acceleration sensor and the step length estimation model, the estimated step length is output, and the S2 is executed again; if not, abnormal step length filtering processing is conducted on the estimated step length, and the S1 is returned.

Description

A kind of pedestrian's step-size estimation method based on intelligent mobile terminal equipment

Technical field

The present invention relates to pedestrian's dead reckoning field, particularly, relate to a kind of pedestrian's step-size estimation method based on intelligent mobile terminal equipment.

Background technology

In pedestrian's dead reckoning system, step-size estimation module is one of main source of error of system.In traditional pedestrian's dead reckoning system, step-length is commonly referred to be constant, and this obviously can introduce a large amount of errors.According to physiological principle, the step-length of different pedestrian is relevant with body weight with its height (or leg is long), therefore, part researchs and proposes the empirical model based on pedestrian's height and body weight information, the step-length of pedestrian is estimated, but, because different rows people walking habits varies, be difficult to accurately be estimated by the step-length of this step-size estimation model to different pedestrian; In addition, research shows, pedestrian's step-length is also relevant to the cadence of pedestrian and acceleration variance, therefore, propose the linear step-length estimation model based on pedestrian's cadence and acceleration variance, this model needs the acceleration information in the different cadence situation of the pedestrian by gathering, and utilize these data to carry out matching estimation to the parameter of linear regression model (LRM), in order to improve the precision of pedestrian's step-size estimation, have to gather a large amount of data in the training process, and different pedestrian needs to train different models, and this brings no small trouble to practical application.In order to reduce the workload that off-line training brings, someone proposes to utilize the satellite-signals such as GPS to carry out the method for on-line amending, but the method is only applicable to outdoor and unobstructed environment, and for situations that environmental satellite signal obviously declines and multipath effect is serious such as indoor, the method will lose efficacy.

In addition, existing most of technology mainly for be the high-precision sensor units such as Inertial Measurement Unit (IMU).But the price of such sensor unit is higher, there is larger difficulty for popular popularization and application.Along with the development of intelligent terminal, increasing sensor is dissolved among smart mobile phone, comprises accelerometer, gyroscope, magnetometer and high-definition camera first-class, and this is provide strong support based on pedestrian's location technology of intelligent mobile phone platform.Wherein, inertial sensor signal can be used for analyzing the movable information of pedestrian; MARG sensor array (comprising magnetometer, accelerometer and gyroscope) is that the attitude information of catching sensor in real time provides help, and then contributes to the attitude analyzing pedestrian; High-definition camera record continuous figurepicture, can utilize optical flow algorithm to calculate the displacement information of pedestrian.Therefore, invent a kind of based on intelligent mobile terminal equipment, there is pedestrian's step-size estimation device that is high-precision, real-time, the pervasive ability of indoor and outdoor surroundings, there is its practical significance.

Summary of the invention

For defect of the prior art, the object of this invention is to provide a kind of ground texture be adapted under various indoor and outdoor surroundings, anti-interference and adaptive ability is given prominence to, and has the pedestrian's step-size estimation method based on intelligent mobile terminal equipment of estimated capacity during high-precision real.

According to a kind of pedestrian's step-size estimation method based on mobile terminal provided by the invention, it is characterized in that, comprising:

S1: step-size estimation core processing unit (113) is according to from the heat transfer agent of acceleration transducer, gyroscope, magnetometer collection with from for obtaining figurethe Optic flow information of the camera collection of picture calculates and stores estimation step-length instantly, and preserves step-size estimation model according to the estimation step size computation stored;

S2: o controller (114) judges whether described step-size estimation model restrains, if convergence, then o controller (114) enters S3 after described estimation step-length instantly being exported, if do not restrain, then enters S6;

S3, collecting sensor signal device (111) only gathers the heat transfer agent of acceleration transducer;

S4, step-size estimation core processing unit (113) calculates new estimation step-length according to the heat transfer agent of described acceleration transducer and described step-size estimation model;

S5, described estimation step-length newly exports by o controller (114), returns S3;

S6, carries out abnormal step-length filtering process to described estimation step-length instantly:

If described estimation step-length is instantly abnormity point, then reject this abnormity point, and described step-size estimation model is reduced to the last result calculated;

If described estimation step-length is not instantly abnormity point, then described estimation step-length is instantly exported;

Then S1 is returned.

As prioritization scheme, described in the core processing unit of step-size estimation described in step S1 (113) calculates, the process of estimation step-length instantly comprises:

Information capture device (112) obtains pedestrian movement's information according to the transducing signal of acceleration transducer, and reach described step-size estimation core processing unit (113), estimation step-length instantly described in described step-size estimation core processing unit (113) calculates according to the heat transfer agent, described pedestrian movement's information and the described Optic flow information that gather from gyroscope, magnetometer;

Described pedestrian movement's packets of information contains meter step information and cadence information,

Wherein, described meter step information obtains according to the amplitude signal of the acceleration obtained at meter beans-and bullets shooter, the amplitude signal of described acceleration: described a _x, a _y, a _zrepresent the acceleration of acceleration in space in three-dimensional.

As prioritization scheme, in step sl, each described Optic flow information comprises corresponding with a timestamp figurepicture; Should figurelight stream vectors (u, v) is comprised in picture;

Information capture device (112) obtains sensor attitude information according to the heat transfer agent of gyroscope, magnetometer.

