Summary of the invention
In view of this, the purpose of the application is to propose parking lot CCTV monitoring system and the side of a kind of Intelligent target tracking
Method.
The parking lot CCTV monitoring system of Intelligent target tracking of the invention, comprising:
Block extraction module is moved, for extracting motion picture region in each frame video pictures;
Motion estimate module, for filtering out movement mesh in the motion picture region of each frame video pictures
Mark;
Motion feature computing module, for extracting the moving parameter information of the moving target, and according to the movement
The moving parameter information of target generates the motion feature of the moving target;
Abnormal object judgment module is classified for the motion feature to total movement target, and by moving target
Quantity is less than the moving target in the classification of normal quantity threshold value as tracking target;
And warning note module, it extracts and sends the video pictures containing tracking target.
Wherein, the movement block extraction module passes through frame difference method, optical flow method or the back based on mixed Gauss model
Scape calculus of finite differences extracts motion picture region from each frame video pictures.
Wherein, the motion estimate module is obtained for the motion picture region extracted in each frame video pictures
The lateral length and longitudinal length of the boundary rectangle in motion picture region, or obtain from the boundary rectangle of motion picture region
Heart point as the shape feature value in the motion picture region, and will be moved to the set of vectors of the motion picture edges of regions
The shape feature value and target shape template matching of picture area, to be sieved from the motion picture region of each frame video pictures
Select moving target.
Wherein, the motion feature computing module determines its boundary rectangle for the moving target of each frame video pictures,
And determine the boundary rectangle of same moving target in temporal former frame video pictures, it calculates boundary rectangle abscissa, indulge
Coordinate, width, height changing value, the motion feature as the moving target.
Wherein, abnormal object judgment module is described outer using the changing value of the boundary rectangle as a various dimensions vector
Connect rectangle abscissa, ordinate, width, height changing value as the value in various dimensions vector in each dimension;To one
The corresponding various dimensions vector of total movement target in section of fixing time in all videos picture executes K-means cluster, according to poly-
Moving target is divided into multiple classification by class result.
The present invention proposes a kind of parking lot CCTV monitoring method of Intelligent target tracking in turn, comprising the following steps:
Block extraction step is moved, for extracting motion picture region in each frame video pictures;
Motion estimate step, for filtering out movement mesh in the motion picture region of each frame video pictures
Mark;
Motion feature calculates step, for extracting the moving parameter information of the moving target, and according to the movement
The moving parameter information of target generates the motion feature of the moving target;
Abnormal object judgment step is classified for the motion feature to total movement target, and by moving target
Quantity is less than the moving target in the classification of normal quantity threshold value as tracking target;
And warning note step, it extracts and sends the video pictures containing tracking target.
Wherein, the movement block extraction step passes through frame difference method, optical flow method or the back based on mixed Gauss model
Scape calculus of finite differences extracts motion picture region from each frame video pictures.
Wherein, the motion estimate step is obtained for the motion picture region extracted in each frame video pictures
The lateral length and longitudinal length of the boundary rectangle in motion picture region, or obtain from the boundary rectangle of motion picture region
Heart point as the shape feature value in the motion picture region, and will be moved to the set of vectors of the motion picture edges of regions
The shape feature value and target shape template matching of picture area, to be sieved from the motion picture region of each frame video pictures
Select moving target.
Wherein, the motion feature calculates step, determines its boundary rectangle for the moving target of each frame video pictures,
And determine the boundary rectangle of same moving target in temporal former frame video pictures, it calculates boundary rectangle abscissa, indulge
Coordinate, width, height changing value, the motion feature as the moving target.
Wherein, abnormal object judgment step is described outer using the changing value of the boundary rectangle as a various dimensions vector
Connect rectangle abscissa, ordinate, width, height changing value as the value in various dimensions vector in each dimension;To one
The corresponding various dimensions vector of total movement target in section of fixing time in all videos picture executes K-means cluster, according to poly-
Moving target is divided into multiple classification by class result.
The present invention is suitable for being not provided with the unmanned parking lot of field management maintenance personnel, in order to guarantee inner part of parking lot vehicle
Park and driving conditions in people, vehicle safety and order, massive video is obtained for inner part of parking lot CCTV system photographs
Picture can be automated, intelligent knowledge by extracting moving target therein and its consecutive variations amount being extracted and clustered
Wherein there is no the vehicle or personage's moving target of abnormal case or security risk, such as the vehicle scratched that collides,
Drive in the wrong direction vehicle, the vehicle on the stifled road of stagnation for a long time etc., and by the personnel of normal routine walking, for a long time delay or aggregation people
Member etc., and perhaps vehicle is unfolded to track the video pictures frame that the personnel or vehicle will be present to abnormal moving target personnel
It is pushed to parking lot backstage manager, to greatly reduce the workload of parking lot CCTV monitoring, improves speed and effect
Rate, the case where avoiding the occurrence of monitoring dead angle or lag.
Specific embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Convenient for description, part relevant to related invention is illustrated only in attached drawing.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
As shown in Figure 1, the present invention provides a kind of parking lot CCTV monitoring system of Intelligent target tracking.Unmanned parking lot
CCTV system generally by laying photographic device throughout, wired or wireless video signal transmission network, background video
Server and monitor scope composition.The monitoring system provided by the invention can be set in the background video server.
Each frame video pictures that the monitoring system receives and stores for background video server mention moving target therein
It takes its motion feature and is analyzed, intelligence finds and identify the vehicle or the people that wherein there is abnormal case or security risk
Object target, and the video pictures for the vehicle or human target that abnormal case or security risk will be present send monitoring to and show
Show device, and give necessary prompt, checks and dispose for the administrative staff on backstage;It is realized by providing video pictures to mesh
Target tracking, reduces the workload of direct surveillance, improves the efficiency and accuracy of identification, and monitoring dead angle occurs for prevention.
Specifically, referring to fig. 2, the parking lot CCTV monitoring system of Intelligent target of the invention tracking, comprising:
Block extraction module is moved, for extracting motion picture region in each frame video pictures;
Motion estimate module, for filtering out movement mesh in the motion picture region of each frame video pictures
Mark;
Motion feature computing module, for extracting the moving parameter information of the moving target, and according to the movement
The moving parameter information of target generates the motion feature of the moving target;
Abnormal object judgment module is classified for the motion feature to total movement target, and by moving target
Quantity is less than the moving target in the classification of normal quantity threshold value as tracking target;
And warning note module, it extracts and sends the video pictures containing tracking target.
Wherein, the movement block extraction module passes through frame difference method, optical flow method or the back based on mixed Gauss model
Scape calculus of finite differences extracts motion picture region from each frame video pictures.Movement block extraction module takes from the background video
Business device obtains continuous videos image frame acquiring with set rate, arranging sequentially in time, sequentially in time can be with table
Up to for the n-th-m frame, the n-th-m+1 frame ..., the (n-1)th frame, n-th frame, (n+1)th frame ... the n-th+m-1 frame, the n-th+m frame.Assuming that we are by
N frame video pictures are as current video picture frame, then based on the upper first or posterior video of n-th frame video pictures frame and time
Image frame, by means such as frame difference method in the prior art, optical flow method or background subtractions based on mixed Gauss model,
The motion picture region in current video picture frame can be therefrom extracted, i.e. current video picture frame is upper first relative to the time
Or there are the regions that pixel changes for posterior video pictures frame.
It may include the vehicle of movement, personage etc., the main needle of the present invention from the motion picture region that video pictures frame extracts
Anomaly analysis, identification and tracking are realized to the moving target that there is abnormal vehicle or personage, so, motion estimate mould
Block filters out moving target from the motion picture region of each frame video pictures.The motion estimate module is directed to
The motion picture region extracted in each frame video pictures obtains lateral length and the longitudinal direction of the boundary rectangle in motion picture region
Length, or obtain from the central point of motion picture region boundary rectangle to the set of vectors of the motion picture edges of regions, as
The shape feature value in the motion picture region, and by the shape feature value in motion picture region and target shape template ratio
It is right, to filter out moving target from the motion picture region of each frame video pictures.Since personage is compared with vehicle,
Transverse and longitudinal width ratio or edge configuration exist significant different;Therefore, can by target shape template definition personnel or
The transverse and longitudinal width ratio or edge vectors group that vehicle respectively meets;To the motion picture region extracted in each frame video pictures,
Transverse and longitudinal width is compared than the transverse and longitudinal width ratio with template, or by the edge vectors group of its edge vectors group and template into
Row vector difference operation, it can be determined that the motion picture region in each frame video pictures belongs to personage and still falls within vehicle, thus
Extract the moving target of each frame video pictures.
The present invention establishes moving target table to each frame motion picture, and include in the table one frame video pictures of record is complete
Portion's moving target, it may be assumed that
Fn=< ON, 1, ON, 2... ON, i...ON, k>
FnThe moving target table for indicating n-th frame video pictures, if total k moving target in the frame picture, then ON, iIndicating should
I-th of moving target in n-th frame video pictures.
Motion feature computing module, for extracting the moving parameter information of the moving target, and according to the movement
The moving parameter information of target generates the motion feature of the moving target.The motion feature computing module regards each frame
The moving target of frequency picture determines its boundary rectangle, and determines same moving target in temporal former frame video pictures
Boundary rectangle, calculate boundary rectangle abscissa, ordinate, width, height changing value, the movement as the moving target is special
Sign.
Specifically, for adjacent two frames video pictures continuous in time, such as n-th frame video pictures and first the
N-1 frame video pictures utilize the positional relationship and/or above-mentioned shape in video pictures to the moving target in two frame video pictures
Shape characteristic value carries out matching primitives, to determine the same moving target in two frame video pictures.I.e. if the (n-1)th frame video pictures
In a moving target and n-th frame video pictures in a moving target horizontal and vertical change in location all in pre- spacing
From within the scope of and/or the above-mentioned shape feature value of the two moving targets is consistent, then it is assumed that this is n-th frame and the (n-1)th frame two
Same moving target in frame video pictures., whereas if one of horizontal and vertical change in location of the two moving targets exists
Except predetermined distance range or the consistent degree of shape feature value is less than threshold value, then is not considered as this in n-th frame and the (n-1)th frame
Two moving targets are the same moving targets.Respective movement in n-th frame and two frame video pictures of the (n-1)th frame is traversed in this way
Target is compared two-by-two, determines the same movement target on two frames;For existing in former frame but not sent out in a later frame
It is existing in matched moving target, it is believed that its moving target for belonging to disappearance;For existing but not having in former frame in a later frame
It is found that there are matched moving targets, it is believed that belong to newly-increased moving target.For the (n-1)th frame and n-th frame video pictures, lead to
Cross the relationship of the moving target of both target contingency table records:
M(n-1, n)=< (ON-1,1, ON, 1), (ON-1,2, ON, 2) ... (ON-1, i, ON, i) ... (ON-1, L, ON, L)>
M(n-1, n)Indicate the target association table between the (n-1)th frame and n-th frame video pictures, (ON-1,1, ON, 1) indicate (n-1)th
1st associated objects present in frame and n-th frame video pictures, that is, same moving target, similarly, (ON-1, t, ON, i) indicate
I-th of associated objects present in (n-1)th frame and n-th frame video pictures, that is, same moving target.Shared L matched
Moving target.
In turn, same moving target of the motion feature computing module for the (n-1)th frame and n-th frame video pictures, example
Such as (ON-1, i, ON, i), the motion feature of the moving target is calculated, Δ (O is expressed asN-1, i, ON, i).Motion feature computing module needle
To the moving target O of n-th frameN, iIt determines its boundary rectangle, and determines same moving target O in the (n-1)th video picturesN-1, i's
Boundary rectangle, calculate two boundary rectangle abscissas, ordinate, width, height changing value (Δ Xi, Δ Yi, Δ Wi, Δ Hi),
As the motion feature of the moving target, i.e. Δ (ON-1, i, ON, i)=(Δ Xi, Δ Yi, Δ Wi, Δ Hi).Changing value (Δ Xi, Δ
Yi, Δ Wi, Δ Hi) it is four dimensional vectors, wherein Δ Xi, Δ YiDisplacement of the boundary rectangle of the moving target on X, Y-axis;
ΔWiIndicate the width variation of the boundary rectangle of moving target, Δ HiIndicate the high variable quantity of the boundary rectangle of moving target.
In turn, the motion feature computing module is directed in the n-th-m frame, the n-th-m+1 frame ..., the (n-1)th frame, n-th frame, the (n+1)th frame ... n-th
+ m-1 frame, the identified same moving target in a series of video pictures frame such as n-th+m frame, according to the moving target even
Each changing value of the moving target is converted various dimensions vector by the changing value in continuous video pictures frame.It is specific next
It says, for moving target i, various dimensions vector:
ΔO1=... Δ (ON-2, i, ON-1, i), Δ (ON-1, i, ON, i), Δ (ON, i, ON+1, i) ...
A series of changing value i.e. using the moving target in continuous videos image frames is as a various dimensions vector, institute
State boundary rectangle abscissa, ordinate, width, height changing value as the value in various dimensions vector in each dimension.
Abnormal object judgment module, to the corresponding multidimensional of total movement target in certain period of time in all videos picture
Vector is spent, K-means cluster is executed, moving target is divided by multiple classification according to cluster result.For a time
Section (such as 24 hours one day), intercepts whole continuous videos pictures of the period, according to being described above, therefrom obtains each fortune
The various dimensions vector of moving-target, such as Δ Oi-1, Δ Oi, Δ Oi+1Etc..Abnormal object judgment module can be by this section of video frame
In target complete multi-Dimensional parameters changing value, execute such as K-means cluster calculation, be divided into N number of classification.Various dimensions vector
Similar moving target can be generally gathered in one or several classification.Most normal personnel or vehicle are come
It says, the various dimensions vector of variable quantity caused by movement is similar, therefore normal personnel or vehicle can be collected at one
Or in the more classification of several destination numbers.Conversely, for the moving vehicle that there are the abnormal behaviours such as retrograde, stifled road, collision
Target, or the personnel targets for being detained for a long time and assembling on road, and not according to normal straight line walk but
The personnel targets etc. for walking zigzag route, could be separately formed classification.Therefore the classification less for destination number, included
Moving target very likely there is abnormal behaviour.The less classification of destination number can be chosen to be analyzed, it specifically, can be with
One normal quantity threshold value is set, when moving target quantity is less than normal quantity threshold value in classification, then by the movement in the classification
Target is as tracking target.
Warning note module, for the tracking target of identification, this module can extract the picture of the video containing the tracking target
Face, and the video pictures containing tracking target are sent to monitor scope shown in FIG. 1, and give necessary prompt, for after
The administrative staff of platform check and disposition.
The present invention proposes a kind of parking lot CCTV monitoring method of Intelligent target tracking in turn, as shown in figure 3, including following
Step:
Block extraction step is moved, for extracting motion picture region in each frame video pictures;
Motion estimate step, for filtering out movement mesh in the motion picture region of each frame video pictures
Mark;
Motion feature calculates step, for extracting the moving parameter information of the moving target, and according to the movement
The moving parameter information of target generates the motion feature of the moving target;
Abnormal object judgment step is classified for the motion feature to total movement target, and by moving target
Quantity is less than the moving target in the classification of normal quantity threshold value as tracking target;
And warning note step, it extracts and sends the video pictures containing tracking target.
Wherein, the movement block extraction step passes through frame difference method, optical flow method or the back based on mixed Gauss model
Scape calculus of finite differences extracts motion picture region from each frame video pictures.In this step, obtained from the background video server
Continuous videos image frame being acquired with set rate, arranging sequentially in time, can be expressed as the n-th-m sequentially in time
Frame, the n-th-m+1 frame ..., the (n-1)th frame, n-th frame, (n+1)th frame ... the n-th+m-1 frame, the n-th+m frame.Assuming that we draw n-th frame video
Face, then based on the upper first or posterior video pictures frame of n-th frame video pictures frame and time, leads to as current video picture frame
The means such as frame difference method, optical flow method or the background subtraction based on mixed Gauss model in the prior art are crossed, it can be therefrom
The motion picture region in current video picture frame is extracted, i.e. current video picture frame is upper formerly or rear relative to the time
Video pictures frame there are the regions that pixel changes.
Wherein, the motion estimate step is obtained for the motion picture region extracted in each frame video pictures
The lateral length and longitudinal length of the boundary rectangle in motion picture region, or obtain from the boundary rectangle of motion picture region
Heart point as the shape feature value in the motion picture region, and will be moved to the set of vectors of the motion picture edges of regions
The shape feature value and target shape template matching of picture area, to be sieved from the motion picture region of each frame video pictures
Select moving target.
In this step, moving target table is established to each frame motion picture, which, which records in a frame video pictures, includes
Total movement target, it may be assumed that
Fn=< ON, 1, ON, 2... ON, i...ON, k>
FnThe moving target table for indicating n-th frame video pictures, if total k moving target in the frame picture, then ON, iIndicating should
I-th of moving target in n-th frame video pictures.
Specifically, for adjacent two frames video pictures continuous in time, such as n-th frame video pictures and first the
N-1 frame video pictures utilize the positional relationship and/or above-mentioned shape in video pictures to the moving target in two frame video pictures
Shape characteristic value carries out matching primitives, to determine the same moving target in two frame video pictures.I.e. if the (n-1)th frame video pictures
In a moving target and n-th frame video pictures in a moving target horizontal and vertical change in location all in pre- spacing
From within the scope of and/or the above-mentioned shape feature value of the two moving targets is consistent, then it is assumed that this is n-th frame and the (n-1)th frame two
Same moving target in frame video pictures., whereas if one of horizontal and vertical change in location of the two moving targets exists
Except predetermined distance range or the consistent degree of shape feature value is less than threshold value, then is not considered as this in n-th frame and the (n-1)th frame
Two moving targets are the same moving targets.Respective movement in n-th frame and two frame video pictures of the (n-1)th frame is traversed in this way
Target is compared two-by-two, determines the same movement target on two frames;For existing in former frame but not sent out in a later frame
It is existing in matched moving target, it is believed that its moving target for belonging to disappearance;For existing but not having in former frame in a later frame
It is found that there are matched moving targets, it is believed that belong to newly-increased moving target.For the (n-1)th frame and n-th frame video pictures, lead to
Cross the relationship of the moving target of both target contingency table records:
M(n-1, n)=< (ON-1,1, ON, 1), (ON-1,2, ON, 2) ... (ON-1, i, ON, i) ... (ON-1, L, ON, L)>
M(n-1, n)Indicate the target association table between the (n-1)th frame and n-th frame video pictures, (ON-1,1, ON, 1) indicate (n-1)th
1st associated objects present in frame and n-th frame video pictures, that is, same moving target, similarly, (ON-1, i, ON, i) indicate
I-th of associated objects present in (n-1)th frame and n-th frame video pictures, that is, same moving target.Shared L matched
Moving target.
In turn, for the same moving target of the (n-1)th frame and n-th frame video pictures, such as (ON-1, i, ON, i), calculate the fortune
The motion feature of moving-target is expressed as Δ (ON-1, i, ON, i).For the moving target O of n-th frameN, iDetermine its boundary rectangle, and
Determine same moving target O in the (n-1)th video picturesR-1, iBoundary rectangle, calculate two boundary rectangle abscissas, ordinate,
Changing value (the Δ X of width, heighti, Δ Yi, Δ Wi, Δ Hi), as the motion feature of the moving target, i.e. Δ (ON-1, i, ON, i)
=(Δ Xi, Δ Yi, Δ Wi, Δ Hi).Changing value (Δ Xi, Δ Yi, Δ Wi, Δ Hi) it is four dimensional vectors, wherein Δ Xi, Δ
YiDisplacement of the boundary rectangle of the moving target on X, Y-axis;ΔWiIndicate the width variation of the boundary rectangle of moving target,
ΔHiIndicate the high variable quantity of the boundary rectangle of moving target.In turn, in the n-th-m frame, the n-th-m+1 frame ..., (n-1)th
Frame, n-th frame, (n+1)th frame ... the n-th+m-1 frame, the identified same movement mesh in a series of video pictures frame such as n-th+m frame
Mark turns each changing value of the moving target according to the changing value of the moving target in continuous video pictures frame
Turn to various dimensions vector.Specifically, for moving target i, various dimensions vector:
ΔOi=... Δ (ON-2, i, ON-1, i), Δ (ON-1, i, ON, i), Δ (ON, i, ON+1, i) ...
A series of changing value i.e. using the moving target in continuous videos image frames is as a various dimensions vector, institute
State boundary rectangle abscissa, ordinate, width, height changing value as the value in various dimensions vector in each dimension.
Wherein, abnormal object judgment step is described outer using the changing value of the boundary rectangle as a various dimensions vector
Connect rectangle abscissa, ordinate, width, height changing value as the value in various dimensions vector in each dimension;To one
The corresponding various dimensions vector of total movement target in section of fixing time in all videos picture executes K-means cluster, according to poly-
Moving target is divided into multiple classification by class result.For a period (such as 24 hours one day), the time is intercepted
Whole continuous videos pictures of section therefrom obtain the various dimensions vector of each moving target, such as Δ O according to being described abovei-1,
ΔOi, Δ Oi+1Etc..Abnormal object judgment module can by the multi-Dimensional parameters changing value of the target complete in this section of video frame,
Such as K-means cluster calculation is executed, N number of classification is divided into.The similar moving target of various dimensions vector can generally be gathered in one
In a or several classification.For most normal vehicles and personnel, the various dimensions vector of variable quantity caused by movement
It is similar, therefore normal vehicle can be collected in the more classification of one or several destination numbers.Conversely, for presence
It drives in the wrong direction, the moving vehicle target of the abnormal behaviours such as stifled road, collision, and/or for being detained for a long time and assembling on road
Personnel targets, and do not walk according to normal straight line but walk the personnel targets etc. of zigzag route, could be separately formed point
Class.Therefore very likely there is abnormal behaviour in the classification less for destination number, the moving target for being included.It can choose
The less classification of destination number is analyzed, and specifically, a normal quantity threshold value can be set, when moving target number in classification
Amount is less than normal quantity threshold value, then using the moving target in the classification as tracking target.
The present invention is suitable for being not provided with the unmanned parking lot of field management maintenance personnel, in order to guarantee inner part of parking lot vehicle
Park and driving conditions in people, vehicle safety and order, massive video is obtained for inner part of parking lot CCTV system photographs
Picture can be automated, intelligent knowledge by extracting moving target therein and its consecutive variations amount being extracted and clustered
Wherein there is no the moving target of abnormal case or security risk, such as there is and drive in the wrong direction, blocks up the fortune of the abnormal behaviours such as road, collision
Dynamic vehicle target, for the personnel targets be detained for a long time and assembled on road, and do not walk according to normal straight line and
To walk the personnel targets etc. of zigzag route, and to abnormal moving target personnel and vehicle expansion tracking, will be present the vehicle or
The video pictures frame of personnel is pushed to parking lot backstage manager, to greatly reduce the work of parking lot CCTV monitoring
The case where amount, improves speed and efficiency, avoids the occurrence of monitoring dead angle or lag.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the application, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from foregoing invention design, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed herein
Can technical characteristic replaced mutually and the technical solution that is formed.