CN105007433A - Target-based moving object arrangement method enabling energy constraint minimization - Google Patents

Abstract

The invention relates to a target-based moving object arrangement method enabling energy constraint minimization. The method mainly comprises the following steps: 1) performing occupied distribution space processing for all movement tracks by considering sizes and paths of objects; and 2) offering three energy definitions including track split energy, track bump energy and relative time consistency energy, summing the three types of energy on the basis of a weight ratio, and a track and an arrangement mode that enable a minimal sum value are obtained through an energy constraint minimization algorithm. Through the above method, a re-arranged abstract video can be quickly browsed without information loss, space, time and logic relations of moving objects in an original monitor video are stored, a video compression ratio is increased, and users can quickly and logically view a monitor video.

Description

The minimized moving object aligning method of a kind of object-based energy constraint

Technical field:

The present invention relates to the minimized moving object aligning method of a kind of object-based energy constraint, belong to technical field of video monitoring.

Background technology

Due to the progress of science and technology and the raising of economic level, supervision of the cities camera spreads all over every nook and cranny, and the monitor video obtained thus also increases in geometric progression, especially 24 hours round-the-clock monitor videos.But, the consulting to browse of massive video extremely takes time and effort, most of monitor video comprises the object of which movement of little interest on the one hand, as few people's process street or the late into the night people from road surface car motion very sparse, on the other hand, interested moving object to being found from the video of magnanimity, must expending with recording the equal time browsing monitor video, therefore most monitor video never viewed and examination after recording completes.How intactly to browse long monitor video fast, conveniently lock searching object, meet public security, net the various demands of prison and criminal investigation, become the active demand of current monitoring trade.

The video frequency abstract based on key frame traditional at present lost a large amount of action messages by frame sampling technology, uses, especially in the use of monitor video, be difficult to reach desirable effect in reality.Object-based video frequency abstract plays important role in intelligent video analysis and video frequency searching, the activity time of the different time sections in original video is compressed in a shorter video and occurs simultaneously by it, reaches fast browsing and does not lose the object of action message.

Summary of the invention

The object of the present invention is to provide the minimized moving object aligning method of a kind of object-based energy constraint, moving target in original tediously long monitor video is moved on a timeline, be compressed to according to user's request when locus allows in the video of a section shorter and occurred simultaneously, the video frequency abstract generic compared to other, the present invention adopts energy constraint minimization algorithm, make video frequency abstract compression efficiency raising over time and space, readable to strengthen, greatly reduce public security, criminal investigation is used for browsing the time of checking monitor video.

The present invention includes following step:

1) size of binding object itself and path thereof, to take up room distribution process to all movement locus;

2) define three kinds of energy, track desintegration energy, track collision energy and relative time consistency energy, sum up these three energy according to weight ratio, and utilize Energy minimization to obtain making to add and be worth minimum track to occur arrangement mode;

3) synthetic video summary: by seamless spliced in background video for the motion tubes rearranged, produces a video frequency abstract over time and space, this summary is relative compact and contain the video of required activity in original video on Time and place.

Further, step 1) size of binding object itself and path thereof, first distribution space process is taken to all movement locus.Each motion tubes is sued for peace, obtain its corresponding activity value (summing value is taken the logarithm) and be normalized, the summation of these pipelines is put on same pictures, namely its spatial movement distribution (spatial activity distribution) is obtained, show with the form of picture, the place that in picture, brightness is larger, illustrates that the object moved in this track is more, otherwise the object of darker place motion is less.Again every two pipelines are placed separately, ask the degree of correlation of two pipelines.The degree of correlation between pipeline is the similar contract of equivalence of pipeline matrix.According to the compact requirement of the summary of user to summarized radio, determine a suitable threshold value.

$f_{correlation}(b,b^{\&prime;})=\left\{\begin{array}{c}1,correlation\&GreaterEqual;threshold\\ 0,correlation<threshold\end{array}\right.$

When the degree of correlation between pipeline is greater than threshold value, illustrates that these two tracks have overlap, whether unanimously need to proceed detector direction.Otherwise both explanations have the probability meeting or collide extremely low, make at track collision energy without the need to calculating.

Further, step 2) specifically comprise the following steps:

A) the track desintegration energy between pipeline is calculated respectively, track collision energy, and time consistency energy, according to the demand of user, adjust respective weight, and sum up.

