CN103605362A - Learning and anomaly detection method based on multi-feature motion modes of vehicle traces - Google Patents

Abstract

The invention provides a method for learning and anomaly detection of trace modes by utilizing much feature information of a trace. Firstly, in the trace mode learning phase, similarities of motion directions and spatial positions between traces are considered at the same time, a typical trace motion mode is extracted by hierarchical agglomerative clustering, and is provided with high cluster accuracy; and the time efficiency is greatly improved through constructing a Laplacian matrix and reducing the dimensionality of the matrix. Then in the abnormity detection phase, a distribution area of scene starting points is learned through a GMM model, a moving window is used as a basic comparing element, differences of a trace to be detected and a typical trace in position and direction are measured by defining a position distance and a direction distance, and an on-line classifier based on the direction distance and the position distance is established. That the trace belongs to a starting point abnormity, a global abnormity or a local abnormity is determined online through a multi-feature abnormity detection algorithm; and due to the fact that starting point, direction and position feature differences are considered at the same time, and the global abnormity and the local child segment abnormity are considered, the learning and anomaly detection method based on multi-feature motion modes of the vehicle traces is higher in abnormity recognition rate when being compared to traditional methods.

Description

Motor pattern study and method for detecting abnormality based on the many features of track of vehicle

Technical field

What the present invention relates to is motor pattern learning method and the online abnormal track-detecting method based on the many features of track of vehicle.First by hierarchical cluster from coarse to fine, extract orbiting motion pattern, every layer adopts respectively Bhattacharyya distance and the improvement Hausdorff based on line segment interpolation apart from the similarity of weighing direction of motion and locus between track, and introduces Laplacian and shine upon to reduce computation complexity and automatically determine every layer of clusters number.On this basis, consider track to be measured and the motor pattern difference in starting point distribution, position and direction simultaneously, by the starting point distributed model of study and the sorter of position-based distance and direction distance, judge online starting point, the overall situation and local anomaly.

Background technology

In recent years, in the Study of intelligent of video monitoring system, behavioural analysis and identification based on movement objective orbit become study hotspot, and its learning monitoring scene novel trajectory pattern and abnormality detection are the important contents of research.Especially in intelligent traffic monitoring, the driving trace of vehicle is containing abundant characteristic information, vehicle can be along fixing road and the direction running of appointment under normal circumstances, movement locus shows higher repeatability and similarity, by the normal trace motion model with study, relatively just can detect automatically the abnormal behaviours such as retrograde, U-shaped turning, compare traditional mark manually abnormal, improved greatly abnormality detection efficiency.

In track pattern learning method, by unsupervised clustering algorithm, extract the method for exemplary trajectory motor pattern and be used widely, conventional have spectral clustering, hierarchical clustering, fuzzy K mean cluster and a k-medoids algorithm etc.But traditional track sorting algorithm only considers to utilize single track characteristic to weigh similarity between track, is applied under complicated monitoring scene, and trajectory model discrimination is low.

In method for detecting abnormality, be mainly to set up normal trace model, learning model parameter, mates track to be measured to judge whether with model abnormal.Topmost method has two kinds: the method based on single Gauss model and the method based on HMM model.(1) the former is by the Statistical distribution model of a series of single Gauss model study normal trace, set up Bayes classifier, then by the online method for detecting abnormality increasing progressively, identify abnormal behaviour, but only considered trajectory range malposition, do not consider that direction is abnormal; (2) the latter sets up locus model by C-HMM, each normal trace cluster is divided into several regions, with GMM, learn the model parameter of each HMM state, set abnormal threshold value, input using track to be measured as model judges that track is abnormal, the method is larger abnormal of checkout discrepancy roughly, for complicated local subsegment, is extremely difficult to identification.

This patent is to above problem, and this patent has proposed to utilize a plurality of characteristic informations of track to carry out the method for trajectory model study and abnormality detection.First in the track pattern learning stage, this patent is considered direction of motion and the locus similarity between track simultaneously, and the Agglomerative Hierarchical Clustering that carries out layering extracts typical orbiting motion pattern, therefore has higher cluster accuracy rate; By structure Laplacian matrix dimensionality reduction, greatly improved time efficiency.Then in the abnormality detection stage, this patent is first by GMM model learning scene starting point distributed areas, using Moving Window as basic comparing unit again, definition position distance and direction, apart from weighing the difference of track to be measured in position and direction, are set up the online classification device based on direction distance and positional distance; By the many feature abnormalities detection algorithm proposing judge online that the starting point of track is abnormal, global abnormal and local anomaly, because consider starting point, direction and the position feature difference of track simultaneously, consider that again global abnormal and local subsegment are abnormal, therefore compare classic method, this patent has higher abnormal discrimination.

Summary of the invention

This patent mainly comprises two aspects: first invented a kind of unsupervised many characteristic locuses pattern learning method and extracted typical orbiting motion pattern; Then invented on this basis that a kind of online many feature abnormalities detection method detects that the starting point of track is abnormal simultaneously, local anomaly, global abnormal.

One, unsupervised many characteristic locuses pattern learning method

First the present invention provides a kind of pattern learning method based on the many features of track, by consider direction of motion and locus between track simultaneously, weigh the similarity between track, the Agglomerative Hierarchical Clustering that carries out layering extracts typical orbiting motion pattern, and improves hierarchical clustering algorithm efficiency by introducing Laplacian matrix.Concrete pattern learning framework as shown in Figure 1.

