CN111127446B - Gait analysis-oriented plantar pressure image partitioning method - Google Patents
Gait analysis-oriented plantar pressure image partitioning method Download PDFInfo
- Publication number
- CN111127446B CN111127446B CN201911366465.8A CN201911366465A CN111127446B CN 111127446 B CN111127446 B CN 111127446B CN 201911366465 A CN201911366465 A CN 201911366465A CN 111127446 B CN111127446 B CN 111127446B
- Authority
- CN
- China
- Prior art keywords
- pressure
- sole
- plantar pressure
- model
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Abstract
The invention relates to a gait analysis-oriented plantar pressure image partitioning method, which comprises the following steps: obtaining plantar pressure data and carrying out early-stage processing; clustering the sole pressure data after the early-stage processing to obtain a sole pressure image of a footprint area and performing integral projection to obtain pressure centers of a heel and a sole; establishing a plantar pressure characteristic point positioning model, and positioning characteristic points of a plantar pressure image; partitioning the plantar pressure image according to the relative position relation between each area of the foot and the plantar pressure characteristic points, wherein the plantar pressure characteristic point positioning comprises two stages of model training and characteristic point searching; in the model training stage, a sufficient number of plantar pressure images are selected as a training set, the accurate positions of the characteristic points are marked manually, and a plantar pressure characteristic point positioning model is established; and in the characteristic point searching stage, a plantar pressure characteristic point positioning model is initialized according to the pressure centers of the heel and the sole, and then the accurate position of the plantar pressure image characteristic point is found through model searching.
Description
Technical Field
The invention relates to the field of gait analysis, in particular to a method for partitioning a plantar pressure image for gait analysis.
Background
When a human body stands still or walks dynamically, the sole of a foot can be subjected to the reaction force of the human body on the ground, wherein the reaction force vertical to the contact plane of the sole and the ground is the most obvious force, namely the pressure of the sole. When the foot structure of the human body is diseased or dysfunctional, the distribution of the plantar pressure can also be greatly changed. The modern medical means can assist doctors to diagnose the disease cause of sick feet of patients and appoint a rehabilitation treatment scheme by dividing the sole pressure according to anatomy and comparing the sole pressure distribution conditions of healthy feet and pathological feet with the pressure distribution conditions of different sole pressure areas of the same patient.
The existing plantar pressure subareas are semi-manual subareas which are manually divided by doctors or scientific researchers through subjective experiences or manually adjusted by software users after a fixed template is mechanically used for preliminarily dividing plantar pressure data by data acquisition software.
Therefore, there is a need for an efficient, low-cost method for plantar pressure partition that is easy to code.
Disclosure of Invention
The invention solves the problems that: the method overcomes the defects of the prior art, provides a gait analysis-oriented plantar pressure image partition method, automatically selects a footprint area from a frame in a plantar pressure image, partitions the plantar pressure image according to a characteristic point positioning method, and has the advantages of high efficiency, low cost and quick realization.
The technical scheme adopted by the invention is as follows: obtaining plantar pressure data and carrying out early-stage processing; clustering the sole pressure data after the early-stage processing to obtain a sole pressure image of a footprint area; performing integral projection on the pressure image of the sole to obtain the pressure centers of the heel and the sole; establishing a plantar pressure characteristic point positioning model, and positioning characteristic points of a plantar pressure image; and (4) partitioning the plantar pressure image according to the relative position relationship between each area of the foot and the plantar pressure characteristic points. The method comprises the following steps of (1) positioning plantar pressure characteristic points, wherein the plantar pressure characteristic point positioning comprises two stages of model training and characteristic point searching; in the model training stage, selecting a sufficient number of plantar pressure images as a training set, manually marking the accurate positions of the characteristic points in the training samples, establishing a plantar pressure characteristic point positioning model, and only performing the model training stage once; in the characteristic point searching stage, a plantar pressure characteristic point positioning model is initialized according to the pressure centers of the heel and the sole, and then the accurate position of the plantar pressure image characteristic point is found through model searching.
