CN108416276A - The abnormal gait detection method of side gait video based on people - Google Patents
The abnormal gait detection method of side gait video based on people Download PDFInfo
- Publication number
- CN108416276A CN108416276A CN201810143888.2A CN201810143888A CN108416276A CN 108416276 A CN108416276 A CN 108416276A CN 201810143888 A CN201810143888 A CN 201810143888A CN 108416276 A CN108416276 A CN 108416276A
- Authority
- CN
- China
- Prior art keywords
- gait
- abnormal
- feature
- video
- people
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
- G06V40/25—Recognition of walking or running movements, e.g. gait recognition
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The abnormal gait detection method of the invention discloses a kind of side gait video based on people; side gait video is acquired by common general camera or mobile phone; therefrom extract step change feature, the forward lean feature of people; and single category support vector machines model is trained using the characteristic information of normal gait, it is normal or abnormal that can quickly and effectively detect corresponding gait.The present invention is not necessarily to special inspecting equipment, and without carrying out gait detection in specific place, flexibility ratio is high.The present invention considers forward lean feature when by gait video extraction gait feature, and the detectability of abnormal gait detection model can effectively be avoided to be influenced by abnormal training data, improves Detection accuracy.Present invention only requires the single category support vector machines models of feature training using normal gait information, and training sample is few, can rapidly and accurately detect that gait is normal/abnormal, to the diagnosis for carrying out abnormal gait that assists a physician, improve the working efficiency of doctor.
Description
Technical field
The present invention relates to Gait Recognition fields, and in particular to extracts feature in a kind of side gait video from people and combines
The method that abnormal gait detection model realizes abnormal gait detection.
Background technology
The posture of gait, that is, people's walking, abnormal gait are then the improper walkings occurred when a human body occurs abnormal
Posture.Common includes pain, central nervous system exception and skeletal musculature damage the reason of leading to abnormal gait.It is abnormal
The type of gait is various, and typical abnormal gait includes spastic hemiplegia gait, spastic paraplegia gait, sensory ataxia
Gait, festinating gait, myopathic gait, steppage gait, hysteria gait etc..The appearance of part typical gait reflects feature disease
In the presence of by carrying out observation analysis to abnormal gait, symptomatic diagnosis can be carried out to patient, such as common flurried of Parkinsonian
Open gait, freezing of gait etc..Therefore, the detection of abnormal gait is the important evidence of diagnosis.Although at present about people
Gait research, but all concentrate on using gait carry out identification on, abnormal gait detection on research it is less, and
The gait that these methods are all based on normal person carries out identification classification, is not particularly suited for general abnormal gait detection.
On clinical medicine, frequently with directly observation or the objective acquisitions such as electromyogram are based on to the abnormal gait detection of patient
The quantitative analysis of data.The acquisition of abnormal gait is carried out with analysis usually using external hardware instruments, angle measurement is such as used
The movement locus of device synthetical collection major joint analyzes the gait cycle rule of people；Or it is adopted by wearable gait collector
Collect plantar pressure, is tipped according to the mutation analysis of plantar pressure data, the exception of the classifications such as heel is walked, toed-out and toed-in
Gait.These methods analyze the gait of patient based on hardware device, higher to equipment requirement, need particular place into
Row, flexibility ratio is relatively low, is unfavorable for the abnormal gait detection of daily life.Individual researchs are (as application No. is 2017107435559
Patent of invention《Abnormal gait detection method and abnormal gait detecting system》) normal abnormal by human body contour outline extraction feature progress
The characteristics of classifying, but only accounting for changing on lower part of the body stride, the upper part of the body in some abnormal gaits such as festinating gait is not considered
The characteristics of forward lean, while used training pattern is easy the abnormal gait data influence in by training data, Bu Nengyou
Effect detects rich and varied abnormal gait.
Therefore, it is necessary to be improved to the prior art.
Invention content
The technical problem to be solved by the present invention is to propose that a kind of side gait based on people without professional detection device regards
The abnormal gait detection method of frequency.
To solve the above problems, the present invention proposes a kind of abnormal gait detection method of the side gait video based on people,
Include the following steps：
Step 1 obtains side gait video, and extracts silhouette contour images sequence from the gait video of side；
Step 2, according to the silhouettes image sequence of gained in step 1, structure step width and height than argument sequence A,
The x-axis horizontal distance at head and two sufficient centers and height ratio lean forward argument sequence B, and by argument sequence A and argument sequence B extractions
The feature of gait information is described；
Step 3 collects the side gait video of normal gait as training sample；The training sample passes sequentially through step
1 and step 2 extract the feature of its gait information, and the feature extracted is inputted into single category support vector machines and is trained, structure
Build abnormal gait detection model；
Step 4 obtains side gait video to be detected, passes sequentially through step 1 and step 2 extracts the spy of its gait information
Sign, and will be detected in the abnormal gait detection model constructed by this feature input step 3.