Specific implementation step of the present invention is as follows:

1, similarity measurement between many feature extractions and track

The present invention makes full use of track position and direction character information is weighed the similarity between track, adopts respectively IMHD distance and Bhattacharyya apart from the similarity of calculating between track.Pretreated effective track can be expressed as:

$T_i=\{{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="HSA0000095043070000013.png,HSA0000095043070000031.png"\; state="\{\{state\}\}">t</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="HSA0000095043070000013.png,HSA0000095043070000023.png"\; state="\{\{state\}\}">t</figure-callout>}}_2,...,t_j,...,t_{N_i}\}=\{(x_1,y_1),(x_2,y_2),...,(x_j,y_j),...,(x_{N_i},y_{N_i})\}$

T wherein _jrepresent track T _ij sampled point, N _irepresent course length.(x _j, y _j) represent that j sampled point is at the two-dimensional position coordinate of the plane of delineation.

1) orbiting motion direction similarity measurement

Trajectory direction feature extraction as shown in Figure 2, defines m _j=(x _j+1-x _j, y _j+1-y _j), represent direction vector between neighbouring sample point; m ₀=(1,0), represents direction level vector of unit length to the right.Track T _ij sampled point deflection can be expressed as:

${\mathrm{\&theta;}}_j=\left\{\begin{array}{c}\frac{{\cos }^{-1}(m_j\&CenterDot;m_0)}{\left|m_j\right|\&CenterDot;\left|{\mathrm{<figure-callout\; id="0"\; label="m"\; filenames="HSA0000095043070000022.png,HSA0000095043070000023.png"\; state="\{\{state\}\}">m</figure-callout>}}_0\right|}\&CenterDot;\frac{180}{\mathrm{\&pi;}},\mathrm{if}{y}_{j+1}-y_j\&GreaterEqual;0\\ (2\mathrm{\&pi;}-\frac{{\cos }^{-1}(m_j\&CenterDot;m_0)}{\left|m_j\right|\&CenterDot;\left|{\mathrm{<figure-callout\; id="0"\; label="m"\; filenames="HSA0000095043070000022.png,HSA0000095043070000023.png"\; state="\{\{state\}\}">m</figure-callout>}}_0\right|})\&CenterDot;\frac{180}{\mathrm{\&pi;}},\mathrm{if}{y}_{j+1}-y_j<0\end{array}\right.(1\&le;j<N_i-1)---\left(1\right)$

θ wherein _j∈ (0,360), is divided into N (N=18) individual uniformly-spaced sub-range (I trajectory direction angular region (0,360) ₁, I ₂... I _n).Each sub-range length is △ θ=360/N=20 degree, track T _iall deflection θ _jbe mapped to corresponding sub-range.Track T _jdeflection is distributed in interval I _qprobability be p _q=M _q/ M (M _q: θ _j∈ I _qnumber, M: track T _ideflection number).

Track T _idirection character can be expressed as

track T has been described _istatistics directional information.Adopting Bhattacharyya apart from weighing direction of motion similarity between track is:

$\mathrm{Dire}(T_i,T_j)={[1-\underset{q=1}{\overset{N}{\mathrm{\&Sigma;}}}\sqrt{\mathrm{Dire}{\left(T_i\right)}_q\mathrm{Dire}{\left(T_j\right)}_q}]}^{1/2}\&Element;\left[\mathrm{0,1}\right]---\left(2\right)$

Dire (T wherein _i) _qand Dire (T _j) _qrepresent respectively track T _iand T _jdeflection be distributed in the probability in q deflection interval.Dire (T _i, T _j) more close to 1, represent that two orbiting motion directions are more similar; Dire (T _i, T _j) be more bordering on 0, contrary.

2) trajectory range location similarity is measured

By target following, can directly obtain track position feature, this patent adopts the improvement Hausdorff distance (IMHD) based on line segment interpolation to weigh the similarity of trajectory range position, and track line segment interpolation model as shown in Figure 3, represents track T with broken line _iand T _j: $\stackrel{\&OverBar;}{T_i}=\{\stackrel{\&OverBar;}{t_1{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="HSA0000095043070000013.png,HSA0000095043070000023.png"\; state="\{\{state\}\}">t</figure-callout>}}_2},\stackrel{\&OverBar;}{t_2{\mathrm{<figure-callout\; id="3"\; label="t"\; filenames="HSA0000095043070000023.png"\; state="\{\{state\}\}">t</figure-callout>}}_3},...,\stackrel{\&OverBar;}{t_{a-1}{t}_a},...,\stackrel{\&OverBar;}{t_{N_{i-1}}{t}_{N_i}}\},\stackrel{\&OverBar;}{T_j}=\{\stackrel{\&OverBar;}{t_1{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="HSA0000095043070000013.png,HSA0000095043070000023.png"\; state="\{\{state\}\}">t</figure-callout>}}_2},\stackrel{\&OverBar;}{t_2{\mathrm{<figure-callout\; id="3"\; label="t"\; filenames="HSA0000095043070000023.png"\; state="\{\{state\}\}">t</figure-callout>}}_3},...,\stackrel{\&OverBar;}{t_{b-1}{t}_b},...,\stackrel{\&OverBar;}{t_{N_{j-1}}{t}_{N_j}}\},$ In IMHD algorithm, track T _isampled point t _ato T _jbee-line is expressed as:

Wherein || || represent Euclidean distance between sampled point, t _abe mapped to

the vertical interpolation point of corresponding line segment, in Fig. 3, dot, known according to mathematical principle, if existed

for t _ato T _jbee-line, otherwise traversal track T _jall sampled points, find minor increment

Last track T _ito T _jdistance be:

track T _iand T _jlocus distance be:

D _IMHD(T _i，T _j)=max(h(T _i，T _j)，h(T _j，T _i)) (4)

2, many characteristic locus classification

This patent is introduced the Laplacian mapping thought of spectral clustering in Agglomerative Hierarchical Clustering algorithm, and structure Laplacian matrix also carries out Eigenvalues Decomposition.According to the perturbation theory of matrix, by the automatic hard clustering number of the difference between eigenwert k; Original higher-dimension track data is mapped to a bit in new k dimensional feature space simultaneously, with the unit of low dimensional feature vector, usually represents original track data, greatly reduce the dimension of track.In conjunction with many characteristic similarities of track comparison above, a kind of many feature hierarchies clustering algorithm based on Laplacian Matrix has been proposed, as shown in empty frame in Fig. 1.

