CN104732236A - Intelligent crowd abnormal behavior detection method based on hierarchical processing - Google Patents

Abstract

The invention provides an intelligent crowd abnormal behavior detection method based on hierarchical processing. The method comprises the crowd state pre-judging process and the crowd abnormal behavior detection process. According to the crowd state anticipation process, crowd states are prejudged by utilizing a crowd state change index R; according to the crowd abnormal behavior detection process, crowd abnormal behaviors are judged by utilizing crowd kinetic energy En and crowd motion direction entropy Hn. The intelligent crowd abnormal behavior detection method based on hierarchical processing has a good detecting effect on the crowd abnormal behaviors, learning and training are not needed, flexibility is high, the calculation amount is low, the real-time property can be well ensured, and the method is suitable for crowded places such as squares, metros, railway stations, airports and the like with the high public security requirement.

Description

A kind of crowd's abnormal behaviour intelligent detecting method based on layered shaping

Technical field

The invention belongs to Intelligent Video Surveillance Technology field, particularly relate to a kind of crowd's abnormal behaviour intelligent detecting method based on layered shaping.

Technical background

In recent years, along with the Mass disturbance endangered public security frequently occurs, the demand of people to public safety is more and more higher, so more and more receive the concern of people to the crowd's abnormal behaviour analysis under crowd's scene and visual analysis.Wherein crowd behaviour analysis is the challenging problem of computer vision field most.

No. 201110090467.6th, Chinese patent discloses a kind of detection method of the crowd's abnormal behaviour based on improvement social force model, but this method is only applicable to highdensity crowd's scene.Chinese patent the 201210403819.3rd, 201310437367.5, the crowd's anomaly detection method proposed in 201310494769.9 all relates to the process of learning training, and result is comparatively coarse, algorithm complex is high, and operand is large, is therefore difficult to ensure real-time.

Summary of the invention

In order to solve the problem, the object of the present invention is to provide a kind of crowd's abnormal behaviour intelligent detecting method based on layered shaping.

In order to achieve the above object, the crowd's abnormal behaviour intelligent detecting method based on layered shaping provided by the invention comprises crowd state anticipation and crowd's unusual checking two processes, wherein: crowd state anticipation process comprises the following step carried out in order:

1) mixed Gauss model is utilized to extract crowd's prospect to video input;

2) to said extracted to crowd's prospect utilize median filter to carry out medium filtering process, then calculate crowd's foreground area area S;

4) difference got to adjacent two frame videos and obtain frame difference region area S ₂;

4) using step 2) in crowd's foreground area area S of obtaining as foreground mask operator to step 3) in the frame difference region area S that obtains ₂do mask computing, obtain crowd's moving area area S thus ₁;

5) according to step 2) in crowd's foreground area area S of obtaining and step 4) in crowd's moving area area S of obtaining ₁calculate crowd state variability index R, wherein

6) utilize crowd state variability index R to carry out anticipation to crowd state, work as R<R ₀time, explanation crowd is in normal condition, R ₀represent crowd state threshold value; Work as R>R ₀time, illustrate that crowd movement's state changes, unusual condition may be had to occur, detect further by following step;

Described crowd's unusual checking process comprises the following step carried out in order:

A) Harris Corner Detection is carried out to video input;

B) L-K optical flow tracking is carried out to the above-mentioned Harris angle point detected, obtain crowd movement vector field w _i={ u, v}, wherein w _irepresent the motion vector field of i-th unique point in the n-th frame;

C) according to step B) in crowd movement's vector field of obtaining, calculate the speed of each unique point,

$v_i=\sqrt{u^2+v^2}$

D) according to step C) speed of unique point that obtains, calculate crowd movement's ENERGY E of the n-th frame further _n, wherein m _irepresent the quality of i-th unique point of the n-th frame, suppose the quality m of all unique points here _i=1, n-th frame has j unique point;

F) according to step B) in crowd movement's vector field of obtaining, obtain crowd movement's directional spreding histogram further: h (i)={ k _i, 0<i<=8}, wherein h (i) represents the number of the motion vector that each direction comprises, and is altogether divided into 8 directions;

F) according to step e) in crowd movement's directional spreding histogram of obtaining obtain crowd's directional spreding probability P (i), $P\left(i\right)=\frac{h\left(i\right)}{m},0<i<=n,i\&Element;N;$ Wherein m is the sum of crowd's motion vector, $m={\mathrm{\&Sigma;}}_1^{8}h\left(i\right);$