The concrete implementation is as follows:
step (1): obtaining plantar pressure data and carrying out early-stage processing, and specifically comprising the following substeps:
step (11): acquiring a plurality of adjacent frames (at least 3 frames) of plantar pressure data containing complete footprints in a time domain by adopting an array type pressure sensor;
step (12): performing time domain mean filtering on the middle frame by using the multi-frame plantar pressure data obtained in the step (11);
step (13): sequentially carrying out maximum filtering processing and interpolation processing on the intermediate frame data obtained in the step (12) to obtain sole pressure data with uniform size and equal row-column spacing;
step (2): clustering the sole pressure data after the preprocessing in the step (1), locating the position of the footprint in the sole pressure data, and obtaining a sole pressure image of the footprint area, wherein the method specifically comprises the following substeps:
step (21): clustering the sole pressure data obtained in the step (1) after the pre-treatment by using a DBSCAN clustering algorithm to obtain a plurality of pressure area blocks, wherein a footprint area may be composed of a plurality of pressure area blocks;
step (22): taking the minimum external moment central point of each pressure area block obtained in the step (21) as the center of the area block, and using a K-means clustering algorithm to aggregate all pressure area blocks contained in each footprint into a large sole pressure area block;
step (23): extracting the pressure data of the footprint area obtained in the step (22) to obtain a sole pressure image containing the complete footprint;
and (3): performing integral projection on the sole pressure image obtained in the step (2) to obtain the pressure centers of the heel and the sole, and specifically comprising the following substeps:
step (31): taking the long side direction of the minimum external torque of the plantar pressure image obtained in the step (2) as a longitudinal axis and the short side direction as a transverse axis, and performing integral projection on the plantar pressure image along the longitudinal axis to obtain a gray level histogram;
step (32): if the gray level histogram obtained in the step (31) contains a plurality of similar peak points, performing mean filtering on the gray level histogram by adopting a sliding window with the length of N;
step (33): because the sole is wider than the heel, the highest peak point of the filtered gray level histogram is the sole center line, and the next highest peak point is the heel center line;
step (34): respectively taking 1/6 row of the total row number of the pressure images of the soles from the upper side and the lower side of the longitudinal middle row of the soles to obtain a sole area, carrying out integral projection on the images of the sole area along a transverse axis to obtain a sole gray histogram, wherein the highest peak point of the sole gray histogram is a sole central column, the next highest peak point is a heel central column, the intersection point of the sole central row and the sole central column is a sole central point, and the intersection point of the heel central row and the heel central column is a heel central point;
and (4): establishing a plantar pressure characteristic point positioning model, and positioning characteristic points of a plantar pressure image; the method specifically comprises two stages of model training and feature point searching:
step (41): in the model training stage, selecting a sufficient number of plantar pressure images as a training set, manually marking the accurate positions of the characteristic points in the training samples, and establishing a plantar pressure characteristic point positioning model, wherein the model training stage is only executed once;
step (42): in the characteristic point searching stage, a plantar pressure characteristic point positioning model is initialized according to the pressure centers of the heel and the sole, and then the accurate position of the plantar pressure image characteristic point is found through an iterative model;
further, the processing method in the model training phase is as follows: selecting N sheets (N)>200 The pressure image of the sole obtained in the step (3) is used as a training set; manually marking n accurate plantar pressure images (n)>= 6) positions of characteristic points (the characteristic points are selected to comprise heels, sole central points and inflection points of sole pressure profiles, the characteristic points and all areas of feet have obvious relative position relations), coordinates after the characteristic points are sequenced are sequentially connected in series to form a shape vector, and a shape vector set obtained by N sole pressure images is marked as X u Wherein the shape vector obtained from the ith plantar pressure image is recorded as X ui ，X ui ＝(x i0 ,…,x i(n-1) ,y i0 ,…,y i(n-1) ) T I =0, \8230, N-1, where x ik 、y ik Respectively representing the horizontal and vertical coordinates of a characteristic point k in the ith plantar pressure training image, wherein k is more than or equal to 0 and less than n, and n is the number of the characteristic points; set of shape vectors X u As input for establishing the part of the plantar pressure characteristic point positioning model, the X set is formed according to the shape vectors u Establishing a plantar pressure characteristic point positioning model for searching plantar pressure characteristic points; the accurate position of the characteristic points in the manual marking training sample and the establishment of the plantar pressure characteristic point positioning model part are only executed once when the movable shape model is established;
furthermore, the plantar pressure feature point positioning model is a template-based feature point positioning method and can be an ASM model or an AAM model; firstly, the model can be processed by Procrustes normalization method, namely, by translation, rotation and scaling transformation operation, and the shape vector is collected on the basis of not changing the point distribution model u Aligning to the same shape vector X, wherein the shape vector X is a feature point template; then, constructing local features for each feature point; at this point, the feature point positioning model is constructed;
further, the method includes the step of collecting the shape vectorsX of the formula u Aligning to the same shape vector X comprises the following steps:
step (411): with a certain training sample X j As a shape reference
For other training samples X i Performing translation, rotation and scaling to make all samples as close to the reference shape as possible; training sample X i And shape reference>
Proximity between uses Euclidean distance definition->
Obtaining a transformed shape vector of X' i ＝M(s,θ)[X i ]-t, where s is the scaling scale, θ is the rotation angle, t is the translation vector;
step (412): calculating all transformed training samples X' u As a new shape reference
And calculates the current shape reference>
And the last shape reference->
Translation, rotation and zoom deviations between;
step (413): iteration steps (411) and (412), if the deviation is smaller than a specified threshold or the iteration exceeds a specified maximum iteration number, stopping the iteration; and taking the shape reference obtained at the last time as an average shape vector X after all the training samples are aligned.