If the set that contains m effective track is Ω _traj={ T ₁, T ₂..., T _m, algorithm 1 is described below:

1) the thick cluster stage---based on direction of motion cluster

Step1 uses Bhattacharyya apart from calculating track T _i, T _jbetween direction of motion similarity Dire (T _i, T _j), the similar matrix W with gaussian kernel function structure based on direction of motion _dire∈ R ^{m * m}, w wherein _ij=exp (Dire (T _i, T _j) ²/ 2 σ ²), σ is scale parameter.

Step2 structure standard Laplacian matrix

d wherein ₁for diagonal matrix, matrix element is

Step3 is to L _direcarry out Eigenvalues Decomposition, eigenwert is numbered to λ by descending sort ₁>=λ ₂>=...>=λ _m, calculate the poor of adjacent feature value, if i eigenwert and i+1 eigenwert difference are maximum, determine thick cluster number

Step4 structure m * k matrix L=[l ₁, l ₂..., l _k], l wherein ₁, l ₂..., l _kfor front k eigenwert characteristic of correspondence vector.Each row of L is carried out to unit processing, obtain matrix X, wherein

track set after dimensionality reduction is Ω _traj'={ x ₁, x ₂..., x _m, x ₁, x ₂..., x _meach row vector of homography X, represents R respectively ^ka bit of space.

Step5 is to x ₁, x ₂..., x _mcarry out Agglomerative Hierarchical Clustering, use similarity between two bunches of minimum distance calculation, when finally merging into k bunch, iteration stops, and obtains k centre cluster { O ₁, O ₂, O _i..., O _k.

2) the thin cluster stage---based on locus cluster

Step6 is to cluster O in the middle of each _ithe former track of middle correspondence uses IMHD apart from the similar matrix W constructing based on locus _iMHD, w wherein _ij=exp (D _iMHD(T _i, T _j) ²/ 2 σ ²).Repeat Step2 to Step5 step, determine each cluster O _icluster number q _i, construct low dimensional feature space, last cluster obtains cluster { C ₁, C ₂, C _k, total cluster number

Two, online many feature abnormalities detection method

1, abnormal track is described

The present invention is from the character of abnormal trajectory generation, in conjunction with many feature abnormalities detection method below, without any priori in the situation that, the degree and the character that according to abnormal track, depart from proper motion pattern, defined starting point abnormal, three kinds of common Exception Types of global abnormal and local anomaly.Instantiation as shown in Figure 4, has track motor pattern C in figure ₁and C ₂, track T _cand T _dlocus and direction of motion all run counter to pattern C ₁, be global abnormal; And track T _aand T _blocus meet respectively pattern C ₁and C ₂, but T _adirection and pattern C ₁on the contrary, T _bfrequent break-in, institute thinks local anomaly.

2, online many feature abnormalities detect

Fig. 5 is abnormality detection general frame, and according to, local anomaly abnormal to starting point and global abnormal definition, first the present invention learns scene start position by GMM model and distribute.Then set Moving Window that a length is k as basic comparing unit, each motor pattern after on-line study cluster

locus and the distribution of direction of motion, set up position-based distance and direction distance classification device, learning model parameter; Online many feature abnormalities detection-phase, weighs difference between track to be measured and normal trace motor pattern from starting point, position and three levels of direction, judges whether it is abnormal track, and which kind of type judgement is.

1) set up track starting point distributed model

Study moving target enters the starting point distributed areas of monitoring scene, significant to abnormal behaviour identification.The present invention distributes by normal trace cluster start position in two-dimentional GMM model learning scene, sets up starting point position distribution model.First to training the point set that rises of track

with K mean cluster, obtain the initial parameter of GMM model, each gauss component parameter (p of recycling EM Algorithm Learning GMM _l, u _l, ∑ _l).Enter scene track starting point z=(x ₁, y ₁) ^tthe probability that meets GMM model profile is:

$P\left(z\right)=\underset{l=i}{\overset{k}{\mathrm{\&Sigma;}}}{p}_{l}G(z,u_l,{\mathrm{\&Sigma;}}_l)---\left(5\right)$

Wherein k represents single gauss component number, p _lfor single gauss component prior probability, meet

the probability density function of single gauss component is:

$G(z,u_l,{\mathrm{\&Sigma;}}_l)=\frac{1}{\sqrt{{\left(2\mathrm{\&pi;}\right)}^d\left|{\mathrm{\&Sigma;}}_l\right|}}\exp [-\frac{1}{2}{(z-u_l)}^T{{\mathrm{\&Sigma;}}_l}^{-1}(z-u_l)]---\left(6\right)$

Each track sample starting point (x ' ₁, y ' ₁) as input z _i=(x ' ₁, y ' ₁) ^t, try to achieve the probability density P (z under GMM model _i), get minimum probability value as GMM model threshold

2) set up the online classification device of position-based distance and direction distance

To the every class normal trace cluster C after cluster _iresample, make the track in each cluster

equal in length is l _i, obtain each cluster C _iposition represent pattern

wherein

if the Moving Window that length is k is as the basic comparing unit of online abnormality detection, definition position distance H _positionweigh track to be measured

and the locus match condition of normal trace motor pattern between basic comparing unit.