As prioritization scheme, described step-size estimation core processing unit (113) comprises ground light stream Chooser, displacement trasducer, sum-average arithmetic device and step-length model generation and storer;

Described step S1 comprises further:

S11, the described ground light stream Chooser disturbing factor filtered in described Optic flow information obtains the Optic flow information of denoising, and is each light stream vectors coupling adaptive weighting in the Optic flow information of described denoising wherein, p represents pixel sequence number; v _tp () represents light stream vectors, represent figurethe light stream vectors on picture top;

S12, described displacement trasducer calculates acquisition displacement information according to the Optic flow information of described sensor attitude information and denoising, and institute's displacement information comprises the motion vector (r, s) corresponding with timestamp, and the motion vector that a timestamp obtains is:

$\left\{\begin{array}{c}{r}_t\left(p\right)=\frac{f_y\·h}{f_x(y-c_y)\sin \mathrm{\θ}+f_x{f}_y\cos \mathrm{\θ}}\·u_t\left(p\right)\\ s_t\left(p\right)=\frac{f_y\·h}{{[(y-c_y)\sin \mathrm{\θ}+f_y\cos \mathrm{\θ}]}^2}\·v_t\left(p\right)\end{array}\right.,$

Wherein, p represents pixel sequence number; H represents camera height overhead; Y represents the row-coordinate of pixel; θ represents the angle on camera plane and ground; f _xand f _yrepresent the focal length of camera, c _xand c _yrepresent the principal point of camera;

S13, sum-average arithmetic device is averaging after each displacement information weighting summation in institute's displacement information, obtains each figurethe estimation displacement sample of picture;

S14, step-length model generation and storer calculate according to timestamp, described estimation displacement sample and described pedestrian movement's information, preserve and estimation step-length instantly described in exporting.

As prioritization scheme, the process that the core processing unit of step-size estimation described in step S4 (113) calculates described estimation step-length newly comprises:

Information capture device (112) obtains pedestrian movement's information according to the transducing signal of acceleration transducer, and reaches described step-size estimation core processing unit (113);

Described pedestrian movement's packets of information is containing meter step information and cadence information, and described step-size estimation core processing unit (113) calculates described estimation step-length newly according to described cadence information and described step-size estimation model;

Wherein, described meter step information obtains according to the amplitude signal of the acceleration obtained at meter beans-and bullets shooter, and the amplitude signal of described acceleration is: described a _x, a _y, a _zrepresent the acceleration of acceleration in space in three-dimensional.

As prioritization scheme, in described step S1 according to the estimation step size computation stored and the process of preserving step-size estimation model be specially: the estimation step-length sample point corresponding according to the estimation step-length stored carries out curve fitting, obtain a matched curve, the expression formula of preserving this matched curve is described step-size estimation model.

As prioritization scheme, abnormal step-length filtering process described in step S6 judges whether to export and specifically comprises further:

A. the first decision threshold Tr1 and the second decision threshold Tr2 is preset;

B. the estimation step-length sample point that described estimation step-length correspondence one is newly new, calculate the root-mean-square error between this new estimation step-length sample point and described matched curve, judge whether to be less than the first decision threshold Tr1, if, then export the described estimation step-length newly that this sample point is corresponding, if not, then next step is entered;

C. the distance between described new estimation step-length sample point and described matched curve is calculated, judge whether to be less than the second decision threshold Tr2, if, then export the described estimation step-length newly that this sample point is corresponding, calculate and upgrade the maximal value M in the distance of preserving between all estimation step-length sample points and described matched curve, if not, then reject this estimation step-length sample point and described step-size estimation model be reduced to the last result calculated;

D. upgrading the second decision threshold Tr2 is M/ α, wherein preset constant α >1.

Compared with prior art, the present invention has following beneficial effect:

The present invention is based on light stream algorithm for estimating, final realization is based on the real-time step-size estimation of high-precision pedestrian of intelligent mobile terminal equipment, be adapted to the ground texture under various indoor and outdoor surroundings, anti-interference and adaptive ability is given prominence to, for the dead reckoning in pedestrian's positioning system provides effective help.The present invention passes through the step-size estimation model of optimization and adjustment System, improves the precision of step-size estimation; By the Dynamic controlling of feedback signal, effectively reduce the acquisition and processing of redundant signals, reduce system power dissipation.

Accompanying drawing explanation

In order to be illustrated more clearly in the technical scheme of the embodiment of the present invention, to use required in describing embodiment below accompanying drawingbe briefly described, obviously, in the following describes accompanying drawingonly some embodiments of the present invention, for those skilled in the art, under the prerequisite not paying creative work, can also according to these accompanying drawingobtain other accompanying drawing. in accompanying drawing:

fig. 1it is the system frame of pedestrian's step-size estimation device in embodiment figure;

fig. 2it is the principle of work signal of collecting sensor signal device in embodiment figure;

fig. 3it is the principle of work signal of information capture device in embodiment figure;

fig. 4it is the principle of work signal of step-size estimation core processing unit in embodiment figure;

fig. 5it is the principle of work signal of o controller in embodiment figure;

fig. 6that in embodiment, pedestrian's step-size estimation device uses sight signal figure;

fig. 7it is step-size estimation model optimization effect in embodiment figure;

in figure110-intelligent mobile terminal equipment, 111-collecting sensor signal device, 112-information capture device, 113-step-size estimation core processing unit, 114-o controller, 61-camera.