Further, the specific operation of constructing the local feature for each feature point is: calculating the average local texture of the characteristic points i (i =0,1, \8230;, n-1) in the training sample
Sum variance S i (ii) a Firstly, on two sides of the ith characteristic point of the jth (j =0,1, \ 8230;, N-1) training sample, respectively selecting N pressure points along a direction perpendicular to a connecting line of the front characteristic point and the rear characteristic point of the point to form a vector with the length of 2n +1, and deriving a pressure value contained in the vector to obtain a local texture g ij The same operation is performed on all samples in the training set to obtain N local textures of the ith feature point, and the mean value is calculated>
And variance->
Further, the characteristic point searching stage comprises the steps of initializing a model according to the pressure centers of the heel and the sole, and finding out the accurate position of the characteristic point of the sole pressure image through an iterative model; the method specifically comprises the following substeps:
step (421): calculating the pressure centers (x) of heel and sole in the target plantar pressure image by taking the pressure centers of heel and sole as the reference 1 ,y 1 ) And (x) 2 ,y 2 ) And a shape reference
Center of pressure of middle heel and sole (x' 1 ,y' 1 ) And (x' 2 ,y' 2 ) A translation, rotation, scaling aberration in between, for a shape reference>
Transforming to obtain the initial position X of the model c ；
Step (422): searching a new position of each feature point in the plantar pressure image; first, the initial position X of the active shape model is set c Covering on the plantar pressure image, for the ith characteristic point in the model, taking the ith characteristic point as the center in the direction vertical to the connecting line of the front characteristic point and the rear characteristic point, and respectively selecting m (m) at two sides>n) pressure points are arranged on the base plate,a search neighborhood with the length of 2m +1 is formed by adding the characteristic point i; sliding windows with the length of n pressure points in the search neighborhood, and calculating the local texture g of each window i And calculating the local texture and the average texture
The mahalanobis distance between them, such that the center point of the window with the smallest mahalanobis distance serves as the new position of the feature point i.
A step (423): calculating translation, rotation and scaling parameters of the shape vector of the initial position of the model and the shape vector after updating the position;
step (424): repeating steps (422) and (423) to calculate a new shape vector X new And the original shape vector X c Distance D of x ，
Wherein x is newi 、y newi Are respectively a new shape vector X new The abscissa and ordinate, x, of the ith feature point of (1) ci 、y ci Are respectively the original shape vector X c The abscissa and the ordinate of the ith feature point of (1). If it is
Or the circulation reaches the maximum times, the search is completed, and the accurate position of each feature point in the plantar pressure image is obtained;
and (5): and (4) according to the relative position relation between each area of the foot and the characteristic points of the sole pressure, combining the characteristic points of the sole pressure image obtained in the step (4), and partitioning the sole pressure image.
The invention has the following beneficial effects:
(1) The plantar pressure partition method can remove contact noise, adhesion noise, network signal noise and acquisition circuit noise of original plantar pressure data, and automatically extracts frames from a plantar pressure image to select a footprint area through twice clustering and a minimum external moment algorithm;
(2) The invention adopts a multiple integral projection method to detect the sole central point and the heel central point, ensures the reliable positioning of the heel center and the sole central point, and simultaneously improves the accuracy of model initialization in the plantar pressure characteristic point searching stage, thereby improving the accuracy of the positioning of the plantar pressure characteristic points and the accuracy of plantar pressure image partition;
(3) The plantar pressure partitioning method selects points with distinct characteristics such as a heel central point, a sole central point, an inflection point of a plantar pressure contour and the like as characteristic points of a plantar pressure image, is favorable for building a plantar pressure characteristic point positioning model according to a training sample set, has obvious relative position relation between the selected characteristic points and each area of the foot, and is also favorable for partitioning the plantar pressure image according to the characteristic points;
(4) The plantar pressure zoning method can obtain accurate plantar pressure zoning results by accurately positioning the characteristic points of the plantar pressure image and then carrying out the zoning on the plantar pressure image according to the relative position relationship between the characteristic points and the areas of the foot;
(5) The model training stage of the plantar pressure characteristic point positioning model adopted by the invention only needs to be executed once, and the trained model can be directly used in the plantar pressure characteristic point searching stage, thereby ensuring that the plantar pressure partitioning method can be executed efficiently.
Drawings
In order to more clearly illustrate the technical solution of the method of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the description below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is a schematic view of an integral projection of a plantar pressure image;
FIG. 3 is a block diagram illustrating a sub-process of the plantar pressure feature point locating portion according to an embodiment of the present invention;
FIG. 4 is a schematic view of the selection of plantar pressure feature points in an embodiment of the present invention;
FIG. 5 shows an initial position X of the active shape model in an embodiment of the present invention c An effect map overlaid on the pressure image of the sole of the foot;
fig. 6 is a schematic diagram illustrating the result of partitioning the pressure image of the sole of a foot according to the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.