Define 4 positional distance (H _position).If track T to be measured and position represent pattern R _imoving Window comparing unit be respectively T ' _p={ t _p..., t _p+k-1| 1≤p≤N-k+1} and R ' _{i, q}={ t _q..., t _q+k-1| 1≤q≤l _i-k+1}.With IMHD distance definition positional distance, be equally:

$\left\{\begin{array}{c}{H}_{\mathrm{position}}(T_p^{\&prime;},R_{i,q}^{\&prime;})=\max (h(T_p^{\&prime;},R_{i,q}^{\&prime;}),h(R_{i,q}^{\&prime;},T_p^{\&prime;}))\\ h\left(T_{p,}^{\&prime;}{R}_{i,q}^{\&prime;}\right)=\frac{1}{k}\underset{a=p}{\overset{p+k-1}{\mathrm{\&Sigma;}}}\mathrm{dist}(t_a,R_{i,q}^{\&prime;})\end{array}\right.---\left(7\right)$

Set up position-based distance H below _positiononline classification device, because of stochastic variable H _position(T ' _p, R ' _{i, q}) be similar to and obey the exponential distribution that parameter is λ, so orbit segment T ' _pbelong to mode position R ' _{i, q}conditional probability be:

$P\left(T_p^{\&prime;}\right|R_{i,q}^{\&prime;})=e^{{-\mathrm{\&lambda;}}_q{H}_{\mathrm{position}}(T_p^{\&prime;},R_{i,q}^{\&prime;})---\left(8\right)}$

According to maximum likelihood assessment level, calculate cluster C _ieach track

represent pattern R with position _ipositional distance between corresponding Moving Window

parameter lambda _qbe estimated as:

${\mathrm{\&lambda;}}_q=\frac{M_i}{\underset{n=1}{\overset{M_i}{\mathrm{\&Sigma;}}}{H}_{\mathrm{position}}(T_{i,q}^{n\&prime;},R_{i,q}^{\&prime;})}---\left(9\right)$

According to bayesian theory, p the Moving Window T ' of track T to be measured _pposition sign syntype R _iq corresponding Moving Window R ' _{i, q}probability be:

$P\left(R_{i,q}^{\&prime;}\right|T_p^{\&prime;})=\frac{P\left(T_p^{\&prime;}\right|R_{i,q}^{\&prime;})P\left(R_i\right)}{\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}P\left(T_p^{\&prime;}\right|R_{i,q}^{\&prime;})P\left(R_i\right)},i=\mathrm{1,2},...,K---\left(10\right)$

Wherein

cluster C _itrack number account for the ratio of total sample number.Cluster C _ieach sample track as input, the parameter threshold of obtaining each Moving Window is

Then set up based on direction distance H _directiononline classification device, define equally direction distance H _directionweigh track T and the normal trace pattern direction of motion match condition between Moving Window.First according to formula (1), obtain orbit segment T ' to be measured _pcorresponding deflection θ _p, direction vector m wherein _p=(x _p+k-1-x _p, y _p+k-1-y _p).Direction distance definition is as follows:

Define 5 direction distance (H _direction).If pattern C _imean motion direction at q Moving Window is

corresponding orientation average departure degree is

utilize Mahalanobis distance, by average θ _{i, q}and variance calculate orbit segment T ' _pto pattern C _ithe direction distance of q Moving Window is:

$H_{\mathrm{direction}}({\mathrm{\&theta;}}_p,{\mathrm{\&theta;}}_{i,q})=\sqrt{\frac{{({\mathrm{\&theta;}}_p-{\mathrm{\&theta;}}_{i,q})}^2}{{\mathrm{\&sigma;}}_{i,q}^2}}---\left(11\right)$

Wherein

for pattern C _imiddle track

at the deflection of q Moving Window, correspondence direction vector is ask pattern C _ieach track

direction distance at corresponding Moving Window get maximal value as the parameter threshold of q Moving Window

3) online many feature abnormalities detection algorithm

Character according to the present invention to abnormal definition is known, when determining that the locus of track occurs extremely or adjacent track section does not belong to same position pattern, can be judged as global abnormal, need not judge that whether direction is abnormal again, can improve online detection efficiency like this; And online detection of track local anomaly be take malposition detection as basis, if orbit segment T ' _pposition distribution belongs to

judge again orbit segment T ' _pdirection of motion whether still belong to

determine whether as local anomaly, as shown in Outlier Detection Algorithm flow process frame as empty in Fig. 5.

Setting Moving Window length is k, and algorithm 2 is described below:

Step1 extracts track starting point z=(x to be measured ₁, y ₁) ^t, calculate P (z), if P (z)>=Ω _gMM, judge that track to be measured enters scene from normal start position region; Otherwise be that starting point is abnormal.

Step2 calculates course length Length to be measured (T), the Moving Window T ' of online extract real-time track T to be measured _p.

