CN110443097A - A kind of video object extract real-time optimization method and system - Google Patents
A kind of video object extract real-time optimization method and system Download PDFInfo
- Publication number
- CN110443097A CN110443097A CN201810413067.6A CN201810413067A CN110443097A CN 110443097 A CN110443097 A CN 110443097A CN 201810413067 A CN201810413067 A CN 201810413067A CN 110443097 A CN110443097 A CN 110443097A
- Authority
- CN
- China
- Prior art keywords
- target area
- image frame
- area
- minimum
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000005457 optimization Methods 0.000 title claims abstract description 34
- 230000011218 segmentation Effects 0.000 claims description 13
- 230000035772 mutation Effects 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 2
- 238000012216 screening Methods 0.000 abstract description 3
- 238000001514 detection method Methods 0.000 description 14
- 230000033001 locomotion Effects 0.000 description 8
- 238000012545 processing Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 7
- 230000000007 visual effect Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000005286 illumination Methods 0.000 description 3
- 230000010365 information processing Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000004807 localization Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000452 restraining effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The present invention relates to a kind of video object extract real-time optimization method, the optimization method includes: the interesting target region in traversal search current image frame;Obtain the external regular figure of minimum in the interesting target region;The target area that the area of the external regular figure of minimum is less than preset first threshold value is determined as noise and is deleted, the target area by denoising is obtained;If occurring target area in the current image frame all to lack, the first threshold is corrected, and re-executes noise screening and deletes.
Description
Technical field
The present invention relates to technical field of video processing, in particular to a kind of video object extract real-time optimization method and it is
System.
Background technique
With the development of science and technology the development of smart city is also maked rapid progress, artificial intelligence technology has been applied to various rows
Among industry, for example, in field of video processing, computer vision technique is used for image procossing and target analysis etc. and is had become to work as
Modern development trend is especially used for the target identification technology of video monitoring system, should can be in complex environment
Also it is capable of the operation of accurate stable, guarantees its real-time again.
In the prior art, mainly there are optical flow method, frame difference method and Background difference etc. for the method for target identification in video,
Wherein, optical flow method is the optical flow field by calculating picture frame, will be a movement mesh with the region recognition of identical light stream vectors
Mark, this method calculation amount is larger and real-time is poor, and light stream field distribution is also easier to be effected by environmental factors;Frame is poor
Method is the motion target area that current image frame is calculated using two frame differences in video flowing frame sequence, and this method is common
One of moving object detection and dividing method, but due to that cannot detect all pixels point in motion target area, it can only
Detection suitable for simple motion;Background difference be it is a kind of utilize the difference of current image frame and background image frame obtain prospect mesh
Calibration method, this method are also common object detection method, but environment more complicated for background, can increase target inspection
The difficulty surveyed and extracted.
Therefore, the video object extract real-time that a kind of real-time is good, identifies accurately and is suitable for complex environment is needed at present
Method and system.
Summary of the invention
The present invention provides a kind of video object extract real-time optimization method, comprising:
Interesting target region in step 1) traversal search current image frame;
Step 2) obtains the external regular figure of minimum in the interesting target region;
The target area that the area of the external regular figure of minimum is less than preset first threshold value is judged to making an uproar by step 3)
It puts and deletes, obtain the target area by denoising;
If occurring target area in the step 4) current image frame all to lack, the first threshold is corrected, is laid equal stress on
Newly execute the step 3).
Preferably, the step 4) further comprises:
The current image frame is compared step 41) with adjacent image frame, analyze in the current image frame whether
In the presence of missing target area;
Step 42) the missing target area if it exists, identifies and records the missing target area in the neighbor map
As the location information and size in frame;
The location information and size that step 43) is obtained using the step 42) adjust the first threshold, and again
The step 3) is executed, the mesh for meeting the size of the principle of optimality is obtained at the position in the current image frame
Mark region.
Preferably, the step 41) further comprises: comparing the mesh in the current image frame and the adjacent image frame
Region quantity is marked, if the target area number of the current image frame is less than the target area number in the adjacent image frame,
Then determine there is missing target area in the current image frame.
Preferably, the step 43) further comprises:
Step 431) reduces the first threshold, re-executes the step 3), and obtain the target area by denoising;
Step 432) identifies the target area in the current image frame at the location information that the step 42) obtains
To increase target area newly, and will the newly-increased target area area with the area that the step 42) obtains compared with to simultaneously further
Adjust the first threshold.
Preferably, the step 432) further comprises:
If the area of the newly-increased target area is greater than the area that the step 42) obtains, illustrate the step 431)
The threshold value th reset1It is less than normal, it needs to increase;
If the area of the newly-increased target area is less than the area or the newly-increased target area that the step 42) obtains
The area in domain is zero, then illustrates the threshold value th that step 431) is reset1It is bigger than normal, it needs to reduce again.
Preferably, the optimization method further comprises:
Step 5) judges whether the target area in the current image frame partial loss situation occurs, if occurring,
Then the part of the loss is restored, is specifically included:
If the area of the external regular figure of minimum in the presently described picture frame of step 51), with the adjacent image
The area ratio of the corresponding external regular figure of minimum is less than preset third threshold value in frame, then determines the external rule of minimum
Then there is mutation in the corresponding interesting target region of figure;
Step 52) compare all external regular figures of minimum in presently described picture frame and Record Comparison number with
It is mutated number;
If step 53) the mutation number accounts for the ratio of the comparison number not less than preset 4th threshold value, determine
There is partial loss in interesting target region in presently described picture frame;
Step 54) utilizes target area corresponding with the external regular figure of minimum of loss in the adjacent image frame
Location information is to presently described picture frame assignment.
Preferably, the optimization method further comprises:
Step 6) obtains the location information of the current all described external regular figure of minimum;
Step 7) judges whether the presently described external regular figure of minimum is Chong Die with other external regular figures of minimum;
If there is overlapping, the corresponding external regular figure of minimum is merged;
Step 8) judges unreasonable segmentation whether occur between the adjacent external regular figure of minimum;It, will if occurring
The corresponding external regular figure of minimum clusters again.
Preferably, the step 8) further comprises: by with the target position in the presently described external regular figure of minimum
Distance is less than the adjacent external regular figure of minimum of preset second threshold, is determined as unreasonable segmentation.
Preferably, the step 1) further comprises pre-processing to the current image frame.
According to another aspect of the present invention, a kind of video object extract real-time optimization system is also provided, including memory,
Processor and storage are on a memory and the computer program that can run on a processor, wherein described in the processor is run
Step as described above is executed when program.
Compared with the existing technology, the present invention achieves following advantageous effects: video object provided by the invention is real-time
Optimization method and system are extracted, the interesting target region that Background difference obtains is carried out using the visual attention location principle of HVS excellent
Change, the noise in interesting target region can not only be effectively removed and repeat to cluster, moreover it is possible to avoid unreasonable segmentation and target
The problems such as missing;Using optimization method provided by the invention and system, enable to the integrality of Objective extraction more preferable, accuracy
More preferably, while also it is able to satisfy real-time demand, the application environment especially suitable for background complexity.
Detailed description of the invention
Fig. 1 is video object extract real-time optimization method flow chart provided by the invention.
Fig. 2 is the method schematic diagram that solution target area provided by the invention is all lost.
Fig. 3 is the method schematic diagram provided by the invention for solving target area partial loss.
Specific embodiment
In order to which the purpose of the present invention, technical solution and advantage is more clearly understood, below in conjunction with attached drawing, to according to this
The video object extract real-time optimization method provided in the embodiment of invention is further described.
Human visual system (HVS) is a highly complex process, and the mechanism for being used for information processing commonly referred to as regards
Feel concern (Visual Attention), typically refers in numerous and complicated external environment, human vision can quickly determine
Position goes out important target area and carries out careful analysis, and other regions are only carried out with coarse analysis and is even ignored, this
Mechanism embodies the visual characteristic that HVS can actively select attentinal contents and be focused on.Inventor passes through numerous studies
Think, this information processing mechanism is used for field of video processing, can effectively promote the recognition capability to target.
Under normal conditions, the video processing procedures such as the classification, tracking for target in video and behavior understanding only relate to figure
As the motion target area in frame, therefore, for the subsequent processing of video, moving target is correctly detected from background
It is particularly significant with splitting, especially when background environment is complex, the difficulty of detection and the extraction of target but will be increased.