G) according to step F) in crowd's directional spreding probability of obtaining, calculate crowd movement direction entropy H further _n, $H_n={\mathrm{\&Sigma;}}_{i=1}^{n}p\left(X_i\right){\log }_2\left(\frac{1}{p\left(x_i\right)}\right);$

H) according to step D) in crowd movement's ENERGY E of obtaining _nwith step G) in the crowd movement direction entropy H that obtains _ncarry out the judgement of crowd's abnormal behaviour, if E _n>E ₀and H _n>H ₀, judge that this crowd there occurs abnormal behaviour, wherein H ₀represent the threshold value of crowd movement direction entropy.

Crowd's abnormal behaviour intelligent detecting method based on layered shaping provided by the invention not only has good Detection results to crowd's abnormal behaviour, and do not need the process of learning training, dirigibility is simultaneously strong, operand is low, real-time can be ensured well, be applicable to square, subway, railway station, airport etc. the crowd is dense and the place that public safety demand is higher.

Accompanying drawing explanation

Fig. 1 is the crowd's abnormal behaviour intelligent detecting method overall flow figure based on layered shaping provided by the invention.

Fig. 2 is provided by the invention based on crowd state anticipation process flow diagram flow chart in crowd's abnormal behaviour intelligent detecting method of layered shaping.

Fig. 3 is provided by the invention based on crowd's unusual checking process flow diagram flow chart in crowd's abnormal behaviour intelligent detecting method of layered shaping.

Fig. 4 is crowd direction division schematic diagram.

Fig. 5 is crowd's direction of motion distribution histogram.

Embodiment

Below in conjunction with the drawings and specific embodiments, the crowd's abnormal behaviour intelligent detecting method based on layered shaping provided by the invention is described in detail.

As shown in Fig. 1-Fig. 3, the crowd's abnormal behaviour intelligent detecting method based on layered shaping provided by the invention comprises crowd state anticipation and crowd's unusual checking two processes, wherein: crowd state anticipation process comprises the following step carried out in order:

1) mixed Gauss model is utilized to extract crowd's prospect to video input;

By finding a large amount of demographic data collection test, when crowd is normal, crowd's moving area remains on a less and stable level, and when abnormal behaviour occurs crowd, crowd's moving area increases rapidly.Crowd's moving area is different from crowd's foreground area, and what the latter described is the region not belonging to background, and what the former described is the region of crowd's movement in foreground area.

2) to said extracted to crowd's prospect utilize median filter to carry out medium filtering process, then calculate crowd's foreground area area S;

5) difference got to adjacent two frame videos and obtain frame difference region area S ₂;

4) using step 2) in crowd's foreground area area S of obtaining as foreground mask operator to step 3) in the frame difference region area S that obtains ₂do mask computing, obtain crowd's moving area area S thus ₁;

5) according to step 2) in crowd's foreground area area S of obtaining and step 4) in crowd's moving area area S of obtaining ₁calculate crowd state variability index R, wherein

6) utilize crowd state variability index R to carry out anticipation to crowd state, work as R<R ₀time, explanation crowd is in normal condition, R ₀represent crowd state threshold value, its value size is by obtaining a large amount of test result statistical study; Work as R>R ₀time, illustrate that crowd movement's state changes, unusual condition may be had to occur, detect further by following step;

Described crowd's unusual checking process comprises the following step carried out in order:

A) Harris Corner Detection is carried out to video input;

B) L-K optical flow tracking is carried out to the above-mentioned Harris angle point detected, obtain crowd movement vector field w _i={ u, v}, wherein w _irepresent the motion vector field of i-th unique point in the n-th frame;

C) according to step B) in crowd movement's vector field of obtaining, calculate the speed of each unique point,

$v_i=\sqrt{u^2+v^2}$

D) according to step C) speed of unique point that obtains, calculate crowd movement's ENERGY E of the n-th frame further _n, wherein m _irepresent the quality of i-th unique point of the n-th frame, suppose the quality m of all unique points here _i=1, n-th frame has j unique point;

G) according to step B) in crowd movement's vector field of obtaining, obtain crowd movement's directional spreding histogram further: h (i)={ k _i, 0<i<=8}, wherein h (i) represents the number of the motion vector that each direction comprises, and is altogether divided into 8 directions;

Crowd movement's directional spreding rule is one of key character of crowd state.Crowd movement direction, in units of 45 degree, is divided into 8 directions, as shown in Figure 4.Crowd movement's directional spreding histogram as shown in Figure 5.

F) according to step e) in crowd movement's directional spreding histogram of obtaining obtain crowd's directional spreding probability P (i), $P\left(i\right)=\frac{h\left(i\right)}{m},0<i<=n,i\&Element;N;$ Wherein m is the sum of crowd's motion vector, $m={\mathrm{\&Sigma;}}_1^{8}h\left(i\right);$

G) according to step F) in crowd's directional spreding probability of obtaining, calculate crowd movement direction entropy H further _n, $H_n={\mathrm{\&Sigma;}}_{i=1}^{n}p\left(X_i\right){\log }_2\left(\frac{1}{p\left(x_i\right)}\right);$

H) according to step D) in crowd movement's ENERGY E of obtaining _nwith step G) in the crowd movement direction entropy H that obtains _ncarry out the judgement of crowd's abnormal behaviour, if E _n>E ₀and H _n>H ₀, judge that this crowd there occurs abnormal behaviour, wherein H ₀represent the threshold value of crowd movement direction entropy, its value size obtains by carrying out statistical study to a large amount of test results.

Claims (1)

1. based on crowd's abnormal behaviour intelligent detecting method of layered shaping, it is characterized in that: it comprises crowd state anticipation and crowd's unusual checking two processes, wherein: crowd state anticipation process comprises the following step carried out in order:

1) mixed Gauss model is utilized to extract crowd's prospect to video input;

2) to said extracted to crowd's prospect utilize median filter to carry out medium filtering process, then calculate crowd's foreground area area S;

3) difference got to adjacent two frame videos and obtain frame difference region area S ₂;

4) using step 2) in crowd's foreground area area S of obtaining as foreground mask operator to step 3) in the frame difference region area S that obtains ₂do mask computing, obtain crowd's moving area area S thus ₁;

5) according to step 2) in crowd's foreground area area S of obtaining and step 4) in crowd's moving area area S of obtaining ₁calculate crowd state variability index R, wherein

6) utilize crowd state variability index R to carry out anticipation to crowd state, work as R<R ₀time, explanation crowd is in normal condition, R ₀represent crowd state threshold value; Work as R>R ₀time, illustrate that crowd movement's state changes, unusual condition may be had to occur, detect further by following step;

Described crowd's unusual checking process comprises the following step carried out in order:

A) Harris Corner Detection is carried out to video input;

B) L-K optical flow tracking is carried out to the above-mentioned Harris angle point detected, obtain crowd movement vector field w _i={ u, v}, wherein w _irepresent the motion vector field of i-th unique point in the n-th frame;

C) according to step B) in crowd movement's vector field of obtaining, calculate the speed of each unique point,

$v_i=\sqrt{u^2+v^2}$

D) according to step C) speed of unique point that obtains, calculate crowd movement's ENERGY E of the n-th frame further _n, wherein m _irepresent the quality of i-th unique point of the n-th frame, suppose the quality m of all unique points here _i=1, n-th frame has j unique point;

E) according to step B) in crowd movement's vector field of obtaining, obtain crowd movement's directional spreding histogram further: h (i)={ k _i, 0<i<=8}, wherein h (i) represents the number of the motion vector that each direction comprises, and is altogether divided into 8 directions;

F) according to step e) in crowd movement's directional spreding histogram of obtaining obtain crowd's directional spreding probability P (i), 0<i<=n, i ∈ N; Wherein m is the sum of crowd's motion vector,

G) according to step F) in crowd's directional spreding probability of obtaining, calculate crowd movement direction entropy H further _n, $H_n={\mathrm{\&Sigma;}}_{i=1}^{n}p\left(X_i\right){\log }_2\left(\frac{1}{p\left(x_i\right)}\right);$

H) according to step D) in crowd movement's ENERGY E of obtaining _nwith step G) in the crowd movement direction entropy H that obtains _ncarry out the judgement of crowd's abnormal behaviour, if E _n>E ₀and H _n>H ₀, judge that this crowd there occurs abnormal behaviour, wherein H ₀represent the threshold value of crowd movement direction entropy.