The improvement of abnormal gait detection method as the side gait video based on people：
The step 2 includes the following steps：
2.1, by each frame silhouettes image in the silhouettes image sequence obtained by step 1 according to each point pixel value
Segmentation obtains human region image；
2.2, divide obtained human region image zooming-out height h, cephalad apex x-axis coordinate x according to step 2.11；
2.3,4/5h to the h height for the human region image that step 2.1 is obtained is intercepted according to direction from top to bottom
Foot area image obtains the sufficient central point x-axis coordinate x of actual step size width w and two2；
2.4, parameter the calculating parameter α and β of gained are calculated according to step 2.3 and step 2.4：
Step width is calculated with height than parameter alpha, calculation formula is
Calculate head level lean forward distance with height than parameter beta, calculation formula is
2.5, the parameter alpha of each frame silhouettes image zooming-out, β are added respectively in argument sequence A and argument sequence B
In, until all silhouettes image procossings finish in silhouettes image sequence, obtain argument sequence A and argument sequence B；
That is, parameter alpha is added successively in argument sequence A, parameter beta is added successively in argument sequence B.
2.6, period of extraction gait in the curve that the argument sequence A obtained from step 2.5 is drawn, first half cycle, after
The transit time 8 of half period, wave crest amplitude, trough amplitude, wave crest variance, the transit time of trough to wave crest and wave crest to trough
A feature；
2.7, extraction indicates mean value, variance, the maximum of forward lean feature in the argument sequence B obtained from step 2.5
Value and 4 features of minimum value；
2.8, the feature that combining step 2.6 and 2.7 extracts constitutes final feature G.
Abnormal gait detection method as the side gait video based on people is further improved：
The step 3 includes the following steps：
3.1, the side gait video for collecting normal gait builds training sample；
3.2, it the training sample that step 3.1 is built is passed sequentially through into step 1 and step 2 extracts the feature of its gait information and obtain
Obtain characteristic parameter training set；
3.3, the training of model is carried out abnormality detection using step 3.2 gained characteristic parameter training set, builds abnormal gait
Detection model.
Abnormal gait detection method as the side gait video based on people is further improved：
The abnormal gait detection model is single category support vector machines.
Abnormal gait detection method as the side gait video based on people is further improved：
The optimization aim of the list category support vector machines model is to ask a center for o, and radius is the minimum spherical surface of R,
Formula is as follows：
And meet condition：
Wherein, u is input data, and i=1 ..., N, N is the number of input data, and C is penalty factor, and ξ is relaxation factor.
Simplify using method of Lagrange multipliers and to constraints, optimization problem can be converted to：
Constraints is：
Wherein, i=1 ..., N, j=1 ..., N, N are the number of input data, and λ is Lagrange multiplier.
The kernel function of this model uses RBF radial basis function：
K(ui, uj)=exp (- γ | ui-uj|2),
Wherein, γ is nuclear parameter, this model value is 0.083.
Compared with prior art, the technical advantages of the present invention are that：
1, the present invention extracts step width and body high specific parameters (that is, parameter alpha), forward lean parameter (that is, ginseng from pedestrian contour
Number β) it is used as gait feature, gait characteristic is effectively described, the extraction of wherein step width has intercepted the foot area of people, in certain journey
Reduce the interference that the dress ornaments such as overcoat extract step width on degree；By extracting forward lean feature, walking is better described
The phenomenon that whether there is forward lean in the process, contributes to the detection of abnormal gait.
2, the present invention carries out mould using normal gait data according to selected gait feature using single category support vector machines
Abnormal gait detection model is established in type training, does not need all kinds of abnormal gaits as training data, you can effectively distinguishes normal step
State and all kinds of abnormal gaits, it is abnormal or normal to detect gait to be measured.
Description of the drawings
The specific implementation mode of the present invention is described in further detail below in conjunction with the accompanying drawings.
Fig. 1 is that the present invention is based on the model foundation block diagrams of the abnormal gait detection method of the side gait video of people；
Fig. 2 is the flow diagram that characterization step is extracted in Fig. 1；
Fig. 3 is in embodiment 1 from human body contour outline extracting parameter h and x1Schematic diagram；
Fig. 4 is in embodiment 1 from human body contour outline extracting parameter w and x2Schematic diagram；
Fig. 5 is the PARAMETER ALPHA change curve of normal gait and abnormal gait；
Fig. 6 is the characteristic parameter β change curves of normal gait and abnormal gait.
Specific implementation mode
With reference to specific embodiment, the present invention is described further, but protection scope of the present invention is not limited in
This.