Inventor to existing video object recognition methods by analysing and comparing and combining feature (such as the field of video monitoring scene
Scape background is relatively fixed and the motion conditions of target area are relatively simple etc.) discovery, the calculating of Background difference it is relatively easy and
Detection effect is preferable, but for background complex environment, such as occur situations such as intense light irradiation, background is unsmooth in background, it is preceding
Scape target detection will appear error detection, repeat the problems such as detecting.
To solve the above-mentioned problems, inventor after study, it is real to propose a kind of video based on HVS information processing principle
When processing method, this method can for Background difference obtain area-of-interest be analyzed, and according to position therein believe
Breath judges the error section in area-of-interest and optimizes, to obtain the target area of more complete and accurate.
Video object extract real-time optimization method provided by the invention is for the video primarily determined by Background difference
The optimization that interesting target region in frame carries out.
If Bn(i, j) is background image, fn(i, j) is current frame image, and two field pictures are carried out difference processing, then are detected
Rule is as follows:
Wherein, T is detection threshold value, SnIt is determined as the region of background, M after (i, j) expression differencenAfter (i, j) expression difference really
It is set to the interesting target region of motion change.
Inventor has found by many experiments, when the background area of video is complex, obtains above by Background difference
The interesting target region obtained will appear the region of partial error detection;And between the interesting target region extracted also
The case where in the presence of segmentation is repeated;Meanwhile above-mentioned interesting target region will appear discontinuous, erroneous segmentation and unreasonable cluster
The problems such as, this will lead to target identification result, and there are a large amount of redundancies, increase the calculation amount of subsequent algorithm.
For this purpose, inventors herein proposing the method that the area-of-interest obtained for above-mentioned Background difference optimizes, below
It will be described in more detail.
In one embodiment of the invention, Fig. 1 is the target extract real-time optimization side that the preferred embodiment of the present invention provides
Method flow chart, as shown in Figure 1, this method specifically includes the following steps:
S10, minimum external regular figure is obtained
Interesting target region is obtained using Background difference, that is, judges that each pixel belongs to back in current image frame
Scape belongs to moving target, and the pixel for wherein belonging to moving target is clustered, the two-value in traversal search cluster result
Region, and the external regular figure (such as rectangle) of minimum of cluster result is obtained, meanwhile, obtain the position letter in corresponding region
Breath, the size including position coordinates and region.
Noise in S20, screening and delete target region
According to the visual attention location principle of HVS, human eye is limited the resolution capability of brightness, can only differentiate with certain bright
The target object of difference is spent, and luminance difference is smaller, can be considered consistent;Meanwhile the mankind obtain often through marginal information
The concrete shape of target object interprets target object etc., therefore only when article size, details light and shade and when presenting in eye
Between length it is all suitable when, could have more clear perception to object detail.
Based on the above principles, inventor thinks when occurring lesser region in object detection results, can be by the region
Area-of-interest caused by being determined as noise, rather than in image scene.Therefore when background environment complexity, if above-mentioned acquisition
Area-of-interest in occur as environment (such as the factors such as illumination, weather) influence caused by noise, not due to these noises
Belong to interesting target region, screened and is deleted.
Assuming that the width for the minimum circumscribed rectangle that step S10 is obtained and high respectively w and h, by making an uproar in analysis target area
Point feature, for example, due to noise be generally distributed it is at random, can by the lesser regional determination of rectangular area be noise region, setting
Threshold value th1, the location information in all target areas of acquisition is traversed, and judged according to following regular one:
W < th1;Or h < th1
If the minimum circumscribed rectangle region obtained meets above-mentioned regular one, it can determine that the region is noise in complex background
Caused noise, rather than the interesting target region in current image frame, can be deleted.
Using above-mentioned regular one, the area of error detection in the interesting target region of Background difference acquisition can be effectively removed
Domain, so that the recognition result obtained to Background difference optimizes.
S30, the whether reasonable simultaneously corrected threshold parameter th of denoising is examined1
When executing noise screening, if it is too small target area occur, it may appear that be mistakenly considered noise and be deleted, cause current
Target missing in picture frame needs the threshold value to above-mentioned setting to be corrected, to select in order to avoid there is above-mentioned mistake
Suitable threshold value th1。
When carrying out threshold correction, it is assumed that current image frame n, by current image frame and with its several adjacent picture frame
(for example