CN1168044C - Far distance identity identifying method based on walk - Google Patents
Far distance identity identifying method based on walk Download PDFInfo
- Publication number
- CN1168044C CN1168044C CNB011441577A CN01144157A CN1168044C CN 1168044 C CN1168044 C CN 1168044C CN B011441577 A CNB011441577 A CN B011441577A CN 01144157 A CN01144157 A CN 01144157A CN 1168044 C CN1168044 C CN 1168044C
- Authority
- CN
- China
- Prior art keywords
- gait
- training
- walking
- feature
- distance
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000012549 training Methods 0.000 claims abstract description 33
- 238000000605 extraction Methods 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims abstract description 15
- 230000002123 temporal effect Effects 0.000 claims abstract description 15
- 238000004458 analytical method Methods 0.000 claims abstract description 11
- 230000005021 gait Effects 0.000 claims description 102
- 239000000284 extract Substances 0.000 claims description 11
- 230000000007 visual effect Effects 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 9
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000008878 coupling Effects 0.000 claims description 6
- 238000010168 coupling process Methods 0.000 claims description 6
- 238000005859 coupling reaction Methods 0.000 claims description 6
- 238000000513 principal component analysis Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims 2
- 230000007704 transition Effects 0.000 claims 1
- 230000033001 locomotion Effects 0.000 abstract description 27
- 230000009466 transformation Effects 0.000 abstract description 4
- 230000037237 body shape Effects 0.000 abstract 1
- 230000000875 corresponding effect Effects 0.000 description 14
- 238000005516 engineering process Methods 0.000 description 11
- 230000008859 change Effects 0.000 description 9
- 238000012544 monitoring process Methods 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 239000013598 vector Substances 0.000 description 7
- 238000012795 verification Methods 0.000 description 7
- 238000011160 research Methods 0.000 description 5
- 238000009825 accumulation Methods 0.000 description 4
- 238000006073 displacement reaction Methods 0.000 description 4
- 108010076504 Protein Sorting Signals Proteins 0.000 description 3
- 230000006399 behavior Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000011410 subtraction method Methods 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 241000408659 Darpa Species 0.000 description 1
- 208000001613 Gambling Diseases 0.000 description 1
- 230000004791 biological behavior Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 210000000554 iris Anatomy 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001681 protective effect Effects 0.000 description 1
- 230000000284 resting effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
Images
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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a far distance identifying method based on walking, which comprises two processes of training and identification, wherein the training process comprises the steps: obtainment of training walking sequences, division of spatial contours, extraction of shape signals and distance signals, main component analyses, extraction of personalized physical characteristics, extraction of visible personalized characteristics as additional characteristics for the final check of walking classification and obtainment of a trained walking database. In the present invention, one far distance identifying method based on walking behavior is achieved by a statistic principle element analyzing method. An improved background reduction method is proposed for extracting walking movement from the background; walking characteristics are extracted from a movement contour with temporal and spatial variation by characteristic temporal transformation; temporal and spatial correlation match or a nearest neighbour rule is used for the identifying process; some personalized characteristics which are relative to personal body shapes and physical constitution are selected and used for the check of final judgment.
Description
Technical field
The invention belongs to area of pattern recognition, particularly based on the far distance identity identifying method of gait.
Background technology
Along with the develop rapidly of modern science and technology, utilize video camera to monitor the every aspect that dynamic scene is widely used in society already, particularly those are to the occasion of safety requirements sensitivity, as national defence, community, bank, military base etc.The vision monitoring of dynamic scene is the forward position research direction that receives much concern in recent years, and its detection from the video camera sequences of images captured, identification, tracking target are also understood its behavior.Although the present rig camera that extends as human vision ubiquity in commerce is used is not given full play to its initiatively supervision instrumentality in real time.Therefore, the visual monitor system of developing the automated intelligent type with practical significance becomes urgent and necessary day by day.This not only requires to replace human eye with video camera, and requires to monitor or control task to finish with computing machine contributor, replacement people.
