Term " first ", " second ", " the 3rd " " in description and claims of this specification and above-mentioned accompanying drawing
Four " etc. (if present) is for distinguishing similar object, without for describing specific order or precedence.Should manage
Solution so data for using can be exchanged in the appropriate case, so that embodiments of the invention described herein for example can be removing
Order beyond those for illustrating herein or describing is implemented.Additionally, term " comprising " and " having " and theirs is any
Deformation, it is intended that covering is non-exclusive to be included, for example, containing process, method, system, the product of series of steps or unit
Product or equipment are not necessarily limited to those steps clearly listed or unit, but may include not list clearly or for this
A little processes, method, product or other intrinsic steps of equipment or unit.
In embodiments of the present invention, method for early warning is applied on the video camera of monitoring system, and video camera can be to video counts
The event of crossing the border according to being processed and to that may occur carries out early warning, can strive for for Security Personnel carries out treatment to the event of crossing the border
Time, and can in time prevent the generation of some events of crossing the border.And multiple video cameras of monitoring system have linkage function, i.e., one
Individual video camera after suspicious pedestrian is detected, can by the feature of the suspicious pedestrian by way of high in the clouds with other video cameras
It is shared, additionally, monitoring system can also link with gate control system, i.e., led by the monitoring that gate control system enters monitoring system in pedestrian
After domain, gate control system will notify that video camera carries out the detect and track of target.
Fig. 1 is referred to, is a kind of method for early warning in the embodiment of the present invention, the method includes:
101st, using in the background model extraction t two field pictures of the conservative modeling method structure of the random sampling based on pixel
All foreground blocks;
In embodiments of the present invention, method for early warning is applied on the video camera of monitoring system, and video camera can be to video counts
The event of crossing the border according to being processed and to that may occur carries out early warning, can strive for for Security Personnel carries out treatment to the event of crossing the border
Time, and can in time prevent the generation of some events of crossing the border.
In embodiments of the present invention, the background that video camera builds using the conservative modeling method of the random sampling based on pixel
Model, foreground blocks can be extracted using the background model from two field picture.
Wherein, the history pixel that each pixel for the two field picture for being shot comprising video camera in background model is randomly selected
The set of value, common, the position of the video camera installed in monitoring system is fixed, and the region for shooting is also fixation,
Video camera can build background model by the pixel value of the pixel on the multiple image for randomly selecting shooting.For example, M (x)=
{V1, V2... ..., Vn, wherein, x represents pixel, ViRefer to i-th history pixel value of pixel x, and pixel x picture
Sample in plain value set M (x) is not continuous, but randomly select.And the sets of pixel values M of the background model for building
In sample nor continuous, be also what is randomly selected.
In embodiments of the present invention, all foreground blocks during video camera can utilize background model to extract t two field pictures, its
In, foreground blocks generally refer to Pedestrians and vehicles in the shooting area of video camera etc., additionally, the background block in two field picture is usually
Refer to the fixed physics for existing, such as building, trees etc. in the shooting area of video camera.
It should be noted that background model is the conservative modeling method of the random sampling based on pixel build, using the back of the body
Scape model extraction foreground blocks, can effectively solve the problems, such as that the time complexity caused by global search is too high.
102nd, whether judged in all foreground blocks comprising target prospect block, target using the cascade sort model for pre-setting
Foreground blocks refer to feature and the matching degree of the feature of the suspicious pedestrian for pre-setting is higher than the threshold value that pre-sets;
In embodiments of the present invention, cascade sort model is pre-set in video camera, wherein, cascade sort model is used for
Distinguish whether the foreground blocks extracted are pedestrian and determine that whether pedestrian is suspicious pedestrian, and video camera will carry out reality to suspicious pedestrian
When track, in the case of the trend that suspicious pedestrian has into warning region, send early warning to control and monitor console and alert.
In embodiments of the present invention, video camera divides after all foreground blocks are extracted using the cascade for pre-setting
Whether class model is judged in all foreground blocks comprising target prospect block, and target prospect block refers to comprising suspicious pedestrian in feature
Feature.
If the 103, including target prospect block in all foreground blocks, video camera is utilized based on histogram intersection core HIK matchings
The Target Acquisition Model of algorithm predicts the route of target prospect block;
In embodiments of the present invention, if video camera determines to include target prospect block in the foreground blocks extracted, using base
The route of target prospect block is predicted in the Target Acquisition Model of histogram intersection core HIK matching algorithms.
104th, if it is determined that target prospect block route point to warning region, then to control and monitor console send early warning alarm.
In embodiments of the present invention, if the route sensing warning region of the target prospect block that video camera determines, should
Video camera sends early warning alarm to control and monitor console, wherein comprises at least target prospect block, early warning place etc. in early warning warning so that prison
The Security Personnel for controlling platform correctly can be processed the alarm in time after early warning alarm is seen.