Example 1
As shown in fig. 1, the core of the present invention is to provide a method for determining pressure image of sole facing gait analysis, which comprises the following steps:
step (1): obtaining plantar pressure data and carrying out early-stage processing, and specifically comprising the following substeps:
step (11): acquiring a plurality of adjacent frames (at least 3 frames) of plantar pressure data containing complete footprints in a time domain by adopting an array type pressure sensor;
step (12): performing time domain mean filtering on the middle frame by using the multiple frames of plantar pressure data obtained in the step (11);
step (13): sequentially carrying out maximum filtering processing and interpolation processing on the intermediate frame data obtained in the step (12) to obtain sole pressure data with uniform size and equal row-column spacing;
step (2): clustering the sole pressure data after the preprocessing in the step (1), locating the position of the footprint in the sole pressure data, and obtaining a sole pressure image of the footprint area, specifically comprising the following substeps:
step (21): clustering the sole pressure data obtained in the step (1) after the pre-treatment by using a DBSCAN clustering algorithm to obtain a plurality of pressure area blocks, wherein a footprint area may be composed of a plurality of pressure area blocks;
step (22): using the minimum external moment center point of each pressure area block obtained in the step (21) as the center of the area block, and gathering the pressure area blocks into a footprint area by using a K-means clustering algorithm;
step (23): extracting the pressure data of the footprint area obtained in the step (22) to obtain a sole pressure image containing the complete footprint;
and (3): according to the method shown in fig. 2, integral projection is carried out on the pressure image of the sole obtained in the step (2) to obtain the pressure center of the heel and the sole, and the method specifically comprises the following sub-steps:
step (31): taking the long side direction of the minimum external torque of the plantar pressure image obtained in the step (2) as a longitudinal axis and the short side direction as a transverse axis, and performing integral projection on the plantar pressure image along the longitudinal axis to obtain a gray level histogram;
step (32): if the gray level histogram obtained in the step (31) contains a plurality of similar peak points, performing mean filtering on the gray level histogram by adopting a sliding window with the length of N;
step (33): because the sole is wider than the heel, the highest peak point of the filtered gray histogram is the sole center row, and the next highest peak point is the heel center row;
step (34): respectively taking 1/6 row of the total row number of the sole pressure images at the upper side and the lower side of the longitudinal middle row of the sole to obtain a sole area, and performing integral projection on the sole area images along a transverse axis to obtain a sole gray histogram, wherein the highest peak point of the sole gray histogram is a sole central column, the next highest peak point is a heel central column, the intersection point of the sole central row and the sole central column is a sole central point, and the intersection point of the heel central row and the heel central column is a heel central point;
and (4): establishing a plantar pressure characteristic point positioning model as shown in fig. 3, and positioning characteristic points of a plantar pressure image; the method specifically comprises two stages of model training and feature point searching;
step (41): in the model training stage, selecting N (N > 200) plantar pressure images as a training set, manually marking the accurate positions of characteristic points in a training sample, and establishing a plantar pressure characteristic point positioning model, wherein the model training stage is only executed once;
step (42): in the characteristic point searching stage, initializing a plantar pressure characteristic point positioning model according to the pressure centers of a heel and a sole, and then finding out the accurate position of a plantar pressure image characteristic point through an iterative model;
further, the model training phase processing method comprises: selecting N plantar pressure images obtained in the step (3) as a training set; manually marking n accurate plantar pressure images (n)>= 6) positions of feature points (the feature points are selected to include a heel, a sole center point, and an inflection point of a sole pressure profile, and there should be a clear relative position relationship between the feature points and each area of the foot), as shown in fig. 4, as an example, 17 feature points are selected in this embodiment, that is: f1 F2, \ 8230;, F16, F17, including heel center point F2 and ball center point F1; the coordinates after the characteristic points are sequenced are sequentially connected in series to form a shape vector, and a shape vector set obtained by N plantar pressure images is recorded as X u Wherein the shape vector obtained from the ith plantar pressure image is recorded as X ui ，X ui ＝(x i0 ,…,x i(n-1) ,y i0 ,…,y i(n-1) ) T I =0, \ 8230, N-1, where x ik 、y ik Respectively representing the kth characteristic point F in the ith plantar pressure training image k (x ik ,y ik ) K is more than or equal to 0 and less than n; set of shape vectors X u As input for establishing the part of the plantar pressure characteristic point positioning model, the X set is formed according to the shape vectors u Establishing a plantar pressure characteristic point positioning model for searching plantar pressure characteristic points; the accurate position of the characteristic points in the manual marking training sample and the establishment of the plantar pressure characteristic point positioning model part are only executed once when the movable shape model is established;
further, the plantar pressure characteristicThe feature point positioning model is a feature point positioning method based on a template, and preferably, a movable shape model is adopted in the embodiment; firstly, the model can be formed by a Procrustes normalization method, namely, a translation, rotation and scaling transformation operation, and a shape vector set X can be collected on the basis of not changing a point distribution model u Aligning to the same shape vector X, wherein the shape vector X is a feature point template; then, constructing local features for each feature point; at this point, the feature point positioning model is constructed;
further, the shape vector is collected into X u The specific steps of aligning to the same shape vector X are as follows:
step (ii) of
Selecting X u The first shape vector is taken as the initial mean shape vector->
Step (ii) of
Each shape vector X ui To an average shape vector>
And (4) alignment, wherein a transformation vector in the alignment process is marked as T = (scos theta, sin theta, T) x ,t y ) T Where s is the scaling, θ is the rotation angle, t x Is the x-axis translation vector, t y Is a y-axis translation vector, and the aligned shape vector is denoted as X' ui ；