Step3 is at each mode position R _iunder, find and orbit segment T ' to be measured _pthe Moving Window of positional distance minimum

bring into corresponding to H _positionsorter, searching makes

the track position pattern of maximum probability $\tilde{i}=\underset{i}{\arg \max }\left(P\left(R_{i,\tilde{q}}^{\&prime;}\right|T_p^{\&prime;})\right)\&CenterDot;$

If Step4 judgement

bing and previous Moving Window T ' _p-1meet same pattern

the orbit segment T ' of track T so to be measured _pposition distribution is normal, belongs to

, enter Step5, judge whether local anomaly; Otherwise output trajectory section T ' _pfor global abnormal, enter Step6.

Step5 calculates Moving Window T ' _pdeflection θ _p, bring associative mode into

based on H _directionsorter, calculate itself and the direction distance of corresponding window

if

direction of motion meets

judge orbit segment T ' to be measured _pfor normal trace section, belong to pattern C _i; Otherwise output trajectory section T ' _pfor local anomaly.

Step6 judges whether Length (T) changes, if changed, extracts next Moving Window, repeats Step2 to Step5; If length is constant, each Moving Window T ' of Bing _pall belong to motor pattern C _i, T is normal trace.

Accompanying drawing explanation

1) the unsupervised trajectory model learning framework of Fig. 1

2) Fig. 2 trajectory direction feature extraction

3) Fig. 3 track line segment interpolation model

4) Fig. 4 global abnormal and local anomaly schematic diagram

5) the online many feature abnormalities of Fig. 5 detect framework

6) Fig. 6 two-way lane traffic scene: a) original background image b) pretreated training track

7) many Dividing Characteristics of Fig. 7 trajectory clustering result: a) based on the thick cluster b of direction of motion) based on the thin cluster in locus

8) Fig. 8 Laplacian proper value of matrix schematic diagram

9) Fig. 9 GMM model learning starting point distributed areas

10) Figure 10 overall situation and local anomaly detect online: a) normal trace b) the abnormal c of lane change) the abnormal d of u turn) abnormal e hovers) oppositely abnormal

embodiment

1, the hierarchical clustering algorithm experimental verification of the many features based on Laplacian matrix

Validity for proof clustering algorithm, a two-way lane traffic video image (as adopting two dimension target track algorithm, is followed the tracks of the moving target of Fig. 6 in a), frame speed is 25 frames/s, image resolution ratio is 320 * 240, gather altogether 296 vehicle movement tracks, by retaining 216 effective tracks after pre-service such as level and smooth grade, as shown in Figure 6 b.

The clustering algorithm that application proposes, first carries out based on the thick cluster of direction of motion training track, and cluster result as shown in Figure 7a, is roughly divided into track the two class O that direction of motion is contrary ₁and O ₂; Then cluster in the middle of each is carried out respectively based on the thin cluster in locus, final cluster result as shown in Figure 7b, represents respectively middle cluster O ₁and O ₂thin cluster result, cluster obtains 5 different classification in the middle of each, has 10 track clusters, just corresponding one by one with the region, 10 tracks of reality.

In every strata class process, clusters number, by by Laplacian proper value of matrix sequencing numbers from big to small, is found and is had the adjacent feature value of maximum disparity automatically to determine.As shown in Figure 8, transverse axis is eigenwert numbering, and the longitudinal axis is characteristic of correspondence value.The thick cluster of ground floor disparity between the 2nd and the 3rd eigenwert is △ λ ₁=0.373, so thick cluster number is 2; The thin cluster of the second layer is all with the biggest gap between the 5th and the 6th eigenwert, is respectively △ λ ₂=0.523 and △ λ ₃=0.463, so thin cluster number is all 5.

In addition, the clustering algorithm proposing because of this patent is the improvement based on Agglomerative Hierarchical Clustering algorithm, so two kinds of methods are compared on the time in cluster accuracy rate and cluster.In experimental data, total number of tracks is 216, and cluster numbers is all 10, and the average result that algorithm operation is 10 times is as shown in table 1.

The comparison of table 1 cluster result

As seen from the results in Table 1, the clustering algorithm that this patent proposes all increases in cluster accuracy rate and time efficiency, be due to the clustering algorithm proposing adopt simultaneously the direction of motion of track and locus feature weigh orbit interval from, carried out hierarchical cluster, therefore improved cluster accuracy rate; In hierarchical cluster from coarse to fine experiment, owing to having introduced Laplacian mapping, the low-dimensional data representation higher-dimension track data of tieing up with 2 peacekeepings 5 respectively, greatly reduces computation complexity, thus cluster time decreased 61.22%, cluster efficiency significantly improves.

2, online many feature abnormalities detection algorithm experimental verification

First judge that starting point is abnormal, by GMM model learning, enter the position distribution of monitoring scene track starting point, judge that whether track starting point to be measured is abnormal.As shown in Figure 9, these scene track starting point distributed areas consist of three single Gaussian distribution.Then establish Moving Window length k=5, by slip Moving Window, judge online global abnormal and local anomaly.As shown in figure 10, the 1st classifies the different vehicle track of tracking as to online testing process, and arrow represents the direction of Vehicle Driving Cycle; The transverse axis of the 2nd each figure of row is Moving Window numbering, and the longitudinal axis is probability, is expressed as the probability that the position distribution of corresponding track to be measured under each Moving Window belongs to each motor pattern, with solid line and dotted line, shows front two maximum probability pattern; The transverse axis of the 3rd each figure of row is Moving Window numbering, the longitudinal axis is deflection, in figure solid line be track to be measured at the direction of motion angle of each Moving Window, dotted line represents that motor pattern that track to be measured meets in locus corresponding to Figure 10 b is most at the mean motion deflection of corresponding window.Figure 10 (a) is normal trace, and the probability that each Moving Window of track to be measured meets pattern 6 is always between 0.9 to 1, and direction of motion meets pattern 6 too; The track of Figure 10 (b) and 10 (c) is global abnormal, as Figure 10 (b2) with (c2), respectively the 7th and 8 Moving Windows start the probability that locus belongs to original pattern 6 and decline, and the probability that belongs to mode 7 and 5 increases progressively, when being less than pattern 6 threshold value, can report to the police abnormal, and corresponding deflection changes as Figure 10 (b3) with (c3); Figure 10 (d) and (e) in trajectory range position belong to pattern 6 and 7 probability always between 0.9 to 1, but the direction distance that can judge track to be measured and associative mode 6 and 7 by direction distance classification device surpasses threshold value, judges that two tracks are local anomaly.