Embodiment 1, the abnormal gait detection method of side gait video based on people pass through as shown in figures 1 to 6 and acquire side
Face gait video, and effectively the features such as the leg speed of people, stride is described, whether leans forward in extraction gait parameter, to exception
Gait makes detection.As shown in Figure 1, specifically including following steps：
Step 1, the side gait video that people is shot using common general camera or mobile phone, need to use three when shooting video
Tripod establishing shot equipment, video format can be MP4 or AVI, shoot the side of people, include at least two gait cycles.One
Complete gait cycle refers to people since the heel contact of side, to the process that lands again of heel of homonymy.It is based on background again
Calculus of finite differences (existing foreground extraction algorithm) extracts silhouette contour images sequence from the gait video of side.
Step 2, as shown in Fig. 2, according to silhouettes image sequence, build respectively step width and height than argument sequence A,
The x-axis horizontal distance at head and two sufficient centers, than the argument sequence B to lean forward, is extracted with height by argument sequence A and argument sequence B
12 feature description gait informations.Eigen extracting method has effectively reflection people in the process of walking in leg speed time, distance
In parameter whether normally, the advantages of the states such as whether body leans forward.As steps described below successively to silhouette contour images sequence
Each frame silhouettes image is handled：
2.1, divide human region：It will be in each frame silhouettes image in the silhouettes image sequence obtained by step 1
Divide human region according to each point pixel value.
Human region segmentation uses image traversal method, defines tetra- changes of x_left, x_right, y_top and y_bottom
Amount indicates ultra-left point x-axis coordinate of the human region on original image, rightest point x-axis coordinate, peak y-axis coordinate and most respectively
Low spot y-axis coordinate.Respectively according to four direction from left to right, from right to left, from top to bottom and from top to bottom, every is calculated successively
The sum of the pixel value of image column or row.Since in binary image, black region thresholding is 0, white area thresholding is 255, therefore,
When it is 0 to calculate the sum of column or row pixel value not, illustrate to detect that human region boundary line, record variable value terminate traversal and look into
It looks for.
The specific method of calculating x_left is in the present embodiment：X_left=0 is initialized, successively according to from left to right direction
The sum of every image column pixel value is calculated, when the sum of pixel value is 0, x_left values add 1, continue to calculate next image column picture
The sum of element value, until the sum of pixel value is not 0 stopping, it is ultra-left point that the row pixel value, which is not 0 pixel, at this time, and x-axis is sat
Mark is x_left at this time, records x_left；Calculate x_right method be：It is wide to initialize the size that x_right is image
It spends (such as 300), calculates the sum of every pixel column by column according to direction from right to left later, when the sum of pixel value is 0, x_
Right values subtract 1, continue to calculate the sum of next image column pixel value, until the sum of pixel value is not 0 to stop, the row picture at this time
Plain value is not rightest point for 0 pixel, and x-axis coordinate is x_right at this time, records x_right；Calculate y_top and
The method of y_bottom is identical as above-mentioned computational methods, therefore does not repeat to describe.
Finally, according to four variables (x_left, x_right, y_top and y_bottom) for calculating gained, silhouette is cut
Contour images obtain human region image, as shown in Figure 3.
2.2, the human region image zooming-out height h divided according to step 2.1, cephalad apex x-axis coordinate x1；
Height h is the dimensional height of the human region image, that is, the difference of y_bottom and y_top in step 2.1.
x1Calculation be：Initialize x1=0, each pixel of human region image the first row is traversed, if pixel
Point value is that 0, x values add 1, continues checking for next pixel, until pixel terminates when not being 0, records x at this time1Value.
2.3, that human region image that step 2.1 is obtained is intercepted its 4/5h to h according to direction from top to bottom is high
Degree foot area image (that is, according to direction from top to bottom intercept its 0 arrive 1/5h height foot area image), to subtract
The interference that the dress ornaments such as small overcoat extract step width.As shown in figure 4, obtaining practical stride width from the foot area image of interception
The sufficient central point x-axis coordinate x of w and two2。
W and x2Computational methods be：Define x1_ left, x1Two variables of _ right indicate the most left of foot area respectively
Point x-axis coordinate, rightest point x-axis coordinate.Respectively according to from left to right and from right to left both direction, every image column is calculated successively
The sum of pixel value.
X is calculated in the present embodiment1The method of _ left is：Initialize x1_ left is calculated every successively according to from left to right direction
The sum of image column pixel value, when the sum of pixel value is 0, x1_ left values add 1, continue to calculate next image column pixel value
The sum of, until the sum of pixel value is not 0 stopping, it is the ultra-left point of foot area that the row pixel value, which is not 0 pixel, at this time,
X-axis coordinate record is in x1In _ left；x1The calculating of _ right is obtained with direction from right to left (in step 2.1 in this way
The computational methods of x_right).
W presses calculation formula w=| x1_right-x1_ left | it calculates, x2By formulaIt calculates.