Operation of alignment is X' ui ＝X ui And T, wherein the calculation method of the transformation vector T comprises the following steps:
wherein, the first and the second end of the pipe are connected with each other,
w is a weight matrix, and the calculation method comprises the following steps:
first, a feature point k (x) in the ith shape is calculated ik ,y ik )、l(x ik ,y ik ) The distance between:
then, a weighted value of the feature point k is calculated
Wherein the variance V Dkl For all N characteristic points k (x) in the plantar pressure image ik ,y ik )、l(x ik ,y ik ) Distance D ikl (i =0, \8230;, N-1); weighting, value w k Representing the stability degree of the characteristic point k; />
Finally, with w k Making a diagonal matrix W for the diagonal, wherein the diagonal matrix is a weight matrix;
step (ii) of
Updating all shape vectors X 'after alignment' ui Is recorded as &>
Step (ii) of
Repeat the step->
Outputting the aligned shape vector to be marked as X after convergence or maximum iteration times;
the convergence determination condition is: calculating a transformation vector T between the average shape vectors of the two times before and after, and converging if the conditions of | s-1| < 0.001, | theta | < 0.001 pi/180, and | T | < 0.01 are simultaneously met;
further, the specific operation of constructing the local feature for each feature point is: calculating the average local texture of the characteristic points i (i =0,1, \8230;, n-1) in the training sample
Sum variance S i (ii) a Firstly, on both sides of the ith characteristic point of the j (j =0,1, \8230;, N-1) training sample, h pressure points are respectively selected along the direction vertical to the connecting line of the front characteristic point and the rear characteristic point of the point to form a vector with the length of 2h +1, and a local texture g is obtained by deriving the pressure values contained in the vector ij The same operation is performed on all samples in the training set to obtain N local textures of the ith feature point, and the mean value is calculated>
And variance->
Further, the establishing of the active shape model includes PCA analysis, which includes the following steps:
step (PCA-1): calculating an average shape vector of the aligned N shape vectors
Step (PCA-2): computing covariance matrices for N shape vectors
Step (PCA-3): calculation assistant partyEigenvalues of the difference matrix lambda i And sorting from big to small, and marking the corresponding eigenvector as p i ,i＝0,1,…,2n-1；
Step (PCA-4): selecting the first k maximum eigenvalues, and forming a principal component analysis matrix P = (P) by using corresponding eigenvectors 0 ,p 1 ,…,p k-1 )；
Step (PCA-5): constructing a movable shape model as
Wherein, b is a k-dimensional shape parameter for controlling the shape change of the feature point; here b is restricted to +>
Further, the feature point searching stage includes the following steps: initializing the movable shape model according to the pressure centers of the heel and the sole, and finding out the accurate position of the characteristic point of the sole pressure image through the iterative model, wherein the specific steps are as follows:
step (421): initializing the movable shape model through affine transformation; the coordinates of the pressure centers of the heel and the sole in the target plantar pressure image are respectively marked as (x) 1 ,y 1 ) And (x) 2 ,y 2 ) The coordinates of the pressure centers of the heel and sole in the active shape model are respectively denoted as (x' 1 ,y' 1 ) And (x' 2 ,y' 2 ) First, calculate the scaling s and the rotation angle θ to make the translation vector t x And t y To be 0, the movable shape model is zoomed and rotated to obtain a temporary shape, and the pressure center coordinates of the heel and the sole of the foot which are recorded with the temporary shape are respectively (x ″) 1 ,y″ 1 ) And calculating a translation vector t x ,t y Then, let s =0 and θ =0, the temporary shape is translated to obtain the initial position X of the model c ；
Step (422): searching a new position of each feature point in the plantar pressure image; first, the initial position X of the active shape model is set c Overlaid on the plantar pressure image, as shown in particular in FIG. 5, wherein point F c 1，F c 2，…,F c 16，F c 17, are all initial positions X c The characteristic point of (1). For the ith characteristic point in the model, taking the ith characteristic point as the center in the direction vertical to the connecting line of the front characteristic point and the back characteristic point, and respectively selecting m (m) at two sides>h) Forming a search neighborhood with the length of 2m +1 by the pressure points and the characteristic point i; sliding windows with the length of h pressure points in the search neighborhood, and calculating the local texture g of each window i Calculating the Mahalanobis distance between the local texture and the average texture, and taking the central point of the window with the minimum Mahalanobis distance as the new position of the characteristic point i;
step (423): updating the attitude parameters; calculating a shape vector X of an initial position of the active shape model c And the shape vector X after updating the position new The transformation vector T and the transformation parameter b;
step (424): repeating steps (422) and (423) to calculate a new shape vector X new With the original shape vector X c Distance D of x ，
Wherein x is newi 、y newi Are respectively a new shape vector X new The abscissa and ordinate, x, of the ith feature point of (1) ci 、y ci Are respectively the original shape vector X c The abscissa and the ordinate of the ith feature point of (1). If it is
Or the circulation reaches the maximum times, the search is completed, and the accurate position of each feature point in the plantar pressure image is obtained;
and (5): as shown in fig. 6, wherein point F o 1，F o 2，…,F o 16，F o 17, the exact position of each feature point F1, F2, \ 8230;, F16, F17 in the plantar pressure image. According to the relative position relation between each area of the foot and the characteristic points of the sole pressure, the characteristic points of the sole pressure image obtained in the step (4) are combined to divide the sole pressure image into regions; in this embodiment, the plantar pressure image is divided into toe area, footThe 6 areas of the palm area, medial foot, lateral foot, medial heel and lateral heel; the method comprises the following specific steps: f o 3-F o The envelope surface of the No. 17 characteristic point is the whole footprint area; f o 3-F o 4-F o 5-F o 16-F o The envelope surface of the No. 17 characteristic point is a heel area, F o No. 3 and F o The No. 2 characteristic point connecting line can divide the heel area into a heel inner side area and a heel outer side area; f o 5-F o 6-F o 15-F o The envelope surface of the No. 16 characteristic point is a midfoot area, F o No. 1 and F o The connecting line of the No. 2 characteristic points can further divide the midfoot area into a medial midfoot area and a lateral midfoot area; f o 7-F o 8-F o 9-F o 13-F o 14-F o The envelope surface of the No. 15 characteristic point is a sole area; f o 9-F o 10-F o 11-F o 12-F o The envelope surface of the feature point No. 13 is a toe region.