It should be noted that.If the route of the target prospect block that video camera determines points to non-warning region, image
The route of the target prospect block is sent to control and monitor console by machine so that control and monitor console can draw the action rail of the target prospect block
Mark, to realize real-time tracking.
In embodiments of the present invention, the background mould that video camera is built using the conservative modeling method of the random sampling based on pixel
Type extract t two field pictures in all foreground blocks, judged using the cascade sort model for pre-setting be in all foreground blocks
It is no comprising target prospect block, wherein, target prospect block refers to feature high with the matching degree of the feature of the suspicious pedestrian for pre-setting
In the foreground blocks of the threshold value for pre-setting, if including target prospect block in all foreground blocks, based on histogram intersection core HIK
The Target Acquisition Model of matching algorithm predicts the route of target prospect block, and if it is determined that target prospect block route
Warning region is pointed to, then sending early warning to control and monitor console alerts, by the route of real-time estimate target prospect block, can be effective
The real-time tracking for realizing target prospect block and can effectively carry out early warning.
In order to be better understood from the technical scheme in the embodiment of the present invention, Fig. 2 is referred to, be pre- in the embodiment of the present invention
Another embodiment of alarm method, including:
201st, it is each in the set determination t two field pictures of the history pixel value according to the pixel included in background model
Individual pixel is foreground blocks pixel or background block pixel;
In embodiments of the present invention, video camera is by according to the set of the history pixel value of the pixel included in background model
Determine that each pixel in t two field pictures is foreground blocks pixel or background block pixel, specifically, video camera will be right
Each pixel performs following step:In determining several history pixel values of pixel i in background model, in picture
Vegetarian refreshments i it is current Euclidean circle region in pixel value number, Euclidean circle region refer to pixel i in t two field pictures
Centered on pixel value, the arest neighbors radius for pre-setting is the pixel value region that radius is formed;If in the current Europe of pixel i
History pixel value number in family name's circle region is less than or equal to the threshold value for pre-setting, then pixel i is foreground blocks pixel;If
History pixel value number in the current Euclidean circle regions of pixel i is more than the threshold value for pre-setting, then pixel i is the back of the body
Scape block pixel.
Wherein, the t frames refer to present frame.
202nd, the foreground blocks pixel determined in t two field pictures is divided into foreground blocks;
In embodiments of the present invention, video camera is it is determined that each pixel in t two field pictures is foreground blocks pixel
Or after background block pixel, the foreground blocks pixel of determination is divided into foreground blocks, wherein, if the mode of division refers to one
Individual pixel is foreground blocks pixel with any one pixel adjacent thereto, then two pixels are divided into same
In foreground blocks.
In embodiments of the present invention, video camera is divided into after foreground blocks by foreground pixel point, will be using pre-setting
Cascade sort model whether judge in all foreground blocks comprising target prospect block, wherein, cascade sort model is straight conversion
Scheme (Census Transform Histogram, CENTRIST) feature-forward direction feature selecting (Forward Feat in side
Selection, FFS)-disaggregated model that (Cascade) is constituted is cascaded, wherein, conversion histogram CENTRIST features are a kind of
Feature Descriptor based on profile, it is simple with computing, the advantage of the ability of local and global feature is taken into account, and to illumination and rotation
Transformation is changed with robustness.
Wherein, FFS is a kind of strong classification based training algorithm, based on iterating and greedy algorithm thought, is characterized in each
The weight of sample need not update after iteration, and each Weak Classifier has identical franchise to final classification results.And make
With FFS and Cascade structure disaggregated models because the two has very big advantage on detection rates are improved, one is ensure that
On fixed verification and measurement ratio, the time complexity of classification is reduced to greatest extent.
In embodiments of the present invention, above-mentioned cascade sort model includes pedestrian's disaggregated model and histogram intersection core HIK
Matching algorithm, wherein, wherein, pedestrian's classification mode is to train what is obtained by the way of offline, is the prison according to the video camera
The multitude of video data of the camera acquisition in control system, extract different appearances of the pedestrian under a variety of background environments
State, dressing and with the pedestrian's positive sample blocked, while the negative sample for not containing pedestrian is extracted, by strengthening study (Boost
Learning method) determines the feature of suspicious pedestrian.
Wherein, pedestrian's disaggregated model be based on above-mentioned CENTRIST features and FFS features, and using by the way of cascading come
Classified.
In order to be better understood from the technical scheme in the embodiment of the present invention, the instruction of pedestrian's disaggregated model is described in detail below
Practice process, including:
Step 1, the substantial amounts of monitor video data of collection, therefrom extract various attitudes, the pedestrian sample of dressing, and unification is returned
One positive sample for turning to preset size, for example:Sample size is 36*72 sizes.Choose street, square, building etc. and do not contain and appoint
The background block of what pedestrian is equally normalized to the sample of preset size as sample, for example, can also be 36*72 sizes.Choose
The negative sample that big figure (480*720) without pedestrian learns as enhancing.Positive negative sample is pre-processed, the conversion of each sample is extracted
Histogram CENTRIST features, constitute positive Negative training sample feature set.It is sky to initialize current cascade classifier.