Embodiment

Hereafter combine accompanying drawingin the mode of specific embodiment, the present invention is described in detail.Following examples will contribute to those skilled in the art and understand the present invention further, but not limit the present invention in any form.It should be pointed out that the embodiment that can also use other, or the amendment on 26S Proteasome Structure and Function is carried out to the embodiment enumerated herein, and can not depart from the scope and spirit of the present invention.

as Fig. 1shown system architecture frame figuremobile terminal in the embodiment of a kind of pedestrian's step-size estimation method based on intelligent mobile terminal equipment provided by the invention can be panel computer, smart mobile phone, and comprises the intelligent mobile terminal equipment comprising collecting sensor signal device (111), information capture device (112), step-size estimation core processing unit (113) and o controller (114).

Pedestrian's step-size estimation device described in the present embodiment is integrated in smart mobile phone, but the present invention is not limited thereto, and also can be integrated in other intelligent electronic device, or as independent a kind of step-size estimation device.

Described collecting sensor signal device (111) wherein, if for the convergence of step-size estimation model, then only gathers and exports the transducing signal of camera, if step-size estimation model is not restrained, then gathers and exports the transducing signal of all the sensors.Described sensor comprises: acceleration transducer, gyroscope, magnetometer and for obtain figurethe camera of picture.Described in described information capture device (112) basis figurepicture calculates and obtains Optic flow information.

Described information capture device (112), if for described step-size estimation model convergence, then calculate according to the transducing signal received from described collecting sensor signal device (111) and send pedestrian movement's information to step-size estimation core processing unit (113);

If described step-size estimation model is not restrained, then calculate according to the described transducing signal received from described collecting sensor signal device (111) and send pedestrian movement's information, Optic flow information and sensor attitude information to step-size estimation core processing unit (113).

Described pedestrian movement's packets of information is containing meter step information and cadence information.

Described step-size estimation core processing unit (113), if do not restrain for described step-size estimation model, then calculates according to described pedestrian movement's information, Optic flow information and sensor attitude information and exports step-size estimation model and estimation step-length instantly;

If described step-size estimation model convergence, then only calculate and export new estimation step-length.

Described o controller (114), for judging whether the step-size estimation model received from described step-size estimation core processing unit (113) restrains,

If do not restrain, then abnormal step-length filtering process is carried out to the estimation step-length instantly received from described step-size estimation core processing unit (113) and judge whether to export;

If convergence, then export described estimation step-length newly.

Step-size estimation core processing unit (113) wherein controls for the core algorithm realizing pedestrian's step-size estimation, as Fig. 4shown in, step-size estimation core processing unit (113) comprising:

(1) ground light stream Chooser, obtains the Optic flow information after denoising for filtering disturbing factor in described Optic flow information, and mates an adaptive weighting for the Optic flow information after each denoising.Realize camera collection thus figurethe rejecting of the interference such as leg and pin in picture.

(2) displacement trasducer, obtains displacement information sample for calculating according to the Optic flow information after described sensor attitude information and denoising.Realize thus by the conversion of light stream to actual displacement.

(3) sum-average arithmetic device, for being averaging after the weighted sum of each displacement message sample, obtaining and estimating displacement sample.Realize thus estimating that the output of displacement sample and Gaussian noise suppress.

(4) step-length model generator and storer, for:

When described step-size estimation model is not restrained, according to described estimation displacement sample and pedestrian movement's information acquisition estimation step-length instantly, obtain step-size estimation model according to described estimation step-length instantly, estimation step-length instantly described in output and step-size estimation model; When the convergence of described step-size estimation model, only export new estimation step-length according to described pedestrian movement's information acquisition.When model is not restrained, utilization stores and constantly updates the estimation step-length sample point added, constantly Optimal Step Size estimation model, until obtain the step-size estimation model of convergence, and the step-size estimation model that preservation generates is in real time.

Described o controller (114) is further used for: judge whether described step-size estimation model restrains, and judged result is fed back to described transducing signal collector (111).Described transducing signal collector (111) further in order to: if described step-size estimation model convergence, only gather the transducing signal of acceleration transducer; If described step-size estimation model is not restrained, gather the transducing signal of all the sensors.Using model generation and optimizing phase as the training stage, then as Fig. 5shown in, o controller (114) specifically comprises abnormal step-length wave filter and step-length model detector further.Abnormal step-length wave filter is used for carrying out abnormal step-length filtering process to estimation step-length instantly, realizes the elimination of exceptional sample point, improves accuracy and the reliability of the estimation model generated in training.Step-length model detector is used for judgement and whether completes training, namely for judging the convergence of step-size estimation model:

When step-length model detector detects that step-size estimation model is restrained, then to described abnormal step-length filter feedback, abnormal step-length wave filter directly exports described estimation step-length after receiving the feedback signal of model convergence, step-length model detector terminates the feedback signal of training to the transmission of described collecting sensor signal device (111) simultaneously, and described collecting sensor signal device (111) only gathers the heat transfer agent of acceleration transducer according to the feedback signal that this model is restrained;

When step-length model detector detects that step-size estimation model is not restrained, then to described abnormal step-length filter feedback, abnormal step-length wave filter carries out abnormal step-length filtering process to described estimation step-length instantly after receiving the feedback signal that model do not restrain, and continues the feedback signal of training to the transmission of described collecting sensor signal device (111).

After the convergence of step-size estimation model being detected, namely the training stage stops, and the follow-up step-size estimation model of this convergence that can directly adopt calculates estimation step-length.Because without the need to material calculation estimation model again, the transducing signal collection of acceleration transducer, gyroscope, magnetometer just there is no need, o controller (114) in the present embodiment achieves the control to different sensors signals collecting, thus avoid redundant signals Acquire and process, accelerate to estimate that step-length exports, reduce system power dissipation.Described collecting sensor signal device (111) is further used for: when receiving the feedback signal of described end training, only gather the transducing signal of acceleration transducer; When receiving the feedback signal of described continuation training, gather the transducing signal of all the sensors.

fig. 1give the frame of described real-time step-size estimation device (110) figure.Whole step-size estimation device is divided into several module.Disparate modules connects each other, define a closed-loop system, utilize unified interface to interconnect and transmit the input and output of current block, carried out the state transition of intermodule by decision-feedback between modules, this is conducive to the robustness and the degree of accuracy that improve whole system, and this is also an advantage of pedestrian's step-size estimation device provided by the invention.