2.4, parameter (i.e. h, x of gained are calculated according to step 2.3 and step 2.41, w and x2) calculating parameter α and β：
By custom formulaStep width is calculated with height than parameter alpha, custom formulaCalculate head water
The flat distance that leans forward is with height than parameter beta (i.e. forward lean parameter)；
2.5, by the addition of the parameter alpha of each frame silhouettes image zooming-out in argument sequence A, parameter beta is added in parameter
In sequence B, until silhouettes image sequence in all silhouettes image procossings finish, obtain final argument sequence A and
B。
Record the video of normal gait and abnormal gait (festinating gait) in the present embodiment respectively using camera, and according to upper
It states step to be handled, obtains corresponding argument sequence A and B.As shown in figure 5, the argument sequence A that left figure is normal gait is corresponded to
Curve graph, right figure be the corresponding curve graphs of abnormal gait argument sequence A, abscissa be frame number, ordinate is parameter alpha；By
Fig. 5 can be seen that the gait motion of normal person in periodically, and with walking forward, left and right foot switching, gait parameter α fluctuations exist
Between 0.15 to 0.50, peaks and troughs amplitude is basicly stable；The parameter alpha value of festinating gait fluctuates between 0.15 to 0.25,
The characteristics of wave crest amplitude is relatively low, and it is smaller to reflect patient's paces, difficulty in walking.
As shown in fig. 6, left figure is the corresponding curve graphs of argument sequence B of normal gait, right figure is abnormal gait parameter sequence
The corresponding curve graphs of B are arranged, abscissa is frame number, and ordinate is parameter beta；The parameter P value of normal gait is substantially as seen from Figure 6
Between 0.0 to 0.1, fluctuation is little；The parameter P value of festinating gait is between 0.1 to 0.4, and fluctuation is larger, and mean value more normally walks
State is high.By Fig. 5 and Fig. 6 it is found that argument sequence A and B can effectively distinguish normal gait and abnormal gait.
2.6, period G1, the first half cycle of gait are extracted in the curve that the argument sequence A obtained from step 2.5 is drawn
G2, later half period G3, wave crest amplitude G4, trough amplitude G5, wave crest variance G6, the transit time G7 of trough to wave crest and wave crest arrive
The transit time G8 of trough totally 8 features.
As shown in Figure 5, wherein gait cycle G1 is defined as i-th of wave crest to the frame number between the i-th+2 wave crests, quite
In the time of a gait cycle.Frame number is bigger to illustrate that the period is longer, and walking speed is slower, it is understood that there may be abnormal gait.Gait
First half cycle G2 is defined as i-th of wave crest to the frame number between i+1 wave crest.This feature is in normal gait, it should
It is substantially equal to G3 features.When the two differs greatly, it is understood that there may be single leg is abnormal.Gait first half cycle G3 is defined as i+1
Wave crest is to the frame number between the i-th+2 wave crests.Wave crest amplitude G4 is curve peak, actually reflects the step-length of people.When the value
When smaller, illustrate that paces are smaller, it is understood that there may be difficulty of taking a step.Trough amplitude G5 is curve minimum point, actually reflects people's walking
At the time of both legs are overlapped in the process.When the value is larger, it is understood that there may be dysbasia is walked by means of supporter.Wave crest variance G6
For the variance of all wave crests in gait video time, the variation of stride in walking process is actually reflected.The bigger reflection of variance
Patient's walking is abnormal, and uneven length of step is even.The transit time G7 of trough to wave crest is people overlap onto that biped lands from both legs when
Between, the transit time G8 of trough to wave crest is that people lands the time being overlapped to both legs from biped, the practical reflection of the two features
The leg speed of single leg.
2.7, forward lean feature, i.e. the mean value G9 of parameter beta, variance are extracted in the argument sequence B obtained from step 2.5
G10, maximum value G11 and minimum value G12 totally 4 features, can describe people by these features and whether there is in the process of walking
Forward lean or the phenomenon that rock back and forth.
As shown in fig. 6, maximum value G11 is curve peak, minimum value G12 is curve peak.
Note：Above-mentioned mean value G9's and variance G10 is calculated as the prior art, therefore its calculating is not discussed in detail in the present specification
Method.
2.8, it is as follows to constitute final feature G for the feature that combining step 2.6 and 2.7 extracts：
G=[G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12]
Step 3, the side gait video (hereinafter referred to as normal gait video) for collecting normal gait are training sample；
The training sample passes sequentially through the feature that step 1 and step 2 extract its gait information, and by feature input single classification support to
Amount machine is trained, and builds abnormal gait detection model.This method for establishing model, which has, does not need all kinds of abnormal gaits as instruction
Practice data, you can the advantages of effectively distinguishing normal gait and all kinds of abnormal gaits.The step specifically includes following steps：
3.1, the side gait video for collecting normal gait video and abnormal gait (is hereinafter referred to as abnormal gait to regard
Frequently)；Randomly select 70% structure training sample in collected normal gait video, and by remaining 30% normal gait video with
Collected abnormal gait video mix constitutes test sample.
3.2, the training sample that step 3.1 is built is passed sequentially through into step 1 and step 2 extracts the spy of its side gait video
It levies (that is, feature of normal gait information) and obtains characteristic parameter training set.