Step 2, based on positive and negative sample characteristics collection, going out one using FFS Algorithm for Training has strong point of certain classification capacity
Class device, and add it to current cascade classifier.
Step 3, Negative training sample collection is processed using current cascade classifier, reject judicious negative sample.Use again
Current cascade classifier processes the negative sample of above-mentioned enhancing study, and the negative sample of misclassification is inserted into Negative training sample concentrates,
The negative sample of training is updated with this.Then proceed to perform step 2 untill cascade classifier reaches certain accuracy of detection.
Whether judged in foreground blocks comprising mesh using the cascade sort model that pre-sets to be better understood from video camera
The technical scheme of foreground blocks is marked, step 203 and step 204 is referred to:
203rd, the foreground blocks for being characterized as pedestrian are extracted from all foreground blocks using pedestrian's disaggregated model;
In embodiments of the present invention, video camera will be extracted from all foreground blocks using pedestrian's disaggregated model and be characterized as pedestrian
Foreground blocks.
Fig. 3 is referred to, in being the embodiment of the present invention, the schematic diagram of pedestrian's disaggregated model, in pedestrian's disaggregated model,
H1, H2 ... ..., Hr represent category level of pedestrian's disaggregated model to foreground blocks, and judgement from H1 to Hr is increasingly to be directed to
The judgement of details, for example:The judgement of H1 be foreground blocks profile, by profile determine be pedestrian or non-pedestrian, H2's sentences
It is disconnected then whether be comprising head for meeting conventional size and shape etc. in foreground blocks.d(1)、f(1)、d(2)、f(2)、……、d
R (), f (r) are the mark of the feature of foreground blocks, for example:One feature of foreground blocks is by being defined as inhuman (non-after H1
Human), then the foreground blocks need not be further continued for determining whether pedestrian, if the feature of the foreground blocks can by determination after H1
Can be pedestrian, then be input into H2 by by the feature d (1) of the foreground blocks, f (1), and if be judged as inhuman, stop to the foreground blocks
Judgement, the otherwise foreground blocks are probably pedestrian, and continuation is judged by next stage H, until last defeated from pedestrian's disaggregated model
Go out, determine whether the foreground blocks are pedestrian.
204th, the feature of the foreground blocks for being characterized as pedestrian is calculated according to histogram intersection core HIK matching algorithms and is pre-set
Suspicious pedestrian feature matching degree, if being more than the foreground blocks of threshold value for pre-setting comprising matching degree, it is determined that matching degree
It is target prospect block more than the foreground blocks of the threshold value for pre-setting;
In embodiments of the present invention, video camera, will be according to histogram intersection after the foreground blocks for being characterized as pedestrian are extracted
Core HIK matching algorithms calculate the feature of the foreground blocks for being characterized as pedestrian and the matching degree of the suspicious pedestrian for pre-setting, if comprising
Matching degree is more than the foreground blocks of the threshold value for pre-setting, it is determined that the foreground blocks that matching degree is more than the threshold value for pre-setting are target
Foreground blocks.
It should be noted that the feature of suspicious pedestrian is saved in embodiments of the present invention, in video camera, and the suspicious row
The feature of people can be that video camera analyzes server obtain, or control and monitor console by taking the photograph from substantial amounts of video data
Linkage function between camera, during the feature of the suspicious pedestrian being input into by monitoring personnel shared into video camera, is protected by video camera
Deposit.
Wherein, the server of control and monitor console determines that the mode of the feature of suspicious pedestrian is:Monitoring personnel opens the prison of video camera A
Control video, clicks on the Data Enter button that system is provided, and (query time section refers to comprising crime to the query time section of set information
The time period of process), system will be presented the video of the query time section in the form of single frames, it is assumed that the current time chosen
It is 15:00 to 15:05 this period, system can sample and provide the image of video camera A feedbacks this period, about 50 frames, monitor
Member only needs to that target rectangle frame is marked into (the mark of the sample of a suspect by mouse on the image for have a suspect
Number is determined by monitoring personnel, but be have to be larger than or equal to 1, marked the accuracy generation influence that number will be on detecting and following the trail of),
Click in confirmation typing, system will generate the feature of suspicious object.
Monitoring personnel also can be set the scope of the camera of the feature for sharing the suspicious pedestrian, for example, being escaped according to suspect
Its area size that may be escaped from substantially is estimated to the current time, the region representation of system default is with a specified ground
In a border circular areas centered on point, system provides the slider bar of zone radius, and monitoring personnel can come elastic with sliding button
Selection region size, it is determined that after, all video cameras that the region includes will receive server end by video camera linkage function
The feature of the suspicious pedestrian for sending, and beginning carries out target detection in respective monitor area.