The visual analysis of people's motion is active research theme in the computer vision field, and effective important development trend especially that combines of biometrics identification technology and people's motion analysis, one of application that it is potential is exactly the vision monitoring to the security sensitive occasion.Remote people's identification---based on the second generation biometrics identification technology of movement vision, more be subjected to researchers' favor in recent years.For example; the DARPA of U.S. Advanced Research Projects administration subsidized in 2000 major project---HID plans (Human Identification at a Distance); its task is exactly exploitation detection, classification and the identification of multimodal monitoring technique to realize remote situation servant, thereby strengthens the automatic protective capability that national defence, occasion such as civilian are avoided the attack of terrorism.Focus is the identification of face picture, gait or specific behavior at present.In general, the identification in the vision monitoring mainly contains the application of two aspects: the access control of (1) special occasions.In some occasions responsive especially (as military base, national important secret unit etc.), only allow that those personnel with special identity authority enter to safety requirements.Use biometrics identification technology in these occasions, the person's that sets up the Lawful access earlier biological characteristic (as face picture, gait etc.) database, when the people invades, gather these features real-time non-contactly, and come to determine exactly whether the visitor has the right that enters this field according to them.(2) auxiliary law tool of solving a case.On the one hand, it can determine whether to have in the monitoring scene exist (as suspect etc.) of specific people in real time.Public security department can set up some suspects and escaped criminal's biological characteristic, load onto visual monitor system in their the more frequent area (as gambling house, community) of discrepancy and discern the people that each enters the monitoring field, judge whether it is the suspect, if then report to the police automatically at once; The law tool that also can be used as on the other hand afterwards is used to solve a case, such as the bank raid incident, therefore criminal's face picture may be hidden in the image of camera record, but its gait as seen, can judge whether it is real criminal to the discriminating that the suspect carries out gait.
Face picture and gait are two main biological characteristics that are considered to be used for the identification of visual monitor system medium and long distance at present.Face picture identification has obtained extensive studies, but the face picture may be hidden sometimes, perhaps presents less resolution when certain distance and is difficult to identification and identification; And gait has been given prominence to its superiority in this regard, especially under remote situation, the sentience of people's gait, non-infringement, untouchable advantage have made it become a biological behavior characteristic that shows unique characteristics to be used for remote identification.First generation biometrics identification technology has obtained actual application as recognition technologies such as fingerprint, irises in some commerce and legal agency, and also is in the starting stage at present based on the research of the second generation biometrics identification technology of vision.As a kind of new behavioural characteristic, gait also has and is difficult to hide and camouflage, is easy to advantage such as seizure, and its unique appreciable behavioural characteristic when also being certain distance.Gait Recognition is intended to not consider that the posture of walking according to people under the situations such as clothes, visual angle, background carries out people's identification.Because gait is a kind of motor pattern of change in time and space, so its deal with data amount is relatively large.Certainly, as other biological characteristic, gait also is subjected to some such as drunk, conceived, the influence of physical factors such as the joint is injured (is seen documents [1] H Murase and R.Sakai, Moving objectrecognition in eigenspace representation:gait analysis and lip reading, PatternRecognition Letter, 1996,17:155-162.[2] P.Huang, C.Harris and M.Nixon, Human gait recognition in canonical space using temporal templates, in IEE Proc.Of Vision Image and Signal Processing, April 1999,146 (2): 93-100).
Summary of the invention
The purpose of this invention is to provide a kind of automatic gait recognition method based on pivot analysis, it has solved and has utilized the gait behavior to realize remote identification problem.
For achieving the above object, a kind of contactless far distance identity identifying method based on gait comprises training and discerns two processes, and described training process comprises step:
Obtain the training gait sequence;
Space profiles is cut apart;
Extract the shape distance signal;
Principal component analysis (PCA) is used to obtain the feature space matrix;
Personalized corporal characteristic is extracted, and extracts visual individualized feature as supplementary features, is used for the terminal check of gait classification;
The gait data storehouse that acquisition has been trained;
Described identifying comprises:
For the test gait sequence, cut apart the back through space profiles and extract the shape distance signal;
Carry out projection at feature space;
Utilize the gait data storehouse of having trained to carry out temporal and spatial correlations coupling or the regular Gait Recognition that combines with the gait classification checking procedure of individualized feature of arest neighbors.