3.3, characteristic parameter training set is inputted single category support vector machines to be trained, training obtains abnormal gait detection
Model.
Single category support vector machines (One-class-SVM) are a kind of existing abnormality detection models, are suitable for positive and negative sample
The list classification task such as cloth imbalance, abnormality detection.Common abnormality detection model further includes orphan in addition to single category support vector machines
Vertical forest, the local models such as factors check that peel off.The main thought of single category support vector machines is one multi-dimensional sphere model of training
, if data point is fallen in hypersphere, belong to this when needing the data new to one to judge including all positive samples
Class is not belonging to if falling outside ball, is abnormal data.The present invention is joined using existing single category support vector machines using feature
Number training sets are trained single category support vector machines, build hypersphere, and the hypersphere is by the spy containing normal gait information
Sign is surrounded, to realize the detection of abnormal gait.The optimization aim of model is to ask a center for o, and radius is the minimum of R
Spherical surface：
And meet condition：
Wherein, u is input data, and i=1 ..., N, N is the number of input data, and C is penalty factor, and ξ is relaxation factor,
F, that is, object function.
It carries out simplifying (note using method of Lagrange multipliers Solve problems and to constraints：Method of Lagrange multipliers is normal
With optimization problem solving method, it is existing mathematical method, therefore its computational methods is not discussed in detail in the present specification), it can will be excellent
Change problem is converted to：
Constraints is：
Wherein, i=1 ..., N, j=1 ..., N, N are the number of input data, and λ is Lagrange multiplier, and L, that is, glug is bright
Day function.
The kernel function of this model uses RBF radial basis function：
K(ui, uj)=exp (- γ | ui-uj|2),
Wherein, γ is nuclear parameter, this model value is 0.083, K, that is, radial basis function.
3.4, the test sample (or practical side gait video to be detected) that step 3.1 is built is passed sequentially through into step 1
The feature of its gait information is extracted with step 2, and by the abnormal gait detection model constructed by feature input step 3.3, if number
It is fallen in trained hypersphere according to characteristic point, then it is normal gait to export+1, and it is abnormal gait otherwise to export -1.To quickly examine
Whether abnormal survey gait.
Experiment, influence of the forward lean feature to abnormal gait Detection accuracy：
Test sample is made of 85 parts of normal gait videos and 35 parts of abnormal gait videos in this experiment, wherein 35 parts of exceptions
Gait video is made of 10 parts of cyllopodias, 10 parts of spastic hemiplegia gaits, 5 parts of sensory ataxia gaits, 10 parts of festinating gaits.
It extracts in test sample after the final feature G of each side gait video, feature G inputs embodiment 1 is established different successively
Normal gait detection model is detected, and testing result is as shown in table 1：
Table 1
Comparative example cancels step 2.7 in embodiment, remaining is equal to embodiment；I.e. final feature G only includes feature
G1-G8 (cancels mean value G9, variance G10, maximum value G11 and minimum value G12)；Using identical training sample in embodiment, carry
Its final single category support vector machines of feature G training is taken, to build abnormal gait detection model；Extract test specimens successively again
In this after final feature G of each side gait video, and the abnormal gait detection model that feature G inputs comparative example is established
It is detected, testing result is as shown in table 1.
As shown in Table 1, increase forward lean feature (mean value G9, variance G10, maximum value G11 and minimum value G12) model afterwards
The accuracy rate of testing result greatly improves.From the point of view of the testing result of normal gait data, the correct accuracy rate for detecting positive sample
It is improved before not increasing forward lean feature relatively, reduces erroneous judgement；From the point of view of the testing result of abnormal gait data, increase
After adding forward lean feature, the rate of accuracy reached of negative sample is correctly detected as to 100.0%, wherein particularly, for there are bodies
The festinating gait for the feature that leans forward, model inspection effect are significantly improved.
In summary, the present invention acquires side gait video by common general camera or mobile phone, therefrom extracts people's
Step change feature, forward lean feature, and single category support vector machines model is trained using the characteristic information of normal gait, it can
Quickly and effectively the corresponding gait of detection is normal or abnormal.The present invention is not necessarily to special inspecting equipment, without being carried out in specific place
Gait detects, and flexibility ratio is high.The present invention considers forward lean feature when by gait video extraction gait feature, and can be effective
It avoids the detectability of abnormal gait detection model from being influenced by abnormal training data, improves Detection accuracy.Present invention only requires
Using the single category support vector machines model of feature training of normal gait information, training sample is few, can rapidly and accurately detect
Gait is normal/abnormal, to the diagnosis for carrying out abnormal gait that assists a physician, improves the working efficiency of doctor.
Finally, it should also be noted that it is listed above be only the present invention several specific embodiments.Obviously, this hair
Bright to be not limited to above example, acceptable there are many deformations.Those skilled in the art can be from present disclosure
All deformations for directly exporting or associating, are considered as protection scope of the present invention.