The validity of the online many feature abnormalities detection algorithm proposing for checking, adopts respectively path matching algorithm based on single Gauss model and the locus model matching process based on C-HMM to compare.Secondary tracking obtains the track of vehicle of Fig. 6 scene again, and Bing people is adds various abnormal next abundant detection algorithm efficiency.Finally in experimental traces data, hand labeled normal trace be 265, abnormal track is 89.In abnormal behaviour detects, by precision ratio and recall ratio, weigh abnormality detection efficiency more effective, comparative result is as shown in table 2.

The comparison of table 2 method for detecting abnormality result

The experimental result of table 2 shows, the abnormal track identification efficiency of method that this patent proposes is higher, and especially recall ratio, has improved respectively 20.22% and 25.84% than two kinds of traditional methods.Because classic method is only considered the differences in spatial location of track and large abnormal of rough detection shape difference, the abnormal and local subsegment of direction is extremely difficult to identify.And the Outlier Detection Algorithm that this patent proposes can be weighed the direction of motion of track to be measured and the difference of locus, by slip Moving Window, detect local anomaly and global abnormal simultaneously, therefore abnormal discrimination is higher.

Claims (2)

1. unsupervised many characteristic locuses pattern learning method, said method comprising the steps of:

A. similarity measurement between many feature extractions and track

Adopt respectively IMHD distance and Bhattacharyya apart from the similarity of calculating between track, pretreated effective track can be expressed as:

$T_i=\{t_1,t_2,...,t_j,...,t_{N_i}\}=\{(x_1,y_1),(x_2,y_2),...,(x_j,y_j),...,(x_{N_i},y_{N_i})\}$

T wherein _jrepresent track T _ij sampled point, N _irepresent course length, (x _j, y _j) represent that j sampled point is at the two-dimensional position coordinate of the plane of delineation; Track definition is m _j=(x _j+1-x _j, y _j+1-y _j), represent direction vector between neighbouring sample point; m ₀=(1,0), represents direction level vector of unit length to the right; Track T _ij sampled point deflection can be expressed as:

${\mathrm{\&theta;}}_j=\left\{\begin{array}{c}\frac{{\cos }^{-1}(m_j\&CenterDot;m_0)}{\left|m_j\right|\&CenterDot;\left|m_0\right|}\&CenterDot;\frac{180}{\mathrm{\&pi;}},\mathrm{if}{y}_{j+1}-y_j\&GreaterEqual;0\\ (2\mathrm{\&pi;}-\frac{{\cos }^{-1}(m_j\&CenterDot;m_0)}{\left|m_j\right|\&CenterDot;\left|m_0\right|})\&CenterDot;\frac{180}{\mathrm{\&pi;}},\mathrm{if}{y}_{j+1}-y_j<0\end{array}\right.(1\&le;j<N_i-1)---\left(1\right)$

θ wherein _j∈ (0,360), trajectory direction angle θ _jbe divided into N=18 uniformly-spaced sub-range (I ₁, I ₂... I _n), each sub-range length is △ θ=360/N=20 degree, track T _iall deflection θ _jbe mapped to corresponding sub-range, track T _ideflection is distributed in interval I _qprobability be p _q=M _q/ M, wherein M _q: θ _j∈ I _qnumber, M: track T _ideflection number;

Track T _idirection character can be expressed as

track T has been described _istatistics directional information, adopts Bhattacharyya apart from direction of motion similarity between measurement track to be:

$\mathrm{Dire}(T_i,T_j)={[1-\underset{q=1}{\overset{N}{\mathrm{\&Sigma;}}}\sqrt{\mathrm{Dire}{\left(T_i\right)}_q\mathrm{Dire}{\left(T_j\right)}_q}]}^{1/2}\&Element;\left[\mathrm{0,1}\right]---\left(2\right)$

Dire (T wherein _i) _qand Dire (T _j) _qrepresent respectively track T _iand T _jdeflection be distributed in the probability in q deflection interval, Dire (T _i, T _j) more close to 1, represent that two orbiting motion directions are more similar; Dire (T _i, T _j) be more bordering on 0, contrary;

Trajectory range location similarity is measured: by target following, can directly obtain track position feature,

the improvement Hausdorff distance (IMHD) of employing based on line segment interpolation weighed the similarity of trajectory range position, and track line segment represents track T with broken line _iand T _j: $\stackrel{\&OverBar;}{T_i}=\{\stackrel{\&OverBar;}{t_1{t}_2},\stackrel{\&OverBar;}{t_2{t}_3},...,\stackrel{\&OverBar;}{t_{a-1}{t}_a},...,\stackrel{\&OverBar;}{t_{N_{i-1}}{t}_{N_i}}\},\stackrel{\&OverBar;}{T_j}=\{\stackrel{\&OverBar;}{t_1{t}_2},\stackrel{\&OverBar;}{t_2{t}_3},...,\stackrel{\&OverBar;}{t_{b-1}{t}_b},...,\stackrel{\&OverBar;}{t_{N_{j-1}}{t}_{N_j}}\},$ In IMHD algorithm, track T _isampled point t _ato T _jbee-line is expressed as:

Wherein || || represent Euclidean distance between sampled point,

t _abe mapped to

the vertical interpolation point of corresponding line segment, if existed

for t _ato T _jbee-line, otherwise traversal track T _jall sampled points, find minor increment

last track T _ito T _jdistance be:

track T _iand T _jlocus distance be:

D _IMHD(T _i，T _j)=max(h(T _i，T _j)，h(T _j，T _i)) (4)

B. many characteristic locus classification

If the set that contains m effective track is Ω _traj={ T ₁, T ₂..., T _m, Step1 uses Bhattacharyya apart from calculating track T _i, T _jbetween direction of motion similarity Dire (T _i, T _j), the similar matrix W with gaussian kernel function structure based on direction of motion _dire∈ R ^{m * m}, w wherein _ij=exp (Dire (T _i, T _j) ²/ 2 σ ²), σ is scale parameter; Step2 structure standard Laplacian matrix

d wherein ₁for diagonal matrix, matrix element is

step3 is to L _direcarry out Eigenvalues Decomposition, eigenwert is numbered to λ by descending sort ₁>=λ ₂>=...>=λ _m, calculate the poor of adjacent feature value, if i eigenwert and i+1 eigenwert difference are maximum, determine thick cluster number step4 structure m * k matrix L=[l ₁, l ₂..., l _k], l wherein ₁, l ₂..., l _kfor front k eigenwert characteristic of correspondence vector, each row of L is carried out to unit processing, obtain matrix X, wherein

track set after dimensionality reduction is Ω _traj'={ x ₁, x ₂..., x _m, x ₁, x ₂..., x _meach row vector of homography X, represents R respectively ^ka bit of space; Step5 is to x ₁, x ₂..., x _mcarry out Agglomerative Hierarchical Clustering, use similarity between two bunches of minimum distance calculation, when finally merging into k bunch, iteration stops, and obtains k centre cluster { O ₁, O ₂, O _i..., O _k; Step6 is to cluster O in the middle of each _ithe former track of middle correspondence uses IMHD apart from the similar matrix W constructing based on locus _iMHD, w wherein _ij=exp (D _iMHD(T _i, T _j) ²/ 2 σ ²), repeat Step2 to Step5, determine each cluster O _icluster number q _i, construct low dimensional feature space, last cluster obtains cluster { C ₁, C ₂, C _k, total cluster number

2. online many feature abnormalities detection method, said method comprising the steps of:

A. online many feature abnormalities detect

First by GMM model, learn scene start position and distribute, then set Moving Window that a length is k as basic comparing unit, each motor pattern after on-line study cluster

locus and the distribution of direction of motion, set up position-based distance and direction distance classification device, learning model parameter; Online many feature abnormalities detection-phase, weighs difference between track to be measured and normal trace motor pattern from starting point, position and three levels of direction, judges whether it is abnormal track, and judgement is that starting point is abnormal, which kind of type in local anomaly and global abnormal three;

B. set up track starting point distributed model:

By normal trace cluster start position in two-dimentional GMM model learning scene, distribute, set up starting point position distribution model; First to training the point set that rises of track with K mean cluster, obtain the initial parameter of GMM model, each gauss component parameter (p of recycling EM Algorithm Learning GMM _l, u _l, ∑ _l), enter scene track starting point z=(x ₁, y ₁) ^tthe probability that meets GMM model profile is:

$P\left(z\right)=\underset{l=l}{\overset{k}{\mathrm{\&Sigma;}}}{p}_{l}G(z,u_l,{\mathrm{\&Sigma;}}_l)---\left(5\right)$

Wherein k represents single gauss component number, p _lfor single gauss component prior probability, meet

the probability density function of single gauss component is:

$G(z,u_l,{\mathrm{\&Sigma;}}_l)=\frac{1}{\sqrt{{\left(2\mathrm{\&pi;}\right)}^d\left|{\mathrm{\&Sigma;}}_l\right|}}\exp [-\frac{1}{2}{(z-u_l)}^T{{\mathrm{\&Sigma;}}_l}^{-1}(z-u_l)]---\left(6\right)$

Each track sample starting point

as input

try to achieve the probability density P (z under GMM model _i), get minimum probability value as GMM model threshold ${\mathrm{\&Omega;}}_{\mathrm{GMM}}=\underset{i}{\min }\left(P\left(z_i\right)\right);$

C. set up the online classification device of position-based distance and direction distance:

To the every class normal trace cluster C after cluster _iresample, make the track in each cluster

equal in length is l _i, obtain each cluster C _iposition represent pattern

wherein

if the Moving Window that length is k is as the basic comparing unit of online abnormality detection, positional distance H _positionbe used for weighing track to be measured

and the locus match condition of normal trace motor pattern between basic comparing unit;

If track T to be measured and position represent pattern R _imoving Window comparing unit be respectively T ' _p={ t _p..., t _p+k-1| 1≤p≤N-k+1) and R ' _{i, q}={ t _q..., t _q+k-1| 1≤q≤l _i-k+1}, with IMHD distance definition positional distance is:

$\left\{\begin{array}{c}{H}_{\mathrm{position}}(T_p^{\&prime;},R_{i,q}^{\&prime;})=\max (h(T_p^{\&prime;},R_{i,q}^{\&prime;}),h(R_{i,q}^{\&prime;},T_p^{\&prime;}))\\ h\left(T_{p,}^{\&prime;}{R}_{i,q}^{\&prime;}\right)=\frac{1}{k}\underset{a=p}{\overset{p+k-1}{\mathrm{\&Sigma;}}}\mathrm{dist}(t_a,R_{i,q}^{\&prime;})\end{array}\right.---\left(7\right)$

Then set up position-based distance H _positiononline classification device, because of stochastic variable H _position(T ' _p, R ' _{i, q}) be similar to and obey the exponential distribution that parameter is λ, so orbit segment T ' _pbelong to mode position R ' _{i, q}conditional probability be:

$P\left(T_p^{\&prime;}\right|R_{i,q}^{\&prime;})=e^{{-\mathrm{\&lambda;}}_q{H}_{\mathrm{position}}\left(T_p^{\&prime;}{,R}_{i,q}^{\&prime;}\right)}---\left(8\right)$

According to maximum likelihood assessment level, calculate cluster C _ieach track

represent pattern R with position _ipositional distance between corresponding Moving Window $H_{\mathrm{posiuon}}(T_{i,q}^{n\&prime;},R_{i,q}^{\&prime;}),$ Parameter lambda _qbe estimated as:

${\mathrm{\&lambda;}}_q=\frac{M_i}{\underset{n=1}{\overset{M_i}{\mathrm{\&Sigma;}}}{H}_{\mathrm{position}}(T_{i,q}^{n\&prime;},R_{i,q}^{\&prime;})}---\left(9\right)$

According to bayesian theory, p the Moving Window T ' of track T to be measured _pposition sign syntype R _iq corresponding Moving Window R ' _{i, q}probability be:

$P\left(P_{i,q}^{\&prime;}\right|T_p^{\&prime;})=\frac{P\left(T_p^{\&prime;}\right|R_{i,q}^{\&prime;})P\left(R_i\right)}{\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}P\left(T_p^{\&prime;}\right|R_{i,q}^{\&prime;})P\left(R_i\right)},i=\mathrm{1,2},...,K---\left(10\right)$

Wherein cluster C _itrack number account for the ratio of total sample number, cluster C _ieach sample track as input, the parameter threshold of obtaining each Moving Window is

then set up based on direction distance H _directiononline classification device, use equally direction distance H _directionweigh track T and the normal trace pattern direction of motion match condition between Moving Window: first according to formula (1), obtain orbit segment T ' to be measured _pcorresponding deflection θ _p, direction vector m wherein _p=(x _p+k-1-x _p, y _p+k-1-y _p), direction distance is for establishing pattern C _imean motion direction at q Moving Window is corresponding orientation average departure degree is

recycling Mahalanobis distance, by average θ _{i, q}and variance calculate orbit segment T ' _pto pattern C _ithe direction distance of q Moving Window is:

$H_{\mathrm{direction}}=({\mathrm{\&theta;}}_p,{\mathrm{\&theta;}}_{i,q})=\sqrt{\frac{{({\mathrm{\&theta;}}_p-{\mathrm{\&theta;}}_{i,q})}^2}{{\mathrm{\&sigma;}}_{i,q}^2}}---\left(11\right)$

Wherein

for pattern C _imiddle track

at the deflection of q Moving Window, correspondence direction vector is

ask pattern C _ieach track

direction distance at corresponding Moving Window

get maximal value as the parameter threshold of q Moving Window ${\mathrm{\&Omega;}}_{\mathrm{direction}}^{i,q}=\underset{n}{\max }\left(H_{\mathrm{direction}}({\mathrm{\&theta;}}_{i,q}^n,{\mathrm{\&theta;}}_{i,q})\right);$

D. online many feature abnormalities detect:

When determining that the locus of track occurs extremely or adjacent track section does not belong to same position pattern, can be judged as global abnormal, need not judge again that whether direction is abnormal; The online detection of local anomaly be take malposition detection as basis, if orbit segment T ' _pposition distribution belongs to judge again orbit segment T ' _pdirection of motion whether still belong to

determine whether as local anomaly, setting Moving Window length is k, and concrete steps are as follows:

Step1 extracts track starting point z=(x to be measured ₁, y ₁) ^t, calculate P (z), if P (z)>=Ω _gMM, judge that track to be measured enters scene from normal start position region; Otherwise be that starting point is abnormal; Step2 calculates course length Length to be measured (T), the Moving Window T ' of online extract real-time track T to be measured _p;

Step3 is at each mode position R _iunder, find and orbit segment T ' to be measured _pthe Moving Window of positional distance minimum

bring into corresponding to H _positionsorter, searching makes

the track position pattern of maximum probability

if Step4 judgement

bing and previous Moving Window T ' _p-1meet same pattern

the orbit segment T ' of track T so to be measured _pposition distribution is normal, belongs to

enter Step5, judge whether local anomaly; Otherwise output trajectory section T ' _pfor global abnormal, enter Step6; Step5 calculates Moving Window T ' _pdeflection θ _p, bring associative mode into

based on H _directionsorter, calculate itself and the direction distance of corresponding window

if direction of motion meets judge orbit segment T ' to be measured _pfor normal trace section, belong to pattern C _i; Otherwise output trajectory section T ' _pfor local anomaly; Step6 judges whether Length (T) changes, if changed, extracts next Moving Window, repeats Step2 to Step5; If length is constant, and each Moving Window T ' _pall belong to motor pattern C _i, T is normal trace.

Images