$E=\underset{b\&Element;B}{\&Sigma;}{\mathrm{\&alpha;E}}_s\left(\hat{b}\right)+\underset{b,b^{\&prime;}\&Element;B}{\&Sigma;}\left({\mathrm{\&beta;E}}_c\right(\hat{b},{\hat{b}}^{\&prime;})+{\mathrm{\&epsiv;E}}_t(\hat{b},{\hat{b}}^{\&prime;}\left)\right)$

Wherein E _srepresent track desintegration energy, Ec expression activity collision energy, E _trepresent relative time consistency energy, B represents the set of motion tubes, and α, β and ε represent the weighted value arranged according to user's request, and β increases, and the collision in video frequency abstract reduces, and vice versa.

Preferably, described energy constraint mainly comprises following three:

A) track desintegration energy.In original video, if the people of two motions has the interactive action contacted with each other, be difficult to these two people to be separated exactly, often process as a people.When separating after the mutual certain hour of two people, then now suddenly a people will be detected in video more, and moment of occurring as second moving object in the moment two people be separated but not two people occur simultaneously, after having carried out relative time and occurring the calculating of energy, user cannot to have observed in original video the two Action logic relation in summarized radio.Track desintegration energy makes punishment to this situation exactly, retains Action logic relation therebetween.

represent the motion tubes in video frequency abstract, T _x, T _ytwo threshold values, if the position (x that moving object occurs first ₁, y ₁) value be all greater than its corresponding threshold value, the Euclidean distance representing the positional distance Video Edge that this object occurs first is comparatively large, and its position occurred in video frequency abstract is not the position that it occurs first in original video.

B) track collision energy.The motion event do not occurred in the same time is mainly passed through translation on a timeline by video frequency abstract, make it represent at synchronization as much as possible simultaneously, the situations such as this just inevitably occurs track cross, block, especially, when even speed is identical with direction for two people's movement locus, this circumstance of occlusion is particularly serious.

According to the degree of correlation before between our pipeline that calculated, if its degree of correlation does not exceed certain threshold value, illustrate and do not occur overlap between two tracks, even if both times of occurring in summarized radio and speed just the same, also there will not be the overlapping phenomenon being an impediment to observer.On the contrary, when the degree of correlation between pipeline is greater than threshold value, illustrates that these two tracks have overlap, whether unanimously need to proceed detector direction:

$Ec(\hat{b},{\hat{b}}^{\&prime;})=\left\{\begin{array}{c}\underset{x,y,t\&Element;t\hat{b}\&cap;t{\hat{b}}^{\&prime;}}{\&Sigma;}\mathrm{\&chi;}\hat{b}\left(x,y,t\right)\mathrm{\&chi;}{\hat{b}}^{\&prime;}\left(x,y,t\right),f_{correlation}(\hat{b},{\hat{b}}^{\&prime;})=1\\ 0,f_{correlation}(\hat{b},{\hat{b}}^{\&prime;})=0\end{array}\right.$

Wherein to represent in video frequency abstract two different motion tubes respectively, χ ^brepresent the characteristic equation of pipeline.If very little by the track collision energy calculated, illustrate that the direction of motion of the moving object in these two tracks is different, what the two only can be of short duration is staggered, does not affect the observation of observer to former moving object; Otherwise if the track collision energy obtained is greater than threshold value, the direction of both explanations is consistent, and energy is larger, and the time of blocking of both explanations is more, more has negative effect to the observation of observer.

C) relative time consistency energy.Time consistency energy is partial to keep life event order of occurrence in time in original video.In original monitor video, record the time that each moving object occurs, calibrate the order that it occurs, a time consistency penalty value is had between every two motion tubes, two moving objects appearance order in original video is more close, and its penalty value is less, on the contrary, then larger, its time consistency penalty value can be expressed as:

$E_t(\hat{b},{\hat{b}}^{\&prime;})=C\frac{|N\hat{b}-N{\hat{b}}^{\&prime;}|}{T_{total}}$

Wherein C is constant, represent the order that motion tubes b occurs in original video, T _totalrepresent that the frame number disappeared to last motion pedestrian movement detected from the frame number be checked through first is poor.Employing time sequencing calculates the method that Euclidean distance between energy two moving objects every compared to traditional calculating appears in relative time, and amount of calculation reduces a lot, and shorten computing time, judged result is also more accurate.

Further, step 3) utilize Poisson image editing algorithms in conjunction with the adjustment of transparency, the moving object sequence seamless joint rearranged utilizing energy minimization, in background, is eliminated color aliasing, is formed final video frequency abstract.

Accompanying drawing explanation

Fig. 1 is a kind of object-based energy constraint of the present invention minimized moving object aligning method block flow diagram.