Claims (5)
1. the abnormal gait detection method of the side gait video based on people, feature include the following steps：
Step 1 obtains side gait video, and extracts silhouette contour images sequence from the gait video of side；
Step 2, according to the silhouettes image sequence of gained in step 1, structure step width and height than argument sequence A, head
It leans forward argument sequence B with the x-axis horizontal distance at two sufficient centers and height ratio, and by argument sequence A and argument sequence B extraction descriptions
The feature of gait information；
Step 3 collects the side gait video of normal gait as training sample；The training sample pass sequentially through step 1 and
Step 2 extracts the feature of its gait information, and the feature extracted is inputted single category support vector machines and is trained, and structure is different
Normal gait detection model；
Step 4 obtains side gait video to be detected, passes sequentially through step 1 and step 2 extracts the feature of its gait information,
And it will be detected in the abnormal gait detection model constructed by this feature input step 3.
2. the abnormal gait detection method of the side gait video according to claim 1 based on people, it is characterised in that：
The step 2 includes the following steps：
2.1, will be divided according to each point pixel value in each frame silhouettes image in the silhouettes image sequence obtained by step 1
Obtain human region image；
2.2, divide obtained human region image zooming-out height h, cephalad apex x-axis coordinate x according to step 2.11；
2.3, the foot of 4/5h to the h height of the human region image obtained according to direction interception step 2.1 from top to bottom
Area image obtains the sufficient central point x-axis coordinate x of actual step size width w and two2；
2.4, parameter the calculating parameter α and β of gained are calculated according to step 2.3 and step 2.4：
Step width is calculated with height than parameter alpha, calculation formula is
Calculate head level lean forward distance with height than parameter beta, calculation formula is
2.5, the parameter alpha of each frame silhouettes image zooming-out, β are added respectively in argument sequence A and argument sequence B, directly
It is finished to all silhouettes image procossings in silhouettes image sequence, obtains argument sequence A and argument sequence B；
2.6, the period of extraction gait, first half cycle, second half in the curve that the argument sequence A obtained from step 2.5 is drawn
8 spies of transit time of phase, wave crest amplitude, trough amplitude, wave crest variance, the transit time of trough to wave crest and wave crest to trough
Sign；
2.7, in the argument sequence B obtained from step 2.5 extraction indicate the mean value of forward lean feature, variance, maximum value and
4 features of minimum value；
2.8, the feature that combining step 2.6 and 2.7 extracts constitutes final feature G.
3. the abnormal gait detection method of the side gait video according to claim 1 based on people, it is characterised in that：
The step 3 includes the following steps：
3.1, the side gait video for collecting normal gait builds training sample；
3.2, the training sample that step 3.1 is built is passed sequentially through into step 1 and step 2 extracts the feature acquisition spy of its gait information
Levy parameter training collection；
3.3, the training of model, structure abnormal gait detection are carried out abnormality detection using step 3.2 gained characteristic parameter training set
Model.
4. according to the abnormal gait detection method of any side gait videos based on people of claim 1-3, feature
It is：The abnormal gait detection model is single category support vector machines.
5. the abnormal gait detection method of the side gait video according to claim 4 based on people, it is characterised in that：
The optimization aim of the list category support vector machines model is to ask a center for o, and radius is the minimum spherical surface of R, formula
It is as follows：
And meet condition：
Wherein, u is input data, and i=1 ..., N, N is the number of input data, and C is penalty factor, and ξ is relaxation factor；
Simplify using method of Lagrange multipliers and to constraints, optimization problem can be converted to：
Constraints is：
Wherein, i=1 ..., N, j=1 ..., N, N are the number of input data, and λ is Lagrange multiplier；
The kernel function of this model uses RBF radial basis function：
K(ui, uj)=exp (- γ | ui-uj|2),
Wherein, γ is nuclear parameter, this model value is 0.083.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810143888.2A CN108416276B (en) | 2018-02-12 | 2018-02-12 | Abnormal gait detection method based on human lateral gait video |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810143888.2A CN108416276B (en) | 2018-02-12 | 2018-02-12 | Abnormal gait detection method based on human lateral gait video |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108416276A true CN108416276A (en) | 2018-08-17 |