In embodiments of the present invention, before video camera is calculated according to histogram intersection core HIK matching algorithms and is characterized as pedestrian
The matching degree of the feature of scape block and the feature of the suspicious pedestrian for pre-setting, before determining that matching degree is higher than the threshold value for pre-setting
Scape block is specifically as follows for target prospect block:Video camera determines the histogram of the foreground blocks for being characterized as pedestrian, is handed over according to histogram
Collection core HIK matching algorithms calculate the histogram of the foreground blocks and the histogrammic matching degree of the feature of suspicious pedestrian, it is determined that matching
Degree is target prospect block higher than the foreground blocks of the threshold value for pre-setting.
205th, determine that the pixel in target prospect block center is used as the first center pixel in t two field pictures
Point;
In embodiments of the present invention, if video camera extract foreground blocks in include target prospect block, be to the target before
Scape block is tracked, and crossing the border for may occurring is predicted.And if including target prospect in the foreground blocks of video camera extraction
Block, then video camera will first determine that the pixel in target prospect block center is used as imago in first in t two field pictures
Vegetarian refreshments.
206th, with the first central pixel point as the center of circle in t+1 two field pictures, presetting length is in the border circular areas of radius
Foreground blocks pixel in, extract centered on foreground blocks pixel corresponding segment conversion histogram CENTRIST features,
Using histogram intersection core HIK matching algorithms by the conversion histogram of the corresponding segment centered on foreground blocks pixel
CENTRIST features are matched with the conversion histogram CENTRIST features of the suspicious pedestrian for pre-setting, and determine matching degree most
Foreground blocks pixel high is the second central pixel point;
In embodiments of the present invention, video camera determine first central pixel point of the target prospect block in t two field pictures it
Afterwards, will be in t+1 two field pictures with the first central pixel point as the center of circle, the length for pre-setting is obtained in the region of the center of circle for radius
Foreground blocks pixel in, extract centered on foreground blocks pixel corresponding segment conversion histogram CENTRIST features,
Using histogram intersection core HIK matching algorithms by the conversion histogram CENTRIST features of the segment and the suspicious row for pre-setting
The conversion histogram CENTRIST features of people are matched, and determine that matching degree highest foreground blocks pixel is the second center pixel
Point;Fig. 4 is referred to, in being the embodiment of the present invention, the search model of the second central pixel point, wherein, the circular center of circle represents the
One central pixel point Pt, round radius is R, and stain represents foreground blocks pixel.
It should be noted that according to aforesaid way, video camera can in the manner described above determine the 3rd on t+2 two field pictures
Central pixel point, by that analogy, video camera can obtain multiple central pixel points, and the position for passing through the plurality of central pixel point
The route of variation prediction target prospect block.
Wherein, video camera determines that with the first central pixel point be the center of circle, and presetting length is the picture in the center of circle region of radius
Which is that the mode of foreground blocks pixel is in vegetarian refreshments:If video camera determines pixel i in the region in background model
In dry history pixel value, the number of the history pixel value in the current Euclidean circle regions of pixel i, and if in should
The number of the history pixel value in pixel i current Euclidean circle region is less than or equal to the threshold value for pre-setting, then pixel i
It is foreground blocks pixel, if the number of the history pixel value in the current Euclidean circle regions of pixel i is advance more than this
The threshold value of setting, then pixel i is background block pixel.
In embodiments of the present invention, matching hunting zone is defined by using the first central pixel point, Neng Gou
Time complexity is reduced while improving matching accuracy rate.
207th, the position prediction target prospect block according to the first central pixel point and the second central pixel point in two field picture
Route;
In embodiments of the present invention, video camera according to the first central pixel point and the second central pixel point in two field picture
The route of position prediction target prospect block.Wherein, one position of the target prospect block can determine that by the first central pixel point,
Another position of the target prospect block is can determine that by the second central pixel point, is pointed to by the first central pixel point defined location
The direction of the second central pixel point defined location is the route of the target prospect block.
It should be noted that being only with two routes of center pixel point prediction target prospect block in step 207
As a example by illustrate, in actual applications, can also determine in the two field picture of the above one by way of described in step 207
Central pixel point, determines the central pixel point in present frame, and target prospect block is predicted by the change in location of multiple pixels
Route.
208th, if it is determined that target prospect block route point to warning region, then video camera to control and monitor console send early warning
Alarm.
In embodiments of the present invention, however, it is determined that target prospect block route point to warning region, then video camera to
Control and monitor console sends early warning alarm, wherein, route sensing warning region refers to warning region in the action side of target prospect block
To signified region, and the nearest region of the distance target prospect block on this action direction.