From fig. 2can find out that the principle of work of the collecting sensor signal device (111) in described pedestrian's step-size estimation device is the sensor signal gathering smart mobile phone inside, and pre-service is carried out to the signal collected, use for information capture device.This module carries out signals collecting with adjustable sampling rate each sensor to smart mobile phone, and the training stage gathers the signal of four kinds of sensors simultaneously, after having trained, under the allotment of feedback control signal, gathers single-sensor signal.Wherein, MARG sensor array signal, i.e. accelerometer, gyroscope and magnetometer signals, all gather three axis signals, and the pre-service of this signal comprises: calibration, interpolation and filtering; Ground is taken in camera real time record walking process figurepicture, and according to unified timestamp information, right figurepicture carries out frame number screening and pixel compression, to guarantee satisfied real-time calculation requirement.Finally, this signal is exported to information capture device by appointment unit and call format.

fig. 3give the function of the information capture device (112) in described pedestrian's step-size estimation device, it is characterized in that, according to the sensor signal inputted by signal picker, utilize signal processing algorithm to catch movable information, attitude information and Optic flow information needed for pedestrian's step-size estimation, and information is gathered for information matrix outputs in the core processing unit of step-size estimation.

Movable information comprises: meter step information and cadence f _kinformation, its computing method are shown below:

f _k＝1/(t _k-t _k-1) (1)

Wherein, t _krepresent the timestamp of kth step.

Attitude information, i.e. attitude angle information, comprising: crab angle, the angle of pitch and roll angle.In actual use procedure, because smart mobile phone is laterally held in hand, therefore only need calculate roll angle θ, the angle namely between mobile phone plane and ground.What the calculating of attitude angle adopted is gradient descent algorithm, and attitude information hypercomplex number represented converts to and represents by Eulerian angle.

Optic flow information is continuous two width figurechange in displacement information between picture, is embodied in the bivector (u, v) on each pixel.The amplitude of vector represents the size of displacement, in units of pixel; The direction of vector is the direction of displacement, in noiseless situation, should be consistent with the direction that pedestrian advances.

fig. 4show the functional block of the step-size estimation core processing unit in described pedestrian's step-size estimation device figure.It is characterized in that the pedestrian's attitude information utilizing information capture device to calculate and light stream vectors information obtain step-size estimation sample by ground light stream Chooser, displacement trasducer and sum-average arithmetic device; Then, generation step-size estimation model is trained by pedestrian movement's information and corresponding step-size estimation sample.

Ground light stream Chooser is used for removing the impact of the interference such as leg and pin, makes to remain light stream vectors and really can reflect the change in displacement of ground pixel.In actual use procedure, when the angle on smart mobile phone and ground is less than 15 degree, shooting figurejust there will be leg and the pin of pedestrian in picture, due in the method that uses in displacement trasducer, default height is the distance of smart mobile phone to ground, therefore the appearance of leg and pin can cause estimating that displacement sample diminishes.But leg and pin always appear at figurethe bottom of picture, so this device utilizes this feature, gets figurethe top light stream vectors of picture, as with reference to vector, utilizes the adaptive weighting W based on gaussian kernel function _tp () controls the contribution of all light stream vectors in sum-average arithmetic device effectively, shown in formula specific as follows:

$W_t\left(p\right)=\exp \left(\frac{-{\left|\right|v_t\left(p\right)-{\stackrel{\‾}{v}}_t^u\left(p\right)\left|\right|}^2}{2{\mathrm{\σ}}^2}\right)$

Wherein, p represents pixel sequence number; v _tp () represents light stream vectors, represent figurethe light stream vectors on picture top, σ represents the width parameter of gaussian kernel function.By to weight setting thresholding, just can effectively remove figurein picture, the light stream vectors of leg and foot section, namely selects the light stream vectors of above ground portion, ensure that the reliability of system.

Displacement trasducer is used for light stream vectors (u, v) to convert to the actual displacement (r, s) of pedestrian, and computing method are shown below:

$\left\{\begin{array}{c}{r}_t\left(p\right)=\frac{f_y\·h}{f_x(y-c_y)\sin \mathrm{\θ}+f_x{f}_y\cos \mathrm{\θ}}\·u_t\left(p\right)\\ s_t\left(p\right)=\frac{f_y\·h}{{[(y-c_y)\sin \mathrm{\θ}+f_y\cos \mathrm{\θ}]}^2}\·v_t\left(p\right)\end{array}\right.---\left(2\right)$

Wherein, p represents pixel sequence number; H represents the height of smart mobile phone to ground; Y represents the row-coordinate of pixel; θ represents the angle on smart mobile phone plane and ground; f _xand f _yrepresent the focal length of smart mobile phone camera, c _xand c _yrepresent the principal point of smart mobile phone camera.In the present embodiment, pedestrian's step-size estimation device uses sight signal figure is as Fig. 6shown in.