| CN108416276B CN108416276B (en) | 2022-05-24 |
Family
ID=63128601
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810143888.2A Active CN108416276B (en) | 2018-02-12 | 2018-02-12 | Abnormal gait detection method based on human lateral gait video |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108416276B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110322528A (en) * | 2019-06-26 | 2019-10-11 | 浙江大学 | Nuclear magnetic resonance brain image reconstructing blood vessel method based on 3T, 7T |
| CN110728226A (en) * | 2019-10-09 | 2020-01-24 | 清华大学 | Gait quantification system and method based on motion recognition |
| CN111062340A (en) * | 2019-12-20 | 2020-04-24 | 湖南师范大学 | Abnormal gait behavior identification method based on virtual posture sample synthesis |
| CN111144166A (en) * | 2018-11-02 | 2020-05-12 | 银河水滴科技（北京）有限公司 | Method, system and storage medium for establishing abnormal crowd information base |
| CN112597903A (en) * | 2020-12-24 | 2021-04-02 | 珠高电气检测有限公司 | Electric power personnel safety state intelligent identification method and medium based on stride measurement |
| CN113705482A (en) * | 2021-08-31 | 2021-11-26 | 江苏唯宝体育科技发展有限公司 | Body health monitoring and management system and method based on artificial intelligence |
| CN114533044A (en) * | 2022-02-08 | 2022-05-27 | 北京金史密斯科技股份有限公司 | Gait data acquisition system and method and running equipment |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101488185A (en) * | 2009-01-16 | 2009-07-22 | 哈尔滨工程大学 | Partitioned matrix-based gait recognition method |
| CN101571917A (en) * | 2009-06-16 | 2009-11-04 | 哈尔滨工程大学 | Front side gait cycle detecting method based on video |
| US20130090881A1 (en) * | 2011-10-10 | 2013-04-11 | Texas Instruments Incorporated | Robust step detection using low cost mems accelerometer in mobile applications, and processing methods, apparatus and systems |
| CN103927524A (en) * | 2014-04-25 | 2014-07-16 | 哈尔滨工程大学 | Multi-angle gait period detection method |
| CN105326627A (en) * | 2015-11-25 | 2016-02-17 | 华南理工大学 | Trunk center-of-gravity shift based rehabilitation device walking triggering control method |
| CN106295691A (en) * | 2016-08-04 | 2017-01-04 | 江南大学 | The feature selection of single classification SVM and parameter synchronization optimization method |
| CN106600631A (en) * | 2016-11-30 | 2017-04-26 | 郑州金惠计算机系统工程有限公司 | Multiple target tracking-based passenger flow statistics method |
| CN106667494A (en) * | 2017-02-23 | 2017-05-17 | 佛山市量脑科技有限公司 | Moving posture monitoring insole |
| CN106875405A (en) * | 2017-01-19 | 2017-06-20 | 浙江大学 | CT image pulmonary parenchyma template tracheae removing methods based on BFS |
| US20170213081A1 (en) * | 2015-11-19 | 2017-07-27 | Intelli-Vision | Methods and systems for automatically and accurately detecting human bodies in videos and/or images |
| CN107133604A (en) * | 2017-05-25 | 2017-09-05 | 江苏农林职业技术学院 | A kind of pig abnormal gait detection method based on ellipse fitting and predictive neutral net |
| CN107563431A (en) * | 2017-08-28 | 2018-01-09 | 西南交通大学 | A kind of image abnormity detection method of combination CNN transfer learnings and SVDD |
-
2018
- 2018-02-12 CN CN201810143888.2A patent/CN108416276B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101488185A (en) * | 2009-01-16 | 2009-07-22 | 哈尔滨工程大学 | Partitioned matrix-based gait recognition method |
| CN101571917A (en) * | 2009-06-16 | 2009-11-04 | 哈尔滨工程大学 | Front side gait cycle detecting method based on video |
| US20130090881A1 (en) * | 2011-10-10 | 2013-04-11 | Texas Instruments Incorporated | Robust step detection using low cost mems accelerometer in mobile applications, and processing methods, apparatus and systems |
| CN103927524A (en) * | 2014-04-25 | 2014-07-16 | 哈尔滨工程大学 | Multi-angle gait period detection method |
| US20170213081A1 (en) * | 2015-11-19 | 2017-07-27 | Intelli-Vision | Methods and systems for automatically and accurately detecting human bodies in videos and/or images |
| CN105326627A (en) * | 2015-11-25 | 2016-02-17 | 华南理工大学 | Trunk center-of-gravity shift based rehabilitation device walking triggering control method |
| CN106295691A (en) * | 2016-08-04 | 2017-01-04 | 江南大学 | The feature selection of single classification SVM and parameter synchronization optimization method |
| CN106600631A (en) * | 2016-11-30 | 2017-04-26 | 郑州金惠计算机系统工程有限公司 | Multiple target tracking-based passenger flow statistics method |
| CN106875405A (en) * | 2017-01-19 | 2017-06-20 | 浙江大学 | CT image pulmonary parenchyma template tracheae removing methods based on BFS |
| CN106667494A (en) * | 2017-02-23 | 2017-05-17 | 佛山市量脑科技有限公司 | Moving posture monitoring insole |
| CN107133604A (en) * | 2017-05-25 | 2017-09-05 | 江苏农林职业技术学院 | A kind of pig abnormal gait detection method based on ellipse fitting and predictive neutral net |