Contactless remote identification research based on movement vision is used particularly important for the vision monitoring of security sensitive occasion.Utilize the statistics pca method, the present invention has realized a remote identification system based on gait.Improved background subtraction method at first is suggested and is used for extracting gait motion from background; Motion outline shape with change in time and space is through the extraction of feature space conversion with the realization gait feature; Identifying adopts temporal and spatial correlations coupling or arest neighbors rule, and some individualized features relevant with physique with individual body also are selected for the verification of final classification.
Description of drawings
Fig. 1 is based on the identification algorithm block diagram of gait, comprises training and discerns two parts;
Fig. 2 is the process flow diagram of cutting apart of gait motion profile;
Fig. 3 is that the profile of body gait motion is cut apart example, comprises (a) background image; (b) current input image; (c) image after difference and the binaryzation; (d) image after final profile is cut apart;
Fig. 4 is a gait original-shape feature extraction process flow diagram;
Fig. 5 is a gait original-shape feature extraction synoptic diagram, comprises that (a) profile Boundary Extraction reels and (b) normalized distance signal with separating;
Fig. 6 is spatiotemporal motion pattern (first row: input gait sequence of a gait; Second row: the motion outline sequence after cutting apart; The 3rd row: profile Boundary Extraction and centroid calculation; The 4th row: carry out the distance signal sequence after the normalization on amplitude and the length);
Fig. 7 is two kinds of various combination distribution plans of individualized feature, and the point of identical sign is represented the result of same individual's different gait sequences, and has the square sign of same color to represent its corresponding average result with it;
Fig. 8 is the amplitude and the accumulation variance curve figure of eigenwert in the feature space;
Fig. 9 is average shape (a) and preceding 8 characteristic distance signals (b-i) of training sample;
Figure 10 is the projected footprint example of training gait in feature space, for visual, has only shown the situation in three-dimensional feature space.
The embodiment of invention
Describe each related detailed problem in the technical solution of the present invention in detail below in conjunction with accompanying drawing.
Body gait motion segmentation based on background subtraction
From the gait motion that extracts the people the background is a critical step for gait analysis.At present change detecting method mainly contains three kinds of background subtractions, time difference, light stream, because the background subtraction method not only realizes simply, and effective especially for the variation of sensed luminance, so adopted change detection algorithm based on background subtraction in this programme.Generally speaking, it comprises structure, mathematics difference operation, three steps of appropriate threshold selection of background image.
The present invention uses LMedS (Least Median of Squares) method to make up background image reliably from the parts of images sequence.Traditional background image structure the average image service time, but this method is relatively more responsive for noise, illumination etc., and LMedS method robust more comparatively speaking.Make I represent a sequence that comprises the N two field picture, then background image B
XyCan be expressed as
Wherein, q is that (x y) locates gray-scale value to be determined to pixel.
It usually is to finish by the difference of present image and background image that brightness changes.Yet the selection of binary-state threshold but is very difficult, especially for the image of low contrast, is difficult to from noise moving target fully be extracted owing to the brightness variation is too low.In order to address this problem, we have defined an extraction function and have carried out difference operation indirectly
Wherein, (x, y) (x is respectively that present image and background image are at pixel (x, the brightness value of y) locating y) to a with b.This function can detect its susceptibility according to the brightness value of background image to be changed.For every width of cloth image, the distribution of this extraction function is easy to calculate.Utilize traditional histogram method, we can extract from this and determine an appropriate threshold histogram of function.Then, obtain corresponding variation pixel by the difference between this extraction function and the threshold value relatively.
In post-processed, the present invention uses further filtered noise point of morphological operation, and uses a binaryzation connected component to analyze to extract simply connected moving target exactly.The profile of gait motion cut apart process flow diagram and one as a result example respectively as accompanying drawing 2 with shown in the accompanying drawing 3.This method can be applied to cutting apart of any mobile target in complex background fully.