Sum-average arithmetic device is then the adaptive weighting W exported based on ground light stream Chooser _tp (), will figurethe all displacements calculated in picture are weighted sues for peace and is averaged, and just obtains every two width figuredisplacement between picture, then walks the timestamp information obtained, this step experienced according to meter figureimage displacement is sued for peace, and just obtains step-size estimation sample.

The step-size estimation sample that step-length model generator utilizes movable information and obtains, generate step-size estimation model by the analytical approach of linear regression, model equation is shown below:

L＝a·f+b (3)

Wherein, L represents estimation step-length; F represents cadence; A and b is model parameter to be estimated.From formula (3), in theory, just can simulate this model equation by two sample points, and along with the increase of sample point, the precision of estimation model can improve thereupon.Therefore, the present invention can realize often walking generation one at training process and estimate step-length sample point, achieving training process efficiently, also providing powerful support for realizing high-precision real-time step-size estimation.Step-length pattern memory is after model training completes, the step-size estimation model parameter of generation is preserved, when not having model modification instruction (namely not having Optic flow information and attitude information input), the model in storer is automatically utilized to carry out step-size estimation.It should be noted that, in the training stage, this device exports estimation step-length instantly and step-size estimation model simultaneously; And after the training stage completes, this device only exports new estimation step-length.

fig. 5show the principle of work frame of (113) of o controller in described pedestrian's step-size estimation device figure.In the training stage, abnormal step-length detection is carried out to the estimation step-length instantly that step-size estimation core cell exports, meet the requirement of model restrictive condition, then this estimation step-length is instantly exported as estimation step-length, otherwise this estimation step-length instantly of filtering, and step-length model is reduced to last result of calculation.Described abnormal step-length filtering process comprises:

A. the first decision threshold Tr1 and the second decision threshold Tr2 is preset;

B. estimation step-length instantly described in utilizing carries out fitting a straight line as estimation step-length sample point, obtains a fitting a straight line, as Fig. 7shown in be step-size estimation model optimization effect in the present embodiment figure;

C. calculate the root-mean-square error estimated between step-length sample point and described fitting a straight line, judge whether to be less than the first decision threshold Tr1, if so, then export model corresponding to this fitting a straight line, if not, then enter next step;

D. the distance estimated between step-length sample point and described fitting a straight line is calculated, judge whether to be less than the second decision threshold Tr2, if, then retain this sample point, calculate and upgrade maximal value in the distance of preserving between all estimation step-length sample points and described fitting a straight line if not, then remove this estimation step-length sample point and reduce step-length estimation model to last result of calculation;

E. upgrading the second decision threshold Tr2 is M/ α, wherein preset constant α >1;

F. step b to e is repeated, until the convergence of described step-size estimation model.

The convergence of step-length model is detected, if model is restrained, represent that training process terminates, preserve the model parameter generated, to the feedback signal of collecting sensor signal device (111) end of output training, collecting sensor signal device (111) receives the signals collecting stopping gyroscope, magnetometer and camera sensing device after this terminates the feedback signal of training, and the transducing signal and the existing step-size estimation model that only gather acceleration transducer carry out step-size estimation.This device passes through the step-size estimation model of optimization and adjustment System, improves the precision of step-size estimation; By the Dynamic controlling of feedback signal, effectively reduce the acquisition and processing of redundant signals, reduce system power dissipation.

As from the foregoing, the invention provides a kind of intelligent mobile terminal equipment with the real-time step-size estimation ability of high precision pedestrian, comprising: collecting sensor signal device, information capture device, step-size estimation core processing unit and o controller.Wherein the different sensors signal gathering smart mobile phone is responsible for by collecting sensor signal device, and carries out pre-service to the signal gathered; Information capture device is responsible for calculating from sensor signal obtaining pedestrian movement's information, pedestrian's attitude information and Optic flow information; The core algorithm that step-size estimation core processing unit is responsible for whole device realizes, and comprises choosing of ground light stream, is used for suppressing figurethe interference of leg and pin in picture, displacement trasducer converts light stream vectors to actual displacement vector, sum-average arithmetic device is used for suppressing Gaussian noise, step-length model generator and storer utilize step-size estimation regression to generate step-size estimation model, and store the model parameter after having trained, for real-time step-size estimation; O controller is responsible for carrying out filtering to abnormal step-length and being optimized step-size estimation model and detecting, and according to model convergence situation feedback control signal, guiding sensor signal picker carries out self-adapting signal selection, thus improves device counting yield, reduces device power consumption.

Based on same inventive concept, the present embodiment also provides a kind of pedestrian's step-size estimation method based on mobile terminal, it is characterized in that, comprising:

S1: step-size estimation core processing unit (113) is according to from the heat transfer agent of acceleration transducer, gyroscope, magnetometer collection with from for obtaining figurethe Optic flow information of the camera collection of picture calculates and stores estimation step-length instantly, and preserves step-size estimation model according to the estimation step size computation stored;

S2: o controller (114) judges whether described step-size estimation model restrains, if convergence, then o controller (114) enters S3 after described estimation step-length instantly being exported, if do not restrain, then enters S6;

S3, collecting sensor signal device (111) only gathers the heat transfer agent of acceleration transducer;

S4, step-size estimation core processing unit (113) calculates new estimation step-length according to the heat transfer agent of described acceleration transducer and described step-size estimation model;

S5, described estimation step-length newly exports by o controller (114), returns S3;

S6, carries out abnormal step-length filtering process to described estimation step-length instantly:

If described estimation step-length is instantly abnormity point, then reject this abnormity point, and described step-size estimation model is reduced to the last result calculated;

If described estimation step-length is not instantly abnormity point, then described estimation step-length is instantly exported;

Then S1 is returned.