Claims (6)
1. A gait analysis-oriented plantar pressure image partitioning method is characterized by comprising the following steps:
step (1): obtaining plantar pressure data and carrying out early-stage processing;
step (2): clustering the foot sole pressure data processed in the early stage in the step (1), and positioning the position of the foot print in the foot sole pressure data to obtain a foot sole pressure image of a foot print area;
and (3): performing integral projection on the plantar pressure image obtained in the step (2) to obtain a heel and sole pressure center;
and (4): establishing a plantar pressure characteristic point positioning model, and positioning characteristic points of a plantar pressure image;
and (5): according to the relative position relation between each area of the foot and the characteristic points of the sole pressure, the characteristic points of the sole pressure image obtained in the step (4) are combined to divide the sole pressure image into regions;
the step (4) realizes the following processes: the method comprises two stages of model training and feature point searching:
step (41): in the model training stage, selecting N plantar pressure images as a training set, wherein N is more than 200; manually marking the accurate positions of the characteristic points in the training sample, establishing a plantar pressure characteristic point positioning model, and executing the model training stage only once;
step (42): in the characteristic point searching stage, initializing a plantar pressure characteristic point positioning model according to the pressure centers of a heel and a sole, and then finding out the accurate position of a plantar pressure image characteristic point through an iterative model;
the plantar pressure characteristic point positioning model is a characteristic point positioning based on a template, is established by adopting an ASM (automatic sequence modeling) model or an AAM (adaptive analysis and maintenance) model, and is established by the following steps: firstly, a shape vector set X is collected on the basis of not changing a point distribution model through a Procrustes normalization method, namely through translation, rotation and scaling transformation operation u Aligning to the same shape vector X, wherein the shape vector X is a feature point template; then, local features are constructed for each feature point, and the construction of the feature point positioning model is completed;
the shape vectors are collected into X u Aligning to the same shape vector X comprises the steps of:
step (411): with a certain training sample X j As a basis for shape
For other training samples X i Performing translation, rotation and scaling to make all samples approximate to the reference shape; training sample X i And shape reference>
Proximity between uses Euclidean distance definition->
Obtaining a transformed shape vector of X' i ＝M(s,θ)[X i ]-t, where s is the scaling scale, θ is the rotation angle, t is the translation vector;
step (412): calculating all transformed training samples X' u As a new shape reference
And calculates the current shape reference>
And the last shape reference->
Translation, rotation and zoom deviations between;
step (413): iteration steps (411) and (412), if the deviation is smaller than a specified threshold value or the iteration exceeds a specified maximum iteration number, stopping the iteration; the shape reference obtained at the last time is used as an average shape vector X after all training samples are aligned;
the characteristic point searching stage comprises the following steps: initializing the model according to the pressure centers of the heel and the sole, and then finding out the accurate positions of the characteristic points of the sole pressure image through an iterative model; the method comprises the following specific steps:
step (421): calculating the pressure centers (x) of heel and sole in the target plantar pressure image by taking the pressure centers of heel and sole as the reference 1 ,y 1 ) And (x) 2 ,y 2 ) And a shape reference
Center of pressure of middle heel and sole (x' 1 ,y' 1 ) And (x' 2 ,y' 2 ) A translation, rotation, scaling deviation between, for a shape reference>
Transforming to obtain the initial position X of the model c ；
Step (422): searching a new position of each feature point in the plantar pressure image; first, the initial position X of the active shape model is set c Covering on the plantar pressure image, and regarding the ith characteristic point in the model, in the direction perpendicular to the connecting line of the front characteristic point and the rear characteristic point of the ith characteristic pointIt is used as center, and two sides of it respectively select m pressure points, m>h, a search neighborhood with the length of 2m +1 is formed by adding the characteristic point i; sliding windows with the length of h pressure points in the search neighborhood, and calculating the local texture g of each window i And calculating the local texture and average texture
The central point of the window with the minimum Mahalanobis distance is used as the new position of the characteristic point i; />
Step (423): calculating the initial position X of the model c The shape vector of (1) and the translation, rotation and scaling parameters of the shape vector after the position is updated;
step (424): repeating steps (422) and (423) to calculate a new shape vector X new And the original shape vector X c Distance D of x ，
Wherein x is newi 、y newi Are respectively a new shape vector X new The abscissa and ordinate, x, of the ith feature point of (1) ci 、y ci Are respectively the original shape vector X c The abscissa and ordinate of the ith feature point of (a), if
Or when the circulation reaches the maximum times, the search is finished, and the search is finished, so that the accurate position of each feature point in the plantar pressure image is obtained.
2. The plantar pressure image partition method for gait analysis according to claim 1, wherein: the specific implementation process of the step (1) comprises the following steps:
step (11): acquiring a plurality of adjacent frames (at least 3 frames) of plantar pressure data containing complete footprints in a time domain by adopting an array type pressure sensor;
step (12): performing time domain mean filtering on the middle frame plantar pressure data by using the multiple frames of plantar pressure data obtained in the step (11);
step (13): and (4) sequentially carrying out maximum value filtering processing and interpolation processing on the mid-frame plantar pressure data obtained in the step (12) to obtain pre-processed plantar pressure data with uniform size and equal row-column spacing.