| CN107563431A (en) * | 2017-08-28 | 2018-01-09 | 西南交通大学 | A kind of image abnormity detection method of combination CNN transfer learnings and SVDD |
Non-Patent Citations (6)
| Title |
|---|
| HOWARD LEE等: "Video analysis of human gait and posture to determine neurological disorders", 《EURASIP JOURNAL ON IMAGE AND VIDEO PROCESSING》 * |
| KENT SWEETING ET AL: "Gait and posture-assessment in general practice", 《AUST FAM PHYSICIAN》 * |
| 张良伟: "下肢外骨骼机器人智能鞋设计及其应用", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》 * |
| 石欣等: "基于二次特征提取与SVM的异常步态识别", 《仪器仪表学报》 * |
| 薛召军等: "基于支持向量机的步态识别新方法", 《天津大学学报》 * |
| 黄彬等: "用侧影特征分析和识别人的异常步态", 《计算机工程与应用》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111144166A (en) * | 2018-11-02 | 2020-05-12 | 银河水滴科技（北京）有限公司 | Method, system and storage medium for establishing abnormal crowd information base |
| CN110322528A (en) * | 2019-06-26 | 2019-10-11 | 浙江大学 | Nuclear magnetic resonance brain image reconstructing blood vessel method based on 3T, 7T |
| CN110728226A (en) * | 2019-10-09 | 2020-01-24 | 清华大学 | Gait quantification system and method based on motion recognition |
| CN111062340A (en) * | 2019-12-20 | 2020-04-24 | 湖南师范大学 | Abnormal gait behavior identification method based on virtual posture sample synthesis |
| CN111062340B (en) * | 2019-12-20 | 2023-05-23 | 湖南师范大学 | Abnormal gait behavior recognition method based on virtual gesture sample synthesis |
| CN112597903A (en) * | 2020-12-24 | 2021-04-02 | 珠高电气检测有限公司 | Electric power personnel safety state intelligent identification method and medium based on stride measurement |
| CN112597903B (en) * | 2020-12-24 | 2021-08-13 | 珠高电气检测有限公司 | Electric power personnel safety state intelligent identification method and medium based on stride measurement |
| CN113705482A (en) * | 2021-08-31 | 2021-11-26 | 江苏唯宝体育科技发展有限公司 | Body health monitoring and management system and method based on artificial intelligence |
| CN113705482B (en) * | 2021-08-31 | 2024-03-22 | 江苏唯宝体育科技发展有限公司 | Body health monitoring management system and method based on artificial intelligence |
| CN114533044A (en) * | 2022-02-08 | 2022-05-27 | 北京金史密斯科技股份有限公司 | Gait data acquisition system and method and running equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108416276B (en) | 2022-05-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108416276A (en) | The abnormal gait detection method of side gait video based on people | |
| CN103164694B (en) | A kind of human action knows method for distinguishing | |
| US20210315486A1 (en) | System and Method for Automatic Evaluation of Gait Using Single or Multi-Camera Recordings | |
| Jensen et al. | Classification of kinematic swimming data with emphasis on resource consumption | |
| CN111027417B (en) | Gait recognition method and gait evaluation system based on human body key point detection algorithm | |
| Loureiro et al. | Using a skeleton gait energy image for pathological gait classification | |
| CN109271918B (en) | Method for distinguishing people with balance ability disorder based on gravity center shift model | |
| CN111048205A (en) | Method and device for assessing symptoms of Parkinson's disease | |
| Green et al. | Quantifying and recognizing human movement patterns from monocular video images-part II: applications to biometrics | |
| CN108021888A (en) | A kind of fall detection method | |
| CN114307117B (en) | Standing long jump result measuring method and device based on video | |
| Nawar et al. | Topological descriptors for parkinson’s disease classification and regression analysis | |
| Lee et al. | Video analysis of human gait and posture to determine neurological disorders | |
| Yang et al. | Automatic detection pipeline for accessing the motor severity of Parkinson’s disease in finger tapping and postural stability | |
| CN111419237A (en) | Cerebral apoplexy hand motion function Carroll score prediction method | |
| CN104331705B (en) | Automatic detection method for gait cycle through fusion of spatiotemporal information | |
| Dong et al. | An “optical flow” method based on pressure sensors data for quantification of Parkinson's disease characteristics | |
| CN110032976B (en) | Mask processing based novel gait energy map acquisition and identity recognition method | |
| Saponaro et al. | Estimating Physical Activity Intensity And Energy Expenditure Using Computer Vision On Videos | |
| Lin et al. | A Contactless Healthcare System with Face Recognition | |
| CN106446810A (en) | Computer vision method used for mental state analysis | |
| He et al. | Human behavior feature representation and recognition based on depth video | |
| Vega et al. | Experiments on gait analysis by exploiting nonstationarity in the distribution of feature relationships | |
| Zeng et al. | Machine learning based automatic sport event detection and counting | |
| Patil et al. | Early Detection of Hemiplegia by Analyzing the Gait Characteristics and Walking Patterns Using |
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 |