The original-shape feature extraction of gait spatiotemporal mode
An important clue determining the inherent motion of a pedestrian is exactly the variation of human body contour outline shape along with the time.In order to remove information redundance and to reduce computation complexity, we express the change in time and space pattern of gait motion approx with the distance signal sequence that the change of shape of these two-dimentional gait profiles is converted to one dimension.
The extraction process flow diagram of original-shape distance feature and synoptic diagram are respectively shown in accompanying drawing 4 and accompanying drawing 5.After being partitioned into people's motion outline, use the border that can obtain it based on the edge following algorithm of connectedness, calculate the barycenter on its profile border then
Wherein, (x
c, y
c) be center-of-mass coordinate, N
bBe the boundary pixel number, (x
i, y
i) be borderline pixel.
Select crown corresponding point as the reference point, along counterclockwise we can launch the border and are one and put the signal that the distance of its corresponding barycenter is formed by boundary pixel
In order to eliminate the influence of yardstick, length, use L norm and the double sampling method uniformly-spaced signal characteristic of adjusting the distance to carry out normalization on amplitude and the length for training and identification.The original-shape characteristic extraction procedure example of the motion spatiotemporal mode of a gait as shown in Figure 6.The extraction and the expression of simple more gait primitive character will greatly reduce the calculation cost of entire method.
Feature space conversion based on pivot analysis
Given s training classification, each classification is represented the formed raw range burst of a people's gait pattern.Everyone a plurality of gait sequences can optionally increase and not need to change following training process based on pivot analysis.
Make D
I, jBe j the distance signal feature of class i, and N
iBe the distance signal number of class i, then total number of training is N
t=N
1+ N
2+ ... + N
s, whole training set is [D
1,1, D
1,2..., D
1, N1, D
2,1..., D
S, Ns], the average m of this training set
dWith overall variance matrix be
If rank of matrix is N, then utilize the svd theory can draw N nonzero eigenvalue λ
i, λ
2..., λ
NAnd characteristic of correspondence vector e
1, e
2..., e
N
Generally speaking, the several bigger eigenwert characteristic of correspondence vectors in front are corresponding to the bigger variation of training mode, and more the proper vector of high-order is being represented little variation.For the validity of storage and calculating is considered that we use a threshold value T
sIgnore those little eigenwerts and characteristic of correspondence vector thereof
Wherein, W
kIt is the accumulation variance of a preceding k eigenwert.Select k＜N the maximum pairing proper vector of eigenwert, we can construction feature transformation matrix E be [e
1, e
2..., e
k], each distance signal D then
I, jIn feature space, will be projected as 1 P
I, j
P
i，j＝[e
1?e
2…e
k]
TD
i，j (8)
Correspondingly, each gait sequence will be rendered as a track in feature space.As seen, the feature space analysis has greatly reduced the dimension of sample, and it only need keep k main characteristic component the most effective and express original sample.For each gait sequence that participates in training, the averaging projection of distance signal sequence in feature space of its primitive character is
The extraction of individualized feature
In fact, when human eye is discerned a people's gait, usually merged some other visual feature, such as people's leg speed, step-length, height and the bodily form etc.In order to improve the accuracy of identification, the terminal check that is used for gait classification corresponding to the individualized feature of these factors as supplementary features has been extracted in this invention.
Everyone gait has a basic driving frequency usually, and this is because the various piece of human body must be with a kind of coordinated mode motion.Consider that tangential movement has bigger variation than vertical movement, thereby stronger distinguishable power arranged that we have selected normalized horizontal velocity and its variance to reflect the different variations of leg speed.The step-length and the bodily form between the individuality also vary with each individual, and we select the maximum of gait motion profile, minimum the ratio of width to height to reflect this variation.In addition, the position that the individual height and the bodily form affect its barycenter, this can recently portray by the height of barycenter and human body.