3. The plantar pressure image partition method for gait analysis according to claim 1, wherein: the specific process of the step (2) comprises the following steps:
step (21): clustering the sole pressure data obtained in the step (1) after the pre-treatment by using a DBSCAN clustering algorithm to obtain a plurality of pressure area blocks, wherein the footprint area consists of a plurality of pressure area blocks;
step (22): using the minimum external moment center point of each pressure area block in the plurality of pressure area blocks obtained in the step (21) as the center of the pressure area block, and gathering the pressure area blocks into a footprint area by adopting a K-means clustering algorithm;
step (23): and (5) extracting the pressure data of the footprint area obtained in the step (22) to obtain a sole pressure image containing the complete footprint.
4. The plantar pressure image partition method for gait analysis according to claim 1, wherein: the specific process of the step (3) comprises the following steps:
step (31): taking the long side direction of the minimum external torque of the plantar pressure image obtained in the step (2) as a longitudinal axis and the short side direction as a transverse axis, and performing integral projection on the plantar pressure image along the longitudinal axis to obtain a gray level histogram;
step (32): if the gray level histogram obtained in the step (31) contains a plurality of similar peak points, performing mean filtering on the gray level histogram by adopting a sliding window with the length of N;
step (33): based on the fact that the sole is wider than the heel, the highest peak point of the filtered gray level histogram is the sole center line, and the next highest peak point is the heel center line;
step (34): 1/6 row of the total row number of the sole pressure images is respectively taken from the upper side and the lower side of the longitudinal central row of the sole to obtain a sole area, the images of the sole area are subjected to integral projection along a transverse axis to obtain a sole gray histogram, the highest peak point of the sole gray histogram is a sole central column, the next highest peak point is a heel central column, the intersection point of the sole central row and the sole central column is a sole central point, and the intersection point of the heel central row and the heel central column is a heel central point.
5. The plantar pressure image partition method for gait analysis according to claim 1, wherein: the processing method of the model training stage comprises the following steps: selecting N plantar pressure images obtained in the step (3) as a training set; the accurate positions of n characteristic points of each plantar pressure image are marked manually, wherein n is the number of the characteristic points, and n is>=6, the selection of the characteristic points comprises a heel, a sole central point and an inflection point of a sole pressure outline, the characteristic points and all areas of the foot have obvious relative position relations, the sequenced coordinates of the characteristic points are sequentially connected in series to form a shape vector, and a shape vector set obtained by N sole pressure images is marked as X u Wherein the shape vector obtained from the ith plantar pressure image is marked as X ui ，X ui ＝(x i0 ,…,x i(n-1) ,y i0 ,…,y i(n-1) ) T I =0, \ 8230, N-1, where x ik 、y ik Respectively representing the kth characteristic point F in the ith plantar pressure image k (x ik ,y ik ) K is more than or equal to 0 and less than n; set of shape vectors X u As input for establishing the part of the plantar pressure characteristic point positioning model, the X set is formed according to the shape vectors u And establishing a plantar pressure characteristic point positioning model for searching plantar pressure characteristic points, wherein the accurate positions of the characteristic points in the training samples are marked manually, and the plantar pressure characteristic point positioning model is established only once when the movable shape model is established.
6. The plantar pressure image partition method for gait analysis according to claim 5, wherein: the specific operation of constructing the local feature for each feature point is as follows: calculating average local texture of characteristic points i in training sample
Sum variance S i I =0,1, \8230;, n-1; firstly, j =0,1, \ 8230;, and N-1 are respectively selected from h pressure points on two sides of the ith characteristic point of the jth training sample along a direction perpendicular to a connecting line of the front characteristic point and the rear characteristic point of the jth training sample to form a vector with the length of 2h < +1 >, and a local texture g is obtained by deriving pressure values contained in the vector ij The same operation is carried out on all samples in the training set to obtain N local textures of the ith characteristic point, and the mean value is calculated and is used for being matched with the value>
And variance->
/>
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911366465.8A CN111127446B (en) | 2019-12-26 | 2019-12-26 | Gait analysis-oriented plantar pressure image partitioning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911366465.8A CN111127446B (en) | 2019-12-26 | 2019-12-26 | Gait analysis-oriented plantar pressure image partitioning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111127446A CN111127446A (en) | 2020-05-08 |
| CN111127446B true CN111127446B (en) | 2023-04-07 |
Family
ID=70503082