As prioritization scheme, described in the core processing unit of step-size estimation described in step S1 (113) calculates, the process of estimation step-length instantly comprises:

Information capture device (112) obtains pedestrian movement's information according to the transducing signal of acceleration transducer, and reach described step-size estimation core processing unit (113), estimation step-length instantly described in described step-size estimation core processing unit (113) calculates according to the heat transfer agent, described pedestrian movement's information and the described Optic flow information that gather from gyroscope, magnetometer;

Described pedestrian movement's packets of information contains meter step information and cadence information,

Wherein, described meter step information obtains according to the amplitude signal of the acceleration obtained at meter beans-and bullets shooter, the amplitude signal of described acceleration: described a _x, a _y, a _zrepresent the acceleration of acceleration in space in three-dimensional.

The defining method of described meter beans-and bullets shooter is the cross zero detecting method of dynamic threshold.

In step sl, each described Optic flow information comprises corresponding with a timestamp figurepicture; Should figurelight stream vectors (u, v) is comprised in picture;

Information capture device (112) obtains sensor attitude information according to the heat transfer agent of gyroscope, magnetometer.

Each from camera collection figurethe all corresponding timestamp of picture, time point time namely collected, and adjacent according to two figurepicture can calculate latter one figurein picture, pixel is relative to previous figurethe relative displacement of pixel in picture, the displacement of each pixel represents with a light stream vectors (u, v).Each figureas upper pixel all to there being a light stream vectors.

Described step-size estimation core processing unit (113) comprises ground light stream Chooser, displacement trasducer, sum-average arithmetic device and step-length model generation and storer;

Described step S1 comprises further:

S11, the described ground light stream Chooser disturbing factor filtered in described Optic flow information obtains the Optic flow information of denoising, and is each light stream vectors coupling adaptive weighting in the Optic flow information of described denoising wherein, p represents pixel sequence number; v _tp () represents light stream vectors, represent figurethe light stream vectors on picture top;

S12, described displacement trasducer calculates acquisition displacement information according to the Optic flow information of described sensor attitude information and denoising, and institute's displacement information comprises the motion vector (r, s) corresponding with timestamp, and the motion vector that a timestamp obtains is:

$\left\{\begin{array}{c}{r}_t\left(p\right)=\frac{f_y\·h}{f_x(y-c_y)\sin \mathrm{\θ}+f_x{f}_y\cos \mathrm{\θ}}\·u_t\left(p\right)\\ s_t\left(p\right)=\frac{f_y\·h}{{[(y-c_y)\sin \mathrm{\θ}+f_y\cos \mathrm{\θ}]}^2}\·v_t\left(p\right)\end{array}\right.,$

Wherein, p represents pixel sequence number; H represents camera height overhead; Y represents the row-coordinate of pixel; θ represents the angle on camera plane and ground; f _xand f _yrepresent the focal length of camera, c _xand c _yrepresent the principal point of camera;

S13, sum-average arithmetic device is averaging after each displacement information weighting summation in institute's displacement information, obtains each figurethe estimation displacement sample of picture;

S14, step-length model generation and storer can obtain displacement corresponding to each timestamp according to timestamp, described estimation displacement sample, real-time displacement at once in people's walking, according to the timestamp of the meter step information then known each meter beans-and bullets shooter in described pedestrian movement's information, the displacement that adjacent meter beans-and bullets shooter timestamp is corresponding is then estimation step-length instantly, just can calculate thus, preserves and estimation step-length instantly described in exporting according to timestamp, described estimation displacement sample and described pedestrian movement's information.

The process that the core processing unit of step-size estimation described in step S4 (113) calculates described estimation step-length newly comprises:

Information capture device (112) obtains pedestrian movement's information according to the transducing signal of acceleration transducer, and reaches described step-size estimation core processing unit (113);

Described pedestrian movement's packets of information is containing meter step information and cadence information, and described step-size estimation core processing unit (113) calculates described estimation step-length newly according to described cadence information and described step-size estimation model;

Wherein, described meter step information obtains according to the amplitude signal of the acceleration obtained at meter beans-and bullets shooter, and the amplitude signal of described acceleration is: described a _x, a _y, a _zrepresent the acceleration of acceleration in space in three-dimensional.

In described step S1 according to the estimation step size computation stored and the process of preserving step-size estimation model be specially: the estimation step-length sample point corresponding according to the estimation step-length stored carries out curve fitting, obtain a matched curve, the expression formula of preserving this matched curve is described step-size estimation model, wherein each described estimation step-length sample point ordinate is corresponding estimation step-length, and horizontal ordinate is corresponding cadence information.

Abnormal step-length filtering process described in step S6 judges whether to export and specifically comprises further:

A. the first decision threshold Tr1 and the second decision threshold Tr2 is preset;

B. the estimation step-length sample point that described estimation step-length correspondence one is newly new, calculate the root-mean-square error between this new estimation step-length sample point and described matched curve, judge whether to be less than the first decision threshold Tr1, if, then export the described estimation step-length newly that this sample point is corresponding, if not, then next step is entered;

C. the distance between described new estimation step-length sample point and described matched curve is calculated, judge whether to be less than the second decision threshold Tr2, if, then export the described estimation step-length newly that this sample point is corresponding, calculate and upgrade the maximal value M in the distance of preserving between all estimation step-length sample points and described matched curve, if not, then reject this estimation step-length sample point and described step-size estimation model be reduced to the last result calculated;

D. upgrading the second decision threshold Tr2 is M/ α, wherein preset constant α >1.

The concrete step-size estimation method of the present embodiment is as follows:

1. gather intelligent mobile phone sensor signal, comprising: accelerometer signal, gyroscope signal, magnetometer signals and figureimage signal, concrete mode of operation as Fig. 6shown in, the Distance geometry angle on smart mobile phone and ground is designated as h and θ respectively.

2. from sensor signal, obtain pedestrian movement's information, pedestrian's attitude information and Optic flow information.Pedestrian movement's information comprises meter step information and cadence information.In order to avoid the impact of mode on algorithm laid by different mobile phone, improve the practicality of system, the amplitude signal of what step-counting signal adopted is acceleration:

$a_{\mathrm{mag}}=\sqrt{a_x^2+a_y^2+a_z^2}---\left(4\right)$

Wherein a _x, a _y, a _zrepresent the acceleration of acceleration in space in three-dimensional.

The defining method of meter beans-and bullets shooter is the cross zero detecting method of dynamic threshold, and the sampled point namely meeting following three conditions is considered to count beans-and bullets shooter:

A. acceleration amplitude signal a _magdynamic threshold δ must be passed from negative to positive _th, be now designated as meter step candidate point;

B. the time interval between continuous two meter step candidate points must be positioned at interval [△ t _min, △ t _max];

C. the maximal value of acceleration amplitude signal between continuous two meter step candidate points and minimum value and dynamic threshold δ _thdifference must be positioned at interval [a _min, a _max].

Wherein, dynamic threshold δ _thget the average of the acceleration amplitude between continuous two meter step candidate points, computing method are as follows:

${\mathrm{\δ}}_{\mathrm{th}}=\frac{1}{\mathrm{\Δ}{t}_k}{\∫}_{t_k}^{t_{k+1}}{a}_{\mathrm{mag}}\left(t\right)\mathrm{dt}---\left(5\right)$

In formula (5), △ t _krepresent the time interval between two meter step candidate points.

And the calculating of cadence is as shown in formula (1).

The attitude acquisition algorithm that what the acquisition of pedestrian's attitude information adopted is based on gradient descent algorithm, according to fig. 6shown smart mobile phone holds mode, and roll angle (roll) is in figureangle θ.

In order to ensure the precision that displacement calculates, what the acquisition of Optic flow information adopted is dense optical flow algorithm, this algorithm pair figureeach pixel of picture all calculates optical flow velocity vector (u, v).

3. due to the interference containing pedestrian's leg and pin in the light stream vectors (u, v) that calculates in step 2, therefore this step utilizes the vectorial difference between light stream vectors to remove figureinterference sections in picture.

4. will in step 3, go the residual vector after disturbing to convert actual displacement to, computing method are as shown in formula (2), smart mobile phone camera inner parameter wherein utilizes the algorithm in camera calibrated tool box to obtain, and calibration parameter value used is respectively:

$\left\{\begin{array}{c}{f}_x=148.77228\\ f_y=217.0334\\ c_y=71.97305\end{array}\right.---\left(6\right)$

Then, cumulative summation is carried out in the displacement of time range between continuous two meter beans-and bullets shooters and just estimated step-length accordingly.

5. utilize in step 4 cadence calculated in the step-length sample and step 2 obtaining estimating to carry out linear regression fit, just obtain the step-size estimation model under the training of current sample point.

6. the step-length sample point in pair step 5 and the step-size estimation model obtained detect, by carrying out root-mean-square error constraint to the fitting degree of sample point and regression model, remove exceptional sample point to the impact of step-size estimation model, Optimized model parameter, concrete steps are as follows:

A. initialization decision threshold Tr1 and Tr2;

B. utilize and estimate that step-length carries out curve fitting to formula (3);

C. calculate the root-mean-square error estimated between step-length sample point and matched curve, and compare with thresholding Tr1, if be less than Tr1, output model parameter a and b, otherwise enter next step;

D. calculate the distance estimating that step-length sample point and matched curve are shown in, and compare with thresholding Tr2, if be less than Tr2, retain this sample point, otherwise remove this estimation step-length sample point and upgrade step-size estimation model, preserve the maximal value M in all distances;

E. upgrading thresholding Tr2 is M/ α, α >1;

F. step b to e is repeated, until convergence.

Fig. 7 is the fitting effect comparison diagram before and after outlier detection, and as can be seen from the figure, after removing exceptional sample point, models fitting successful improves.

7. through the training of about 10-15 step-length sample point, estimation model just can be restrained, and now, training completes, and only need gather acceleration signal, calculate cadence information, finally utilize the step-size estimation model trained directly to carry out real-time step-size estimation according to step 2.

The foregoing is only preferred embodiment of the present invention, those skilled in the art know, without departing from the spirit and scope of the present invention, can carry out various change or equivalent replacement to these characteristic sum embodiments.In addition, under the teachings of the present invention, can modify to adapt to concrete situation and material to these characteristic sum embodiments and can not the spirit and scope of the present invention be departed from.Therefore, the present invention is not by the restriction of specific embodiment disclosed herein, and the embodiment in the right of all the application of falling into all belongs to protection scope of the present invention.

Claims (7)

1., based on pedestrian's step-size estimation method of mobile terminal, it is characterized in that, comprising:

S1: step-size estimation core processing unit (113) calculates according to the heat transfer agent from acceleration transducer, gyroscope, magnetometer collection and the Optic flow information from the camera collection for obtaining image and stores estimation step-length instantly, and preserve step-size estimation model according to the estimation step size computation stored;

S2: o controller (114) judges whether described step-size estimation model restrains, if convergence, then o controller (114) enters S3 after described estimation step-length instantly being exported, if do not restrain, then enters S6;

S3, collecting sensor signal device (111) only gathers the heat transfer agent of acceleration transducer;

S4, step-size estimation core processing unit (113) calculates new estimation step-length according to the heat transfer agent of described acceleration transducer and described step-size estimation model;

S5, described estimation step-length newly exports by o controller (114), returns S3;

S6, carries out abnormal step-length filtering process to described estimation step-length instantly:

If described estimation step-length is instantly abnormity point, then reject this abnormity point, and described step-size estimation model is reduced to the last result calculated;

If described estimation step-length is not instantly abnormity point, then described estimation step-length is instantly exported;

Then S1 is returned.

2. method according to claim 1, is characterized in that, described in the core processing unit of step-size estimation described in step S1 (113) calculates, the process of estimation step-length instantly comprises:

Information capture device (112) obtains pedestrian movement's information according to the transducing signal of acceleration transducer, and reach described step-size estimation core processing unit (113), estimation step-length instantly described in described step-size estimation core processing unit (113) calculates according to the heat transfer agent, described pedestrian movement's information and the described Optic flow information that gather from gyroscope, magnetometer;

Described pedestrian movement's packets of information contains meter step information and cadence information,

Wherein, described meter step information obtains according to the amplitude signal of the acceleration obtained at meter beans-and bullets shooter, the amplitude signal of described acceleration: described a _x, a _y, a _zrepresent the acceleration of acceleration in space in three-dimensional.

3. method according to claim 2, is characterized in that, in step sl, each described Optic flow information comprises the image corresponding with a timestamp; Light stream vectors (u, v) is comprised in this image;

Information capture device (112) obtains sensor attitude information according to the heat transfer agent of gyroscope, magnetometer.

4. method according to claim 3, is characterized in that, described step-size estimation core processing unit (113) comprises ground light stream Chooser, displacement trasducer, sum-average arithmetic device and step-length model generation and storer;

Described step S1 comprises further:

S11, the described ground light stream Chooser disturbing factor filtered in described Optic flow information obtains the Optic flow information of denoising, and is each light stream vectors coupling adaptive weighting in the Optic flow information of described denoising wherein, p represents pixel sequence number; v _tp () represents light stream vectors, represent the light stream vectors on image top;

S12, described displacement trasducer calculates acquisition displacement information according to the Optic flow information of described sensor attitude information and denoising, and institute's displacement information comprises the motion vector (r, s) corresponding with timestamp, and the motion vector that a timestamp obtains is:

$\left\{\begin{array}{c}{r}_t\left(p\right)=\frac{f_y\·h}{f_x(y-c_y)\sin \mathrm{\θ}+f_x{f}_y\cos \mathrm{\θ}}\·u_t\left(p\right)\\ s_t\left(p\right)=\frac{f_y\·h}{[(y-c_y)\sin \mathrm{\θ}+f_y\cos \mathrm{\θ}{]}^2}\·v_t\left(p\right)\end{array}\right.,$

Wherein, p represents pixel sequence number; H represents camera height overhead; Y represents the row-coordinate of pixel; θ represents the angle on camera plane and ground; f _xand f _yrepresent the focal length of camera, c _xand c _yrepresent the principal point of camera;

S13, sum-average arithmetic device is averaging after each displacement information weighting summation in institute's displacement information, obtains the estimation displacement sample of each image;

S14, step-length model generation and storer calculate according to timestamp, described estimation displacement sample and described pedestrian movement's information, preserve and estimation step-length instantly described in exporting.

5. method according to claim 1, is characterized in that, the process that the core processing unit of step-size estimation described in step S4 (113) calculates described estimation step-length newly comprises:

Information capture device (112) obtains pedestrian movement's information according to the transducing signal of acceleration transducer, and reaches described step-size estimation core processing unit (113);

Described pedestrian movement's packets of information is containing meter step information and cadence information, and described step-size estimation core processing unit (113) calculates described estimation step-length newly according to described cadence information and described step-size estimation model;

Wherein, described meter step information obtains according to the amplitude signal of the acceleration obtained at meter beans-and bullets shooter, and the amplitude signal of described acceleration is: described a _x, a _y, a _zrepresent the acceleration of acceleration in space in three-dimensional.

6. method according to claim 1, it is characterized in that, in described step S1 according to the estimation step size computation stored and the process of preserving step-size estimation model be specially: the estimation step-length sample point corresponding according to the estimation step-length stored carries out curve fitting, obtain a matched curve, the expression formula of preserving this matched curve is described step-size estimation model.

7. method according to claim 6, is characterized in that, abnormal step-length filtering process described in step S6 judges whether to export and specifically comprises further:

A. the first decision threshold Tr1 and the second decision threshold Tr2 is preset;

B. the estimation step-length sample point that described estimation step-length correspondence one is newly new, calculate the root-mean-square error between this new estimation step-length sample point and described matched curve, judge whether to be less than the first decision threshold Tr1, if, then export the described estimation step-length newly that this sample point is corresponding, if not, then next step is entered;

C. the distance between described new estimation step-length sample point and described matched curve is calculated, judge whether to be less than the second decision threshold Tr2, if, then export the described estimation step-length newly that this sample point is corresponding, calculate and upgrade the maximal value M in the distance of preserving between all estimation step-length sample points and described matched curve, if not, then reject this estimation step-length sample point and described step-size estimation model be reduced to the last result calculated;

D. upgrading the second decision threshold Tr2 is M/ α, wherein preset constant α >1.