In a word, we have selected following several individualized feature indirectly: normalized horizontal velocity H and its variance V (the average displacement by profile barycenter between the normalization picture frame is similar to), the aspect ratio C of the maximum of the motion outline of yardstickization, minimum the ratio of width to height Max, Min and human body and barycenter.Accompanying drawing 7 has represented the distribution plan example of two kinds of individualized feature various combinations, and wherein the point of identical sign is represented the result of the asynchronous attitude sequence of same individual, and square sign is represented its corresponding average result.As seen, effective combination of individualized feature can bring certain adjudicated ability for gait classification.
The Gait Recognition that classification combines with the individualized feature verification based on temporal and spatial correlations or arest neighbors
Gait Recognition is a traditional classification problem, and this can finish by its similarity of tolerance between reference gait pattern and cycle tests.Two kinds of recognition methodss below the present invention has attempted in feature space, and obtain recognition result exactly in conjunction with the checking procedure of personalized supplementary features.
Temporal and spatial correlations STC (Spatio-temporal correlation) is be correlated with a extend type in the three-dimensional space-time territory of two dimensional image.The input test sequence that a given length is T, it is converted into one-dimensional distance burst I (t) at pretreatment stage, t=1 wherein, 2 ..., T, its being projected as in feature space
P(t)＝[e
1?e
2…e
k]
TI(t) (10)
Then the similarity between the reference sequences R (t) in list entries P (t) and the database can be calculated as
Wherein, R
Ij(t) representative i people's in the K dimensional feature space j is individual with reference to gait pattern.The displacement of considering the time is with Shen Shrink, and we have preserved a plurality of with reference to gait pattern R in advance in database by the method for linear displacement and interpolation
Ij(at+b), wherein parameter a and the selection of b depend on the different of the variation of speed and phase place respectively.Usually, minimize the selected result of i of this distance metric function as identification.
If the displacement of considering the time is with Shen Shrink, then the calculation cost of temporal and spatial correlations will greatly increase.In order to eliminate this problem, we also can fall back on the identification that realizes gait in arest neighbors rule NN (NearestNeighbor).As if adopting normalized Euclidean distance tolerance, then cycle tests P (t) accumulates with the distance of the C of averaging projection (i) of class i and is
Generally, select the minimum value of d (i) as sorting result.
When only using the method for STC or NN to discern, may not obtain result accurately.From the observation of reality, when the distance between two minimum value of above-mentioned similarity measurement function relatively near the time, identification tends to make a mistake, however correct result but is time minimum value usually.Therefore we utilize the terminal check that carries out gait classification at the physical features of the personalization that training period obtained.Promptly when the absolute distance between two minimum value during less than an appropriate threshold, we will increase this step of verification and obtain final court verdict.
In order to verify the recognition performance of this method, we select SOTON gait data storehouse to carry out the Gait Recognition test.This sample database adds up to 28 gait sequences from the University of Southampton of Britain.All images is taken with the speed of per second 25 frames, and original size is 384*288.Everyone sequence is selected for training process, and after pre-service, all sequences all is converted into corresponding one-dimensional distance burst before training and identification.
Training process based on PCA is performed, and accompanying drawing 8 has been showed the amplitude and the corresponding accumulation variance curve of eigenwert in the feature space that is obtained after the training.We have selected preceding 13 eigenwerts (the accumulation variance has been higher than 95%) and characteristic of correspondence vector to come structural attitude spatial alternation matrix.Therefore each distance signal can both be expressed by the linear combination of these 13 main proper vectors.Accompanying drawing 9 has been showed mean distance signal and the first eight character shape distance signal of all training samples, and accompanying drawing 10 has provided the characteristic locus of four training gait sequences.For visual, we have only shown the situation in three-dimensional feature space.As we know from the figure, gait motion has symmetry and periodic characteristics.
Use temporal and spatial correlations coupling and arest neighbors rule, we have assessed the recognition performance of this method, and their recognition result is respectively 90.5% and 89.3%.If the individualized feature corresponding to individual gait is used for carrying out classification check, then discrimination all rises to 100%.
The present invention and list of references [1] [2] have been done a detailed comparison: 1) gait data storehouse: the NIT database that uses in [1] has a significantly limitation, being that everyone wears same clothes and shoes, is infeasible for this identification for reality is used.And UCSD database that uses in [2] and our the SOTON database of use do not have such restriction.In addition, the image background of SOTON database is complicated relatively more.2) feature selecting: the feature of using in [1] is the two dimensional motion contour images.The present invention selects the time of contour shape to change as essential characteristic, and they are converted to the one-dimensional distance signal from two dimensional image, can remove redundance, the minimizing calculation cost of information like this.Simultaneously, some physical traits corresponding to individual's personalization are considered to be used for the verification of gait classification.[2] feature of using in is the light stream image that obtains from gait sequence.As everyone knows, the calculating of light stream is quite complicated, and also very sensitive for the variation of noise and frame per second.3) training process: the training sample of [1] comprises everyone 5 sequences, and the method for this method and [2] is used everyone 1 sequence respectively.The present invention extracts personalized physical features and is used for the gait verification, and [2] increase the separable degree that canonical tanalysis further reduces dimension and optimizes class.4) discrimination: if only consider temporal and spatial correlations identification, the discrimination of [1] is higher relatively.Requirement for restriction when perhaps, this has benefited from picture catching on the one hand can be eliminated the influence of clothes for Gait Recognition to a certain extent; It has preserved more relatively reference gait template on the other hand, and perhaps these reference templates have included the gait pattern that the someone more may change.Have reason to believe that if increase reference template, discrimination of the present invention will improve rapidly.For the method [2] of the basic reference template that has similar number, its discrimination and this method are more approaching.Although the light stream image has implied time-varying information, the temporal and spatial correlations coupling can remedy the deficiency of these resting shape parameters equally well.When [2] increased canonical tanalysis, it had obtained 100% discrimination.Similarly, the present invention has also obtained 100% recognition accuracy after having increased the checking procedure of personalized supplementary features.5) calculation cost: minimum calculation cost is this method biggest advantage, and this mainly has benefited from simple more feature selecting.The primitive character that the present invention uses is the distance signal that contour edge is separated coiling, and size is normalized to 1*480.[1] primitive character is the contour images that is of a size of 64*64.[2] feature of Shi Yonging is the light stream image of 64*64, and optical flow computation itself is exactly very complicated.6) adaptability: the specific (special) requirements in [1] during picture catching explains for practical application that to show be a limitation.Although algorithm of the present invention is simple, certain adaptability is arranged for noise and frame losing.[2] optical flow computation is very sensitive for the velocity variations of noise and frame per second, and calculates also quite expensive.In a word, technology algorithm of the present invention is all having certain superiority aspect feature selecting, calculation cost, recognition accuracy and the adaptability.
Embodiment
Whole proposal mainly comprises training and discerns two processes, and in order to describe the embodiment of this invention in detail, we are controlled to be example with entering of intelligent room and are illustrated.Suppose that intelligent room is one and enters the secret occasion that requires authority, it only allows that some specific people can enter.Therefore, the video camera that we can install and fix in the corridor of room inlet is caught messenger's gait sequence, and utilizes the present invention to realize authentication work based on gait.If this visitor can be discerned effectively, then door can be opened automatically, allows to enter; Otherwise, refusal is entered, perhaps increase the warning measure.
With regard to training process, our purpose is the gait feature database that will at first create corresponding to these privileged people.At first utilize rig camera to catch a series of gait sequences of these people, we are easy to make up the background image in field that video camera is monitored according to these sequences, and can be updated periodically.For these gait sequences, utilize the cutting techniques of above-mentioned body gait motion, the original-shape Feature Extraction technology of body gait spatiotemporal mode to carry out pre-service.The a series of distance signals that obtained then obtain main characteristic component by the training technique of above-mentioned feature space conversion based on pivot analysis.The selected conduct of some gait sequences of different people with reference to gait with its projection in feature space; Simultaneously, utilize everyone different sequences to use the extractive technique of individualized features to obtain these supplementary features corresponding to the individual.In a word, the gait data storehouse after the training should comprise feature space transformation matrix that main characteristic component forms, everyone a plurality of with reference to the gait projection properties and corresponding to individual's adding personalized feature.
With regard to identifying, at first utilize rig camera to catch messenger's gait sequence rapidly, through obtaining the original-shape feature of this cycle tests after the series of preprocessing technology such as identical background constructing, gait extraction, distance feature extraction.This primitive character utilizes the feature space transformation matrix in the database to carry out projection then, and everyone a plurality of in its projection and the database are discerned by temporal and spatial correlations or arest neighbors rule with reference to the gait projection properties, utilize individualized feature that this cycle tests obtains and everyone individualized feature in the database to carry out verification work simultaneously.If this visitor's gait can be correctly validated, then control the unlatching in rooms automatically; Otherwise, refuse to enter or increase warning.
In a word, based on pca method, the present invention proposes a kind of simple and effective automatic Gait Recognition algorithm.At first, improved background subtraction method is used to extract exactly pedestrian's profile; Then, time dependent contour shape passes through the feature space converter technique to realize Feature Extraction; At last, identifying has adopted temporal and spatial correlations coupling or arest neighbors rule, and some additional individualized features are selected for the verification of conclusive judgement.Test findings on the SOTON database has been verified the validity of our algorithms.The present invention is easy to realize, the performance robust.For using based on the remote authentication of gait under the constrained environment, the present invention has advanced the application of the based drive biometrics identification technology of the second generation in the vision monitoring to a certain extent.The present invention can be applied to the visual monitor system of some limited occasions, can be used as also that law tool is auxiliary solves a case etc.
Claims (5)
1. the contactless far distance identity identifying method based on gait comprises training and discerns two processes, and described training process comprises step:
Obtain the training gait sequence;
Space profiles is cut apart;
Extract the shape distance signal;
Principal component analysis (PCA) is used to obtain the feature space matrix;
Personalized corporal characteristic is extracted, and extracts visual individualized feature as supplementary features, is used for the terminal check of gait classification;
The gait data storehouse that acquisition has been trained;
Described identifying comprises:
For the test gait sequence, cut apart the back through space profiles and extract the shape distance signal;
Carry out projection at feature space;
Utilize the gait data storehouse of having trained to carry out temporal and spatial correlations coupling or the regular Gait Recognition that combines with the gait classification checking procedure of individualized feature of arest neighbors.
2. by the described method of claim 1, it is characterized in that described space profiles is cut apart and comprise step:
Utilize the part gait sequence to carry out the structure of background image;
The background image of structure and the image of current input are carried out difference operation;
Image after difference operation and the binaryzation is carried out filtering;
Image to filtering carries out the connected component analysis.
3. by the described method of claim 1, its feature also is to comprise the step of using following formula to carry out after the difference operation and carrying out binary conversion treatment:
4. by the described method of claim 1, it is characterized in that described shape distance signal extracts, and comprises step:
Extract the border of gait profile;
With the boundary transition of the extracting one-dimensional distance signal that all profile frontier points form to its centroid distance of serving as reasons;
The signal of adjusting the distance carries out the normalization on amplitude and the size.
5. by the described method of claim 1, it is characterized in that the extraction of described personalized corporal characteristic, these features have reflected the different of leg speed, step-length and the bodily form between the individuality indirectly.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB011441577A CN1168044C (en) | 2001-12-13 | 2001-12-13 | Far distance identity identifying method based on walk |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB011441577A CN1168044C (en) | 2001-12-13 | 2001-12-13 | Far distance identity identifying method based on walk |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1426020A CN1426020A (en) | 2003-06-25 |
| CN1168044C true CN1168044C (en) | 2004-09-22 |
Family
ID=4677347
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB011441577A Expired - Fee Related CN1168044C (en) | 2001-12-13 | 2001-12-13 | Far distance identity identifying method based on walk |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1168044C (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101226597B (en) * | 2007-01-18 | 2010-04-14 | 中国科学院自动化研究所 | Method and system for recognizing nights pedestrian based on thermal infrared gait |
| CN100585656C (en) * | 2007-03-14 | 2010-01-27 | 中国科学院自动化研究所 | An all-weather intelligent video analysis monitoring method based on a rule |
| CN101334845B (en) * | 2007-06-27 | 2010-12-22 | 中国科学院自动化研究所 | Video frequency behaviors recognition method based on track sequence analysis and rule induction |
| DE102008043953A1 (en) * | 2008-11-21 | 2010-05-27 | Robert Bosch Gmbh | Device for generating and / or processing an object signature, monitoring device, method and computer program |
| CN101587541B (en) * | 2009-06-18 | 2011-02-02 | 上海交通大学 | Character recognition method based on human body contour outline |
| CN101609507B (en) * | 2009-07-28 | 2016-03-09 | 中国科学技术大学 | Gait recognition method |
| CN103778439B (en) * | 2014-01-23 | 2016-08-17 | 电子科技大学 | Human body contour outline reconstructing method based on dynamic space-time information excavating |
| CN105389549B (en) * | 2015-10-28 | 2019-08-13 | 北京旷视科技有限公司 | Object identifying method and device based on human action feature |
| CN106203379A (en) * | 2016-07-20 | 2016-12-07 | 安徽建筑大学 | Human body recognition system for security |
| CN106850955B (en) * | 2016-12-20 | 2019-07-02 | 陕西尚品信息科技有限公司 | A kind of mobile phone identity verification method based on Gait Recognition |
| CN108013998A (en) * | 2017-12-12 | 2018-05-11 | 深圳市罗伯医疗科技有限公司 | A kind of lower limb rehabilitation instrument training method and system |
| CN112818644A (en) * | 2020-01-09 | 2021-05-18 | 赵国良 | Real-time conversion system for writing revision format |
| CN115203663B (en) * | 2022-06-25 | 2023-08-22 | 北京基智科技有限公司 | Small-view-angle long-distance video gait accurate identification identity authentication system |
| CN117097628B (en) * | 2023-10-19 | 2023-12-22 | 中国电子科技集团公司第五十四研究所 | Networking communication behavior identification method based on signal physical characteristic parameters |
-
2001
- 2001-12-13 CN CNB011441577A patent/CN1168044C/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN1426020A (en) | 2003-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20240168563A1 (en) | Information processing device and method, program and recording medium for identifying a gesture of a person from captured image data | |
| Gajjar et al. | Human detection and tracking for video surveillance: A cognitive science approach | |
| Barbosa et al. | Re-identification with rgb-d sensors | |
| Ge et al. | Real-time pedestrian detection and tracking at nighttime for driver-assistance systems | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| Wang et al. | A new attempt to gait-based human identification | |
| CN1168044C (en) | Far distance identity identifying method based on walk | |
| CN101226597B (en) | Method and system for recognizing nights pedestrian based on thermal infrared gait | |
| D'Angelo et al. | People re-identification in camera networks based on probabilistic color histograms | |
| US20080187172A1 (en) | Tracking Apparatus And Tracking Method | |
| CN102521565A (en) | Garment identification method and system for low-resolution video | |
| CN104794451B (en) | Pedestrian's comparison method based on divided-fit surface structure | |
| CN103942577A (en) | Identity identification method based on self-established sample library and composite characters in video monitoring | |
| WO2014092552A2 (en) | Method for non-static foreground feature extraction and classification | |
| Singh et al. | Human activity recognition based on silhouette directionality | |
| CN107230267A (en) | Intelligence In Baogang Kindergarten based on face recognition algorithms is registered method | |
| KR100824757B1 (en) | Gait recognition method | |
| CN111145223A (en) | Multi-camera personnel behavior track identification analysis method | |
| CN111401308B (en) | Fish behavior video identification method based on optical flow effect | |
| Bhuiyan et al. | Person re-identification by discriminatively selecting parts and features | |
| CN109858308B (en) | Video retrieval device, video retrieval method, and storage medium | |
| Taha et al. | Exploring behavior analysis in video surveillance applications | |
| Rafi et al. | A parametric approach to gait signature extraction for human motion identification | |
| Panagiotakis et al. | Shape-based individual/group detection for sport videos categorization | |
| Budzan | Human detection in thermal images using low-level features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20040922 Termination date: 20171213 |