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911366465.8A Active CN111127446B (en) | 2019-12-26 | 2019-12-26 | Gait analysis-oriented plantar pressure image partitioning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111127446B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111289076B (en) * | 2020-05-13 | 2020-08-21 | 成都泰盟软件有限公司 | Human body physiological experiment system capable of automatically collecting human body basic data |
| CN114114969A (en) * | 2020-08-27 | 2022-03-01 | 比亚迪股份有限公司 | Vehicle, vehicle control method and device, storage medium and vehicle-mounted controller |
| CN112037213A (en) * | 2020-09-07 | 2020-12-04 | 深圳市凌云视迅科技有限责任公司 | Method and device for acquiring contour data stable feature points based on statistical histogram |
| CN112529782B (en) * | 2021-02-18 | 2021-06-01 | 棉捷(北京)网络科技有限公司 | Pressure data 3D imaging method and device, terminal equipment and readable storage medium |
| CN113951870A (en) * | 2021-09-26 | 2022-01-21 | 中国科学院深圳先进技术研究院 | Footprint statistical analysis method based on large-scale high-density piezoresistive film |
| CN117179741A (en) * | 2023-07-24 | 2023-12-08 | 北京华益精点生物技术有限公司 | Method for marking plantar position based on pressure and related equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20060097074A (en) * | 2005-03-02 | 2006-09-13 | 삼성전자주식회사 | Apparatus and method of generating shape model of object and apparatus and method of automatically searching feature points of object employing the same |
| WO2008073962A2 (en) * | 2006-12-12 | 2008-06-19 | Rutgers, The State University Of New Jersey | System and method for detecting and tracking features in images |
| CN103440510A (en) * | 2013-09-02 | 2013-12-11 | 大连理工大学 | Method for positioning characteristic points in facial image |
| CN110313916A (en) * | 2019-07-17 | 2019-10-11 | 军事科学院系统工程研究院军需工程技术研究所 | A kind of patient with diabetic feet plantar pressure characteristic recognition method |
| CN110427987A (en) * | 2019-07-17 | 2019-11-08 | 军事科学院系统工程研究院军需工程技术研究所 | A kind of the plantar pressure characteristic recognition method and system of arthritic |
-
2019
- 2019-12-26 CN CN201911366465.8A patent/CN111127446B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20060097074A (en) * | 2005-03-02 | 2006-09-13 | 삼성전자주식회사 | Apparatus and method of generating shape model of object and apparatus and method of automatically searching feature points of object employing the same |
| WO2008073962A2 (en) * | 2006-12-12 | 2008-06-19 | Rutgers, The State University Of New Jersey | System and method for detecting and tracking features in images |
| CN103440510A (en) * | 2013-09-02 | 2013-12-11 | 大连理工大学 | Method for positioning characteristic points in facial image |
| CN110313916A (en) * | 2019-07-17 | 2019-10-11 | 军事科学院系统工程研究院军需工程技术研究所 | A kind of patient with diabetic feet plantar pressure characteristic recognition method |
| CN110427987A (en) * | 2019-07-17 | 2019-11-08 | 军事科学院系统工程研究院军需工程技术研究所 | A kind of the plantar pressure characteristic recognition method and system of arthritic |
Non-Patent Citations (4)
| Title |
|---|
| 一种用于步态分析的足印提取算法;叶强等;《传感技术学报》;20181015(第10期);全文 * |
| 基于因子分析的倒钟摆模型足底压力分布研究;林芳富等;《人类工效学》;20190420(第02期);全文 * |
| 基于柔性压力阵列传感器的人体静态平衡功能评估;武飞等;《仪表技术》;20181115(第11期);全文 * |
| 基于足底压力图像的静态触觉步态识别;梁栋等;《华中科技大学学报(自然科学版)》;20131023(第10期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111127446A (en) | 2020-05-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111127446B (en) | Gait analysis-oriented plantar pressure image partitioning method | |
| WO2022166800A1 (en) | Deep learning network-based automatic delineation method for mediastinal lymphatic drainage region | |
| CN109003278B (en) | Improved active shape model-based CT image aorta segmentation method | |
| Li et al. | Brain tumor detection based on multimodal information fusion and convolutional neural network | |
| KR102154470B1 (en) | 3D Human Hairstyle Generation Method Based on Multiple Feature Search and Transformation | |
| CN108364006B (en) | Medical image classification device based on multi-mode deep learning and construction method thereof | |
| CN107633522B (en) | Brain image segmentation method and system based on local similarity active contour model | |
| CN100528075C (en) | System and method for segmenting the left ventricle in a cardiac MR image | |
| CN110659565B (en) | 3D multi-person human body posture estimation method based on porous convolution | |
| US20180114313A1 (en) | Medical Image Segmentation Method and Apparatus | |
| CN106887019B (en) | Footprint pressure distribution characteristic representation method | |
| CN110120051A (en) | A kind of right ventricle automatic division method based on deep learning | |
| CN108305283A (en) | Human bodys' response method and device based on depth camera and basic form | |
| CN112766142B (en) | Sole pressure image processing method, sole pressure image identification method and gait analysis system | |
| CN109118528A (en) | Singular value decomposition image matching algorithm based on area dividing | |
| CN106296613B (en) | A kind of Dual Energy Subtraction method based on DR machine | |
| CN108876770B (en) | Fundus multispectral image joint registration method and system | |
| CN104598871A (en) | Correlation regression based face age calculating method | |
| US11334995B2 (en) | Hierarchical systems and methods for image segmentation | |
| CN114782715B (en) | Vein recognition method based on statistical information | |
| CN111104902A (en) | Hemiplegia gait classification method based on graph convolution network | |
| CN109003283A (en) | A kind of aorta outline segmentation based on active shape model | |
| CN112258536B (en) | Integrated positioning and segmentation method for calluses and cerebellum earthworm parts | |
| CN110751200B (en) | Shoe stamp height estimation method based on multiple gauss | |
| Wang et al. | Assessment of gait characteristics in total knee arthroplasty patients using a hierarchical partial least squares method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |