CN104616497A - Public transportation emergency detection method - Google Patents

Abstract

The invention discloses a public transportation emergency detection method and belongs to the field of intelligent transportation video monitoring. The public transportation emergency detection method comprises the following steps of 1 denoising an original image, dividing an activity range of people in a vehicle, extracting a video image in the activity range and applying an outline deepening frame difference method and a pixel matching method to extract a video foreground region, 2 extracting feature points in the video foreground region and establishing a foreground motion region vector field by means of an optical flow method, and 3 judging whether an emergency occurs in the vehicle according to change of average region kinetic energy and average feature point distance potential energy in the foreground motion region vector field. Compared with the prior art, the public transportation emergency detection method is high in operation efficiency and efficiency and capable of detecting the disorder situation in the vehicle when the emergency occurs.

Description

Public transport emergency detection method

Technical field

The present invention relates to a kind of public transport emergency detection method, belong to wisdom traffic video monitoring field.

Background technology

On June 7th, 2013, Xiamen bus fire accident causes that 48 people are dead, people more than 30 is injured; During 27 days 12 February in 2014 37 points, one, Guiyang No. 237 buses burn at South Road, Jinyang, Yunyan District, and accident causes 6 dead 35 wounds; On July 6th, 2014, Zhejiang Hangzhou there occurs the tragedy that public transport is together caught fire, and causes 32 passengers injured; On November 21st, 2014, Zhongshan City, Guangdong Province bus is held as a hostage ... visible, in recent years, bus accident was of common occurrence, except the problem that bus spontaneous combustion, blow out etc. is produced by bus self in these accidents, the panic event of the terror more having terrorist to cause.

In today of all-round construction smart city, these events allow the present invention be absorbed in thinking, in the face of the emergency of bus burst, infotech and intelligent video monitoring is how utilized to give warning in advance, reduce casualties, for investigation provides advantageous information evidence, while realizing security purpose, reduce the consumption of social resources.

Application number be 200910110984.8 application for a patent for invention provide a kind of public transportation intelligent terminal, from disclosed technical scheme, that monitoring module has been installed in terminal system, and be arranged near driver's seat, the object of this monitoring module allows driver understand situation in car in real time, in the program, monitoring module does not do emergency process as can be seen here, just the input and output of a simple video.

Application number be 200920298772.2 utility model patent provide a kind of city bus mobile video monitoring and real-time tracking dispatching system, from technical scheme disclosed in it, that embedded video collecting device is installed on bus, and by mobile communications network in real time by the sound in car, image information is transferred to Surveillance center and realizes playing in real time, but this mode not only wastes a lot of energy of supervisor look at screen and to consume Mobile data flow very much, the most important thing is not accomplish that emergency detects, visual fatigue can be produced time screen is checked when people stares at for a long time and accurately real-time judging whether cannot there occurs emergency in car.

Application number be 200910013852.3 application for a patent for invention propose a kind of rapid intelligent public transport stereo monitoring apparatus and method of work thereof, from technical scheme disclosed in it, by Real-time Collection bus interior video information, platform environment and the passenger's video information uploading of waiting.The real-time grasp of control center to passengers quantity can be improved, and then reasonably distribute bus, but he shortcoming is also similar to above-mentioned utility model patent, real-time data traffic is uploaded and is made a lot of part be unreasonable consumption and not accomplish the process to video sequence, also cannot detect in car and whether riot occur give warning in advance, reduce casualties, for investigation provides advantageous information evidence, while realizing security purpose, reduce the consumption of social resources.

Can find out that above prior art has all related to the real-time monitoring in public transport in bus and uploaded, but when there is riot in car, the function of but urgent detection and early warning.

Summary of the invention

Technical matters to be solved by this invention is the defect for background technology, proposes the image processing method of riot in a kind of real-time inspection vehicle.

The present invention is for solving the problems of the technologies described above by the following technical solutions:

A kind of public transport emergency detection method, comprises the following steps:

(1), by original image denoising, mark off the scope of activities of crowd in car, extract the video image in this scope of activities, use and improve frame difference method and pixel matching method extraction video foreground region;

(2), the unique point extracted in video foreground region, set up foreground moving region vector field by optical flow method;

(3), according to the change of average area kinetic energy in the vector field of foreground moving region and average unique point distance potential energy, judge whether emergency occurs in car.

Further, a kind of public transport emergency detection method of the present invention, marks off the scope of activities of crowd in car, is specially described in step (1):

First extract crowd movement region from the movement locus of crowd, adopt formula (1) to be subtracted each other by adjacent two frame video images and obtain moving region profile, then adopt the accumulative all profiles of formula (2) to carry out pursuit movement track L _sum;

L _subi＝frame _i-frame _i-1(1)

$L_{\mathrm{sum}}=\max (\left(\underset{i}{\overset{n}{\mathrm{\Σ}}}{L}_{\mathrm{subi}}\right)>T)---\left(2\right)$

Frame in formula _irepresent the i-th frame of sequence of video images, L _subirepresent that two frame video images subtract each other the profile drawn, the threshold value of T representative setting.

Further, a kind of public transport emergency detection method of the present invention, first carries out mean filter to raw video image before extraction moving region, and wherein average adopts the pixel value in kernel function K*K window on average to export afterwards, and kernel function is:

F represents kernel function, and f.w*f.h represents the size of window.

Further, a kind of public transport emergency detection method of the present invention, the size of window gets 3*3.

Further, a kind of public transport emergency detection method of the present invention, use frame difference method extraction video foreground region specific as follows:

S ₁＝L _sum1(4)

S _i＝S _i-1∩L _subi-1(5)

S _i＝(S _i+L _subi-1)>T (6)

Wherein, S _ibe the foreground area of the i-th two field picture, L _sum1for front cross frame video image subtracts each other the profile drawn; The frame difference method improved is simply efficient, improves the operating rate of algorithm

Further, utilize pixel matching method that extracted foreground area is divided into nonoverlapping fritter, if the number of pixels in fritter is greater than certain threshold value T, just extract this region, otherwise give up, pixel matching method can extract the foreground moving module of video sequence accurately; If the size of fritter is w*h, and w<h, be expressed as:

$\underset{i}{\overset{w}{\mathrm{\&Sigma;}}}\underset{j}{\overset{h}{\mathrm{\&Sigma;}}}{l}_{p(i,j!=0)}>T---\left(7\right)$

Wherein, p (i, j) is the pixel value of (i, j) position, and w is the width of fritter, and h is the length of fritter.

Further, a kind of public transport emergency detection method of the present invention, step (2) is described sets up foreground area motion vector field by optical flow method, it is application optical flow method, extract the direction of motion of foreground area, try to achieve the velocity field of foreground area, according to the velocity vector feature of each unique point, image is analyzed, specific as follows:

Adopt Lucas-Kanade optical flow method, calculate the movement of the position of each angle point between two frames, namely for certain two field picture f _iin certain angle point (x _i, y _i) there is a velocity v _i, according to all angle points, using a two field picture as a motion vector field.

Further, a kind of public transport emergency detection method of the present invention, in step (3):

Average area kinetic energy E _averi:

$E_{\mathrm{sumi}}=\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}\frac{1}{2}{\mathrm{mv}}^2---\left(8\right)$

E _averi＝E _sumi/area (9)

The E tried to achieve in formula (8) _sumifor the kinetic energy in each two field picture foreground area and, m represents the quality of angle point, and v represents the speed of angle point movement, and in formula (9), area represents the size of this two field picture foreground area;

Average characteristics point distance potential energy, i.e. the mean value of the Euclidean distance sum of angle point and car door in certain two field picture:

$d_{\mathrm{sum}}=\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}\sqrt{{(x_i-d_x)}^2+{(y_i-d_y)}^2}/n---\left(10\right)$

In formula, (x _i, y _i) be the position of angle point, (d _x, d _y) representing the position of car door, n is video sequence frame number.

When each average area kinetic energy and average unique point distance potential energy are all undergone mutation, then judge exception.

The present invention adopts above technical scheme compared with prior art, has following technique effect:

Method operating rate of the present invention is fast, and efficiency is high, the riot situation detected in car that can be real-time when there is emergency.

Accompanying drawing explanation

Fig. 1 is overall flow figure of the present invention.

Fig. 2 utilizes pixel matching method that extracted foreground area is divided into nonoverlapping fritter schematic diagram; Wherein (a) is the fritter schematic diagram extracted, and (b) is the fritter schematic diagram after improving, and (c) is the fritter schematic diagram after reduction.

Fig. 3 is kinetic energy histogram and the distance potential energy diagram of video sequence, and wherein (a) is kinetic energy histogram, and (b) is distance potential energy diagram.

Embodiment

Below in conjunction with accompanying drawing, technical scheme of the present invention is described in further detail.Those skilled in the art of the present technique are understandable that, unless otherwise defined, all terms used herein (comprising technical term and scientific terminology) have the meaning identical with the general understanding of the those of ordinary skill in field belonging to the present invention.Should also be understood that those terms defined in such as general dictionary should be understood to have the meaning consistent with the meaning in the context of prior art, unless and define as here, can not explain by idealized or too formal implication.

Car in the process of moving, scene light is outside window according to all changing, and handrail can shake in car, personage also can rock, walk about in car, and therefore the foreground extraction noise not only will left out outside window also wants the rocking of personage in elimination car, walking about of a few peoples judge on final riot the impact that produces.In time there is riot in car, people first reaction be exactly toward door race, therefore prospect motion energy and all can undergo mutation from the distance potential energy of door, just substantially can judge whether riot occurs in car by these two.Therefore target of the present invention is when car is in traveling operation process, during the riot situations such as Ruo Chenei kidnaps, spontaneous combustion, moving target can be isolated by algorithm through image processing techniques timely, judge whether there occurs riot in car by the kinetic energy of moving target and distance potential energy.

The present invention is directed to monitoring technique in the vehicle-mounted bus of wisdom, propose the image processing method of riot in a kind of real-time inspection vehicle, the riot major embodiment in car is crowd's flowing fast.Therefore first this algorithm marks off the main activities scope of crowd, use the frame difference method and pixel matching method extraction foreground area improved, then the unique point in video foreground is extracted, set up motion vector field by optical flow method, finally judge whether riot occurs according to average area kinetic energy and average unique point distance potential energy.

As shown in Figure 1, the process of method is as follows:

1.1 main motion regions (hot map) are extracted

First extract main movement region from the movement locus of crowd, for extracting main movement region, adopt video two frame subtract to obtain profile formula (1), then the algorithm of accumulative all profiles carrys out pursuit movement track L _sum.

L _subi＝frame _i-frame _i-1(1)

$L_{\mathrm{sum}}=\max (\left(\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}{L}_{\mathrm{subi}}\right)>T)---\left(2\right)$

Frame in formula _irepresent the i-th frame of video sequence, L _subirepresent the profile that two frame subtract draw.

There is larger noise in video, therefore first carry out mean filter to original image before treatment, on average export afterwards with the pixel value in K*K window, kernel function is:

F represents kernel function, and f.w*f.h represents the size of window, and the present invention gets the core size of 3*3.

In conjunction with filtering, draw main movement region according to as above algorithm, the band of position that namely crowd is residing in the picture, below will study for this block region.

1.2 sport foregrounds are extracted

For reducing the noise that background area produces subsequent operation, need to extract foreground moving object.Travelling speed of the present invention is fast and little by external influence factors such as illumination, and for Real-time security monitoring provides solid foundation, the average used time processing ten frame pictures is 140--170ms.

Therefore this algorithm visible has real-time and is subject to the feature that ectocine is little, and arthmetic statement is as follows:

S ₁＝L _sum1(4)

S _i＝S _i-1∩L _subi-1(5)

S _i＝(S _i+L _subi-1)>T (6)

S _ibe the prospect of the i-th two field picture, L _sum1for formula (2) gained, the improvement of the method be by former frame gained profile and present frame profile phase with, add the significant degree of profile, the profile of moving object can be extracted more clearly, due to car rocking under steam, in the prospect of extraction, also has certain interference.

Due to the existence of above-mentioned interference, the present invention utilizes pixel matching method that extracted sport foreground is divided into nonoverlapping fritter, if the number of pixels in fritter is greater than certain threshold value T, just extract this region, otherwise give up, what the method can make up above-mentioned algorithm rocks interference, makes the prospect extracted more accurate.If the size of fritter is w*h, because people is in morphologic performance, get w<h.Can be expressed as:

$\underset{i}{\overset{w}{\mathrm{\&Sigma;}}}\underset{j}{\overset{h}{\mathrm{\&Sigma;}}}{l}_{p(i,j!=0)}>T---\left(7\right)$

P (i, j) is the pixel value of (i, j) position.(a) in Fig. 2 fritter for extracting, the black block in connected region is set to foreground area by (b) in Fig. 2, and will independently foreground area cast out, (c) in Fig. 2 is the fritter of reduction.

1.3 feature point extraction

The present invention's application optical flow method, extracts the direction of motion of prospect, tries to achieve the velocity field of foreground image surveyed area, according to the velocity vector feature of each unique point, analyze image, therefore will obtain suitable unique point before this.

1.4 optical flow method extract foreground motion vector

According to the angle point obtained, calculate optical flow field, the present invention uses Lucas-Kanade optical flow method, and it calculates the movement of the position of each angle point between two frames.Namely for certain two field picture f _iin certain angle point (x _i, y _i) there is a velocity v _i, therefore according to all angle points, a two field picture is exactly a motion vector field.

During due to emergency occurring, can there is riot in crowd, and the first reaction is run to car door, and therefore, the present invention utilizes mean kinetic energy to judge whether and riot occurs.

$E_{\mathrm{sumi}}=\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}\frac{1}{2}{\mathrm{mv}}^2---\left(8\right)$

E _averi＝E _sumi/area (9)

Try to achieve in formula (8) for the kinetic energy in each frame and, the present invention supposes that each m is 1, t can directly derive from frequency frame, and owing to being weigh, therefore formula (8) can be revised as in formula (9), area represents the size of this two field picture foreground area.

But still have error to exist, if to do acutely rocking back and forth in original place than a people, still can be detected as kinetic energy comparatively large, therefore the present invention introduces the concept of distance potential energy, i.e. the mean value of the Euclidean distance sum of angle point and car door in certain two field picture.

$d_{\mathrm{sum}}=\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}\sqrt{{(x_i-d_x)}^2+{(y_i-d_y)}^2}/n---\left(10\right)$

(d in formula _x, d _y) representing the position at back door, n is video sequence frame number.

It is visible when each kinetic energy is undergone mutation, if riot, then distance potential energy also can be undergone mutation, if the original place of people is acutely rocked, distance potential energy can keep compared with moderate fluctuation, within a certain period of time because he very large variation can not occur within a certain period of time from position, back door.

As the kinetic energy histogram that (a) in Fig. 3 is video sequence, can learn from figure and have 8 peak values, when each generation peak value, calculate the distance potential energy at this place, visible have the distance potential energy at two places can keep certain stable after kinetic energy is undergone mutation in (b) of Fig. 3, and at this moment the present invention will get rid of and riot and people occur in car, acutely rocks the error that both of these case brings.

The above is only some embodiments of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (7)

1. a public transport emergency detection method, is characterized in that, comprises the following steps:

(1), by original image denoising, mark off the scope of activities of crowd in car, extract the video image in this scope of activities, the frame difference method using profile to deepen and pixel matching method extract video foreground region;

(2), the unique point extracted in video foreground region, set up foreground moving region vector field by optical flow method;

(3), according to the change of average area kinetic energy in the vector field of foreground moving region and average unique point distance potential energy, judge whether emergency occurs in car.

2. a kind of public transport emergency detection method according to claim 1, is characterized in that, mark off the scope of activities of crowd in car, be specially described in step (1):

First extract crowd movement region from the movement locus of crowd, adopt formula (1) to be subtracted each other by adjacent two frame video images and obtain moving region profile, then adopt the accumulative all profiles of formula (2) to carry out pursuit movement track L _sum;

L _subi＝frame _i-frame _i-1(1)

$L_{\mathrm{sum}}=\max (\left(\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}{L}_{\mathrm{subi}}\right)>T)---\left(2\right)$

Frame in formula _irepresent the i-th frame of sequence of video images, L _subirepresent that adjacent two frame video images subtract each other the profile drawn, the threshold value of T representative setting.

3. a kind of public transport emergency detection method according to claim 2, it is characterized in that, before extraction moving region, first carry out mean filter to raw video image, wherein average adopts the pixel value in kernel function K*K window on average to export afterwards, and kernel function is:

F represents kernel function, and f.w*f.h represents the size of window.

4. a kind of public transport emergency detection method according to claim 3, it is characterized in that, the size of window gets 3*3.

5. a kind of public transport emergency detection method according to claim 2, is characterized in that, it is specific as follows that the frame difference method using profile to deepen extracts video foreground region:

S ₁＝L _sum1(4)

S _i＝S _i-1∩L _subi-1(5)

S _i＝(S _i+L _subi-1)>T (6)

Wherein, S _ibe the foreground area of the i-th two field picture, L _sum1for front cross frame video image subtracts each other the profile drawn;

Utilize pixel matching method that extracted foreground area is divided into nonoverlapping fritter, if the number of pixels in fritter is greater than certain threshold value T, just extracts this region, otherwise give up; If the size of fritter is w*h, and w<h, be expressed as:

$\underset{i}{\overset{w}{\mathrm{\&Sigma;}}}\underset{j}{\overset{h}{\mathrm{\&Sigma;}}}{1}_{p(i,j)!=0}>T---\left(7\right)$

Wherein, p (i, j) is the pixel value of (i, j) position, and w is the width of fritter, and h is the length of fritter.

6. a kind of public transport emergency detection method according to claim 1, it is characterized in that, step (2) is described sets up foreground area motion vector field by optical flow method, it is application optical flow method, extract the direction of motion of foreground area, try to achieve the velocity field of foreground area, according to the velocity vector feature of each unique point, image is analyzed, specific as follows:

Adopt Lucas-Kanade optical flow method, calculate the movement of the position of each angle point between two frames, namely for certain two field picture f _iin certain angle point (x _i, y _i) there is a velocity v _i, according to all angle points, using a two field picture as a motion vector field.

7. a kind of public transport emergency detection method according to claim 6, is characterized in that, in step (3):

Average area kinetic energy E _averitry to achieve according to under type:

$E_{\mathrm{sumi}}=\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}\frac{1}{2}{\mathrm{mv}}^2---\left(8\right)$

E _averi＝E _sumiarea (9)

The E tried to achieve in formula (8) _sumifor the kinetic energy in each two field picture foreground area and, m represents the quality of angle point, and v represents the speed of angle point movement, and in formula (9), area represents the size of this two field picture foreground area;

Average characteristics point distance potential energy, i.e. the mean value of the Euclidean distance sum of angle point and car door in certain two field picture:

$d_{\mathrm{sum}}=\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}\sqrt{{(x_i,d_x)}^2+{(y_i-d_y)}^2}/n---\left(10\right)$

In formula, (x _i, y _i) be the position of angle point, (d _x, d _y) representing the position of car door, n is video sequence frame number;

When each average area kinetic energy and average unique point distance potential energy are all undergone mutation, then judge exception.