It should be noted that in embodiments of the present invention, can also be according to the first central pixel point and the second central pixel point
Position in two field picture calculates the translational speed of target prospect block, for example, determining imago in the first central pixel point and second
Distance between position of the vegetarian refreshments in two field picture, moves the distance transform into target prospect block is actual according to the ratio for pre-setting
Dynamic distance, then the business of the time difference between the actual mobile distance of the target prospect block and two frames be the movement of target advance block
Translational speed.
In embodiments of the present invention, the feature of target prospect block, route, position etc. are comprised at least in early warning alarm
Deng, additionally can include the target prospect block translational speed.
It should be noted that if it is determined that the route of target prospect block point to non-warning region, and supervised in real time
Under control scene, then video camera video camera adjacent thereto shares the target prospect block so that other video cameras also can be right
The target prospect block is tracked.The video camera for getting the feature of target prospect block will be checked in video joining place, sent out
It is tracked after existing target prospect block so that be capable of achieving the tracking free of discontinuities to target prospect block.
It should be noted that if it is determined that the developing direction of sidelong glance mark foreground blocks points to non-warning region, video camera can also be by
The route of the target prospect block is sent to the server of control and monitor console so that what server can send according to multiple video cameras
The route of target prospect block draws the movement track of the target prospect block.
Wherein, video camera can select some features to be tracked from the feature of target prospect block, and conventional feature has:Face
Color, edge, local description and multiple features fusion etc., the feature based on color, it has preferably resists screen rotation, non-firm
Property deformation and partial occlusion ability, in the algorithm frequently with target area hue histogram feature as description son, it is excellent
Point is low computation complexity, and local matching accuracy rate is high.
In step 205 to step 207, the target search mould based on histogram intersection core HIK matching algorithms is described
Type predicts the route of target prospect block, different from Mean Shift algorithms of the prior art and Particle Filter
Algorithm, the change in location of target prospect block under the most frame frequencies of acquisition is searched for by characteristic matching, can solve the problem that mesh in the prior art
Mark the drifting problem of track band, it is also possible to reduce the too high time complexity caused because every frame all carries out pedestrian detection.
The Target Acquisition Model of above-mentioned histogram intersection core HIK matching algorithms determines target prospect block in each frame
Position is a kind of algorithm idea based on local optimum, based on target in the limited premise of adjacent interframe moving range, with part
Optimal replacement global optimum, to obtain conevying efficiency higher.But, the mode based on search is a kind of algorithm for estimating, is had
The defect of the accumulation of error, the position of target is that, by the location estimation of target in t frames, inherently there is certain in t+1 frames
Error, then estimate that the position of target in t+2 frames will be come in error accumulation by the position of target in t+1 frames, multiframe
Afterwards, the track of target will produce very big deviation, and the accuracy on early warning produces influence.In order to solve the problem, can adopt
The position of target prospect block is corrected with the mode of detection, to reduce or avoid error, is specifically as follows:In t+x*
In n two field pictures, the Target Acquisition Model based on histogram intersection core HIK matching algorithms determines the position of target prospect block, x and n
It is positive integer;Determine correction target prospect block using background model and cascade classifier, and target prospect block is corrected in utilization
Position is corrected by the position of target prospect block.
Wherein, step 201 is to described in 204 being to determine correction target prospect block using background model and cascade classifier
Mode, described in step 205 to 207 is position that the Target Acquisition Model based on matching algorithm determines target prospect block
Mode.
Wherein, can be using correcting target prospect block and target prospect block is corrected:By detecting the correction for obtaining
There is following relation between having the number i of target prospect block in the number j of target prospect block and the current of search, wherein, correction
The number of target prospect block is the number of the suspicious pedestrian for detecting, and the number of the current existing target prospect block of search is
To have tracked chain number, there is different matching strategies under different situations:
1) i==j;The number of the suspicious pedestrian for now detecting is equal with chain number has been tracked, by position and color
Histogram Matching is adjusted to realize the renewal to tracking chain.
2)i>j;The number of the suspicious pedestrian for now detecting first is carried out less than chain number has been tracked to tracking chain position
Judge, if more than range of video, beyond then deleting the tracking chain.To not matching the tracking chain of detection block by the position that searches
Put information updating tracking chain.
3)i<j;The number of the suspicious pedestrian for now detecting more than having tracked number, by remaining detection after matching
It, into the target of video, is that it creates new tracking chain that object is newly.
The method that search and detection more than are combined can reduce the offset amplitude of target, at utmost be fitted mesh
Target movement locus.