Embodiment:

As shown in Figure 1, step of the invention process forms by calculating pipeline desintegration energy, the calculating track degree of correlation, calculating pipeline collision energy, calculating pipeline time consistency energy and synthetic video summary.Its concrete steps are as follows:

1, pipeline desintegration energy is calculated

Calculate the track desintegration energy between pipeline respectively, track collision energy, and relative time consistency energy, according to the demand of user, adjust respective weight, and sum up.

$E=\underset{b\&Element;B}{\&Sigma;}{\mathrm{\&alpha;E}}_s\left(\hat{b}\right)+\underset{b,b^{\&prime;}\&Element;B}{\&Sigma;}\left({\mathrm{\&beta;E}}_c\right(\hat{b},{\hat{b}}^{\&prime;})+{\mathrm{\&epsiv;E}}_t(\hat{b},{\hat{b}}^{\&prime;}\left)\right)$

Wherein E _srepresent track desintegration energy, Ec represents track collision energy, E _trepresent relative time consistency energy, B represents the set of motion tubes, and α, β and ε represent the weighted value arranged according to user's request, and wherein β increases, and the collision in video frequency abstract reduces, and vice versa.

Track desintegration energy.In original video, if the people of two motions has the interactive action contacted with each other, be difficult to these two people to be separated exactly, often process as a people.When separating after the mutual certain hour of two people, then now suddenly a people will be detected in video more, and moment of occurring as second moving object in the moment two people be separated but not two people occur simultaneously, after having carried out relative time and occurring the calculating of energy, user cannot to have observed in original video the two Action logic relation in summarized radio.Track desintegration energy makes punishment to this situation exactly, retains Action logic relation therebetween.

represent the motion tubes in video frequency abstract, T _x, T _ytwo threshold values, if the position (x that moving object occurs first ₁, y ₁) value be all greater than its corresponding threshold value, the Euclidean distance representing the positional distance Video Edge that this object occurs first is comparatively large, and its position occurred in video frequency abstract is not the position that it occurs first in original video.

2, the track degree of correlation is calculated

The motion event do not occurred in the same time is mainly passed through translation on a timeline by video frequency abstract, make it represent at synchronization as much as possible simultaneously, the situations such as this just inevitably occurs track cross, block, especially, when even speed is identical with direction for two people's movement locus, this circumstance of occlusion is particularly serious.

The size of binding object itself and path thereof, first carry out spatial movement distribution process to all movement locus.Each motion tubes is sued for peace, obtain its corresponding activity value (summing value is taken the logarithm) and be normalized, the summation of these pipelines is put on same pictures, namely its spatial movement distribution (spatial activity distribution) is obtained, show with the form of picture, the place that in picture, color is darker, illustrates that the object moved in this track is more, otherwise the object of darker place motion is less.Again every two pipelines are placed separately, ask the degree of correlation of two pipelines.The degree of correlation between pipeline is the similar contract of equivalence of pipeline matrix.According to the compact requirement of the summary of client to summarized radio, determine a suitable threshold value.

$f_{correlation}(b,b^{\&prime;})=\left\{\begin{array}{c}1,correlation\&GreaterEqual;threshold\\ 0,correlation<threshold\end{array}\right.$

When the degree of correlation is greater than threshold value, illustrate that the track of two pipelines occurs overlapping, need both continuation judgements direction whether consistent.

3, pipeline collision energy is calculated

According to the degree of correlation before between our pipeline that calculated, if its degree of correlation does not exceed threshold value, illustrate and do not occur overlap between two tracks, even if both times of occurring in summarized radio and speed just the same, also there will not be the overlapping phenomenon being an impediment to observer.On the contrary, when the degree of correlation between pipeline is greater than threshold value, illustrates that these two tracks have lap, whether unanimously need to proceed detector direction:

$Ec(\hat{b},{\hat{b}}^{\&prime;})=\left\{\begin{array}{c}\underset{x,y,t\&Element;t\hat{b}\&cap;t{\hat{b}}^{\&prime;}}{\&Sigma;}\mathrm{\&chi;}\hat{b}\left(x,y,t\right)\mathrm{\&chi;}{\hat{b}}^{\&prime;}\left(x,y,t\right),f_{correlation}(\hat{b},{\hat{b}}^{\&prime;})=1\\ 0,f_{correlation}(\hat{b},{\hat{b}}^{\&prime;})=0\end{array}\right.$

Wherein to represent in video frequency abstract two different motion tubes respectively, χ ^brepresent the characteristic equation of pipeline.If very little by the track collision energy calculated, illustrate that the direction of motion of the moving object in these two tracks is different, what the two only can be of short duration is staggered, does not affect the observation of observer to former moving object; Otherwise if the track collision energy obtained is greater than certain threshold value, the direction of both explanations is consistent, and energy is larger, and the time of blocking of both explanations is more, more has negative effect to the observation of observer.

4, pipeline time consistency energy is calculated

Relative time consistency energy.Relative time consistency energy is partial to keep life event order of occurrence in time in original video.In original monitor video, record the time that each moving object occurs, calibrate the order that it occurs, a time consistency penalty value is had between every two motion tubes, two moving objects appearance order in original video is more close, and its penalty value is less, on the contrary, then larger, its time consistency penalty value can be expressed as:

$E_t(\hat{b},{\hat{b}}^{\&prime;})=C\frac{|N\hat{b}-N{\hat{b}}^{\&prime;}|}{T_{total}}$

Wherein C is constant, represent the order that motion tubes b occurs in original video, T _totalrepresent that the frame number disappeared to last moving object campaign detected from the frame number be checked through first is poor.

Employing time sequencing calculates the method that Euclidean distance between energy two moving objects every compared to traditional calculating appears in relative time, and amount of calculation reduces a lot, and shorten computing time, judged result is also more accurate.

Utilizing greedy algorithm, the gross energy matrix obtained is processed, by finding energy minimization path, obtaining the motion tubes appearance order making gross energy minimum.

5, synthetic video summary

Utilize Poisson image editing algorithms, the moving object sequence seamless joint rearranged utilizing energy minimization, in background, is eliminated color aliasing, is formed final video frequency abstract.

The present invention is not limited to above-mentioned preferred forms; anyone can draw other various forms of products under enlightenment of the present invention; no matter but any change is done in its shape or structure; every have identical with the application or akin technical scheme, all drops within protection scope of the present invention.

Claims (4)

1. the minimized moving object aligning method of object-based energy constraint, is characterized in that, comprise the steps:

1) size of binding object itself and path thereof, to take up room distribution process to all movement locus;

2) define three kinds of energy, track desintegration energy, track collision energy and relative time consistency energy, sum up these three energy according to weight ratio, and utilize Energy minimization to obtain making to add and be worth minimum track to occur arrangement mode;

3) synthetic video summary: by seamless spliced in background video for the motion tubes rearranged, produces one over time and space activity video frequency abstract.

2. method according to claim 1, it is characterized in that, step 1) in the size of binding object itself and path thereof, first distribution space process is taken to all movement locus, each motion tubes is sued for peace, obtain its corresponding activity value and be normalized, the summation of these pipelines is put on same pictures, namely its spatial movement distribution is obtained, again every two pipelines are placed separately, ask the degree of correlation of two pipelines, the degree of correlation between pipeline is the similar contract of equivalence of pipeline matrix, according to the compact requirement of the summary of user to summarized radio, determine a suitable threshold value:

When the degree of correlation is greater than threshold value, illustrate that the track of two pipelines is overlapping, need both continuation judgements direction whether consistent; Otherwise both explanations have the probability meeting or collide extremely low, make at track collision energy without the need to calculating.

3. method according to claim 1, is characterized in that, step 2) be specially:

Calculate the track desintegration energy between pipeline respectively, track collision energy, and time consistency energy, according to the demand of user, adjust respective weight, and sum up:

Wherein E _srepresent track desintegration energy, Ec expression activity collision energy, E _trepresent relative time consistency energy, B represents the set of motion tubes, and α, β and ε represent the weighted value arranged according to user's request, and β increases, and the collision in video frequency abstract reduces, and vice versa;

Described energy constraint comprises:

A) track desintegration energy

represent the motion tubes in video frequency abstract, T _x, T _ytwo threshold values, if the position (x that moving object occurs first ₁, y ₁) value be all greater than its corresponding threshold value, the Euclidean distance representing the positional distance Video Edge that this object occurs first is comparatively large, and its position occurred in video frequency abstract is not the position that it occurs first in original video;

B) track collision energy, according to the compact requirement of the summary of client to summarized radio, determines a suitable threshold value,

When the degree of correlation between pipeline is greater than threshold value, illustrates that these two tracks have overlap, whether unanimously need to proceed detector direction:

Wherein to represent in video frequency abstract two different motion tubes respectively, χ _brepresent the characteristic equation of pipeline;

C) relative time consistency energy, its time consistency penalty value can be expressed as:

Wherein C is constant, represent the order that motion tubes b occurs in original video, T _totalrepresent that the frame number disappeared to last moving object campaign detected from the frame number be checked through first is poor.

4. method according to claim 1, it is characterized in that, step 3) utilize Poisson image editing algorithms in conjunction with the adjustment of transparency, the moving object sequence seamless joint rearranged utilizing energy minimization is in background, eliminate color aliasing, form final video frequency abstract.