In embodiments of the present invention, set of the video camera in the history pixel value according to the pixel included in background model
After determining in t two field pictures each pixel and being foreground blocks pixel or background block pixel, will be in t two field pictures really
Fixed foreground blocks pixel is divided into foreground blocks, using the conversion histogram CENTRIST features-FFS-Cascade for pre-setting
Whether the cascade sort model of composition is judged comprising target prospect block in foreground blocks, if in the foreground blocks extracted comprising target before
Scape block, it is determined that in t two field pictures the pixel in target prospect block center as the first central pixel point, the
With the first central pixel point it is the center of circle in t+1 two field pictures, during presetting length is for the foreground blocks pixel in the center of circle region of radius,
In advance centered on foreground blocks pixel corresponding segment conversion histogram CENTRIST features, using histogram intersection core
The conversion histogram CENTRIST features of the segment that HIK matching algorithms will be obtained and the conversion Nogata of the suspicious pedestrian for pre-setting
Figure CENTRIST features are matched, and matching degree highest foreground blocks pixel are determined for the second central pixel point, using first
The route of the position prediction target prospect block of central pixel point and the second central pixel point in two field picture, and if it is determined that
Warning region is pointed in the dynamic direction of the feelings of target prospect block, then video camera sends early warning and reports to control and monitor console so that video camera can
The effective prediction realized to the tracking and realization of target prospect block to the event of crossing the border, reduces the probability of event generation of crossing the border.
And in embodiments of the present invention, video camera can complete the prediction to the event of crossing the border, it is to avoid send data to prison
The time that control platform brings wastes, and greatly improves early warning efficiency.
It should be noted that in embodiments of the present invention, the video camera in monitoring system can also link with gate control system,
When pedestrian enters the monitoring range of monitoring system by gate control system, gate control system notifies that the video camera for pre-setting carries out mesh
Target is detected and tracked, and video camera can carry out early warning according to the method for early warning described in the embodiment of the present invention to the event of crossing the border.
It should be noted that the early warning scheme described in the embodiment of the present invention it is also possible to use in public security monitoring system, and
The real-time tracking of crime one's share of expenses for a joint undertaking can be applied to and followed the trail of afterwards.In order to be better understood from the technical side in the embodiment of the present invention
Case, is described below specific application scenarios:
Monitoring personnel suspects that crime one's share of expenses for a joint undertaking A appears in B areas, then the control and monitor console in B areas is input into the feature of crime one's share of expenses for a joint undertaking A, prison
The server for controlling platform shares on all video cameras in B areas the feature of crime one's share of expenses for a joint undertaking A, and video camera preserves crime one's share of expenses for a joint undertaking A's
Feature.
Video camera D shifts to an earlier date current t frames using the background model that the conservative modeling method of the random sampling based on pixel builds
All foreground blocks in image, according to histogram intersection core HIK matching algorithms calculate be characterized as pedestrian foreground blocks feature with
The matching degree of the feature of the suspicious pedestrian for pre-setting, wherein, crime one's share of expenses for a joint undertaking A is included in suspicious pedestrian, determine that matching degree is more than
The target prospect block E of the foreground blocks of the threshold value for pre-setting, and the feature of the target prospect block E and crime one's share of expenses for a joint undertaking A matching degree
Highest.
Video camera D determines the pixel of the center in target prospect block E in t two field pictures as the first center
Pixel, with the first central pixel point is the center of circle in ground t+1 two field pictures, and presetting length is the prospect in the border circular areas of radius
In block pixel, corresponding segment converts histogram CENTRIST features centered on foreground blocks pixel in advance, using Nogata
Figure common factor core HIK matching algorithms are matched the conversion histogram CENTRIST features of the segment with the feature of crime one's share of expenses for a joint undertaking A,
Determine that matching degree highest pixel, for the second central pixel point, exists according to first central pixel point and the second central pixel point
Position in two field picture determines the route of target prospect block E.
And if the route of target prospect block E points to warning region, early warning alarm is sent to control and monitor console, in case crime
One's share of expenses for a joint undertaking A enters warning region, and if target prospect block E the non-warning region of route, by the route of crime one's share of expenses for a joint undertaking A
Feed back to control and monitor console so that control and monitor console can utilize the route of the crime one's share of expenses for a joint undertaking A of video camera feedback to determine crime one's share of expenses for a joint undertaking A's
Movement track.
Fig. 5 is referred to, is the schematic diagram of the structure of video camera in the embodiment of the present invention, including:
Extraction module 501, the background model for being built using the conservative modeling method of the random sampling based on pixel is extracted
All foreground blocks in t two field pictures;
Judge module 502, after obtaining all foreground blocks in the extraction module 501, using pre-setting
Cascade sort model and pedestrian's disaggregated model judge whether comprising target prospect block in all foreground blocks, before the target
Scape block refers to the foreground blocks of the matching degree higher than the threshold value for pre-setting of feature and the feature of the suspicious pedestrian for pre-setting;
Prediction module 503, if determining in all foreground blocks comprising target prospect block for the judge module 502,
The mode being then combined based on search and detection, according to the target prospect block in the t two field pictures and t+1 two field pictures
Change in location predict the route of the target prospect block;
Sending module 504, if pointing to security area for the route of the target prospect block of the prediction of the prediction module 503
Domain, then send early warning alarm to control and monitor console.
In embodiments of the present invention, extraction module 501 is built using the conservative modeling method of the random sampling based on pixel
Background model extracts all foreground blocks in t two field pictures, and then judge module 502 is by using the cascade sort mould for pre-setting
Whether type and pedestrian's disaggregated model is judged comprising target prospect block in all foreground blocks, wherein, target prospect block refer to feature with
The matching degree of the feature of the suspicious pedestrian for pre-setting is higher than the foreground blocks of the threshold value for pre-setting, and if judge module 502 is true
Target prospect block is included in fixed all foreground blocks, then the Target Acquisition Model prediction based on histogram intersection core HIK matching algorithms
The route of the target prospect block, finally, if the route sensing warning of the target prospect block that prediction module 503 determines
Region, sending module 504 sends early warning and alerts to control and monitor console.
In embodiments of the present invention, the background mould that video camera is built using the conservative modeling method of the random sampling based on pixel
Type extract t two field pictures in all foreground blocks, judged using the cascade sort model for pre-setting be in all foreground blocks
It is no comprising target prospect block, wherein, target prospect block refers to feature high with the matching degree of the feature of the suspicious pedestrian for pre-setting
In the foreground blocks of the threshold value for pre-setting, if including target prospect block in all foreground blocks, based on histogram intersection core HIK
The Target Acquisition Model of matching algorithm predicts the route of the target prospect block, and if it is determined that target prospect block action
Warning region is pointed in direction, then sending early warning to control and monitor console alerts, and is predicted by the route to target prospect block, energy
Enough effectively realize the real-time tracking of target prospect block and can effectively carry out early warning.
It is another embodiment of the structure of video camera in the embodiment of the present invention to refer to Fig. 6, including:
Extraction module 501 described in embodiment as shown in Figure 6, judge module 502, prediction module 503 and sending module
504, and it is similar to the content described in embodiment illustrated in fig. 5, here is omitted.
In embodiments of the present invention, each pixel for the two field picture for being shot comprising the video camera in the background model
The set of the history pixel value that point is randomly selected;
Then the extraction module 501 includes:
First determining module 601, for the set of the history pixel value according to the pixel included in the background model
Determine that each pixel in the t two field pictures is foreground blocks pixel or background block pixel;
Second determining module 602, for determining the t two field pictures in first determining module 601 in each
After pixel is foreground blocks pixel or background block pixel, the foreground blocks pixel that will be determined in the t two field pictures
It is divided into foreground blocks.
Wherein, first determining module 601 is gone through specifically for determining several in the background model of pixel i
In history pixel value, the number of the history pixel value in the current Euclidean circle regions of the pixel i, the Euclidean justifies region
Refer to that the arest neighbors radius for pre-setting is radius shape centered on the pixel value by the pixel i in the t two field pictures
Into pixel value region;If the number is less than or equal to the threshold value for pre-setting, the pixel i is foreground blocks pixel
Point;If the number is more than the threshold value for pre-setting, the pixel i is background block pixel.
In embodiments of the present invention, the cascade sort model is:Conversion histogram CENTRIST features-FFS-
The cascade sort model that Cascade is constituted, and pedestrian's disaggregated model and the histogram intersection core are included in cascade sort model
HIK matching algorithms;
Then the judge module 502 includes:
Cascade sort module 603, after obtaining all foreground blocks in the extraction module 501, using pedestrian
Disaggregated model extracts the foreground blocks for being characterized as pedestrian from all foreground blocks;
Object judgement module 604, after extracting the foreground blocks for being characterized as pedestrian in the cascade sort module 603,
According to the histogram intersection core HIK matching algorithms calculate described in be characterized as pedestrian foreground blocks feature with pre-set can
The matching degree of the feature of pedestrian is doubted, if being more than the foreground blocks of the threshold value for pre-setting comprising matching degree, it is determined that matching degree is more than
The foreground blocks of the threshold value for pre-setting are target prospect block.
In embodiments of the present invention, the prediction module 503 includes:
3rd determining module 605, in determining the foreground blocks for being characterized as pedestrian in the object judgement module 604
Comprising the target prospect block, it is determined that the pixel in the target prospect block center is made in the t two field pictures
It is the first central pixel point;
Matching module 606, after determining first central pixel point for the 3rd determining module 605, in t+1
With first central pixel point as the center of circle in two field picture, presetting length is the foreground blocks pixel in the border circular areas of radius
In, extract the corresponding segment histogram intersection core HIK matching algorithms conversion histogram centered on the foreground blocks pixel
CENTRIST features, will be corresponding centered on the foreground blocks pixel using the histogram intersection core HIK matching algorithms
The conversion histogram CENTRIST features of the conversion histogram CENTRIST features of segment and the suspicious pedestrian for pre-setting
Matched, determined that matching degree highest foreground blocks pixel is the second central pixel point;
Direction prediction module 607, after obtaining second central pixel point in the matching module 606, according to
The row of target prospect block described in the position prediction of first central pixel point and second central pixel point in two field picture
Dynamic direction.
In embodiments of the present invention, the video camera also includes:
Rectification module 608, in t+x*n two field pictures, the mesh based on the histogram intersection core HIK matching algorithms
Mark search model determines the position of the target prospect block, and the x and n are positive integer;
Detection module 609, for determining correction target prospect block using the background model and the cascade classifier, and
The position of the target prospect block is corrected using the position of the correction target prospect block.
In embodiments of the present invention, the first determining module 601 in extraction module 501 is according to the picture included in background model
The set of the history pixel value of vegetarian refreshments determines that each pixel in the t two field pictures is foreground blocks pixel or background
Block pixel, and the foreground blocks pixel determined in the t two field pictures is divided into foreground blocks by the second determining module 602,
Cascade sort module 603 in judge module 502 extracts feature using advance pedestrian's disaggregated model from all foreground blocks
It is the foreground blocks of pedestrian, and row is characterized as according to histogram intersection core HIK matching algorithms are calculated by object judgement module 604
The matching degree of the feature of the foreground blocks of people and the feature of the suspicious pedestrian for pre-setting, if being more than what is pre-set comprising matching degree
The foreground blocks of threshold value, it is determined that the foreground blocks that matching degree is more than the threshold value for pre-setting are target prospect block, if object judgement mould
Block 604 is characterized as determining mould comprising the target prospect block, the in prediction module 503 the 3rd in the foreground blocks of pedestrian described in determining
Block determines that the pixel in the target prospect block center is used as the first central pixel point in the t two field pictures, and by
, with first central pixel point as the center of circle in t+1 two field pictures, presetting length is in the border circular areas of radius for matching module 606
Foreground blocks pixel in, extract the conversion histogram CENTRIST of corresponding segment centered on the foreground blocks pixel special
Levy, using the histogram intersection core HIK matching algorithms by the conversion Nogata of the corresponding segment centered on the foreground blocks pixel
The conversion histogram CENTRIST features of the suspicious pedestrian that figure CENTRIST features pre-set with this are matched, it is determined that matching
Degree highest foreground blocks pixel is the second central pixel point, by direction prediction module 607 according to first central pixel point and
The route of target prospect block described in position prediction of second central pixel point in two field picture.In the embodiment of the present invention
In, in t+x*n two field pictures, rectification module 608 is based on the Target Acquisition Model of the histogram intersection core HIK matching algorithms
Determine the position of the target prospect block, the x and n is positive integer, and by detection module 609 is using the background model and is somebody's turn to do
Cascade classifier determines correction target prospect block, and utilizes the position of the correction target prospect block to the position of the target prospect block
Corrected.
In embodiments of the present invention, set of the video camera in the history pixel value according to the pixel included in background model
After determining in t two field pictures each pixel and being foreground blocks pixel or background block pixel, will be in t two field pictures really
Fixed foreground blocks pixel is divided into foreground blocks, using the conversion histogram CENTRIST features-FFS-Cascade for pre-setting
Whether the cascade sort model of composition judges include target prospect block in all foreground blocks, if including mesh in the foreground blocks extracted
Mark foreground blocks, it is determined that the pixel of target prospect block center is in t two field pictures as the first central pixel point,
With the first central pixel point as the center of circle in t+1 two field pictures, presetting length is the foreground blocks pixel in the center of circle region of radius
In point, the conversion histogram CENTRIST features of the corresponding segment centered on the foreground blocks pixel are extracted, using described
Histogram intersection core HIK matching algorithms are by the conversion histogram of the corresponding segment centered on the foreground blocks pixel
CENTRIST features are matched with the conversion histogram CENTRIST features of the suspicious pedestrian for pre-setting, it is determined that matching
Degree highest foreground blocks pixel is the second central pixel point.And according to the first central pixel point and the second central pixel point in frame
The route of the position prediction target prospect block in image, and if it is determined that the dynamic direction of feelings of target prospect block point to security area
Domain, then video camera send early warning to control and monitor console and report so that video camera can effectively realize to the tracking of target prospect block and
The determination to the event of crossing the border is realized, the probability of event generation of crossing the border is reduced.
It will be appreciated by those skilled in the art that all or part of step in realizing above-described embodiment method can be by
Program is completed come the hardware for instructing correlation, and described program can be stored in a kind of computer-readable recording medium, above-mentioned to carry
To storage medium can be read-only storage, disk or CD etc..
A kind of method for early warning provided by the present invention and video camera are described in detail above, for the one of this area
As technical staff, according to the embodiment of the present invention thought, will change in specific embodiments and applications, it is comprehensive
Upper described, this specification content should not be construed as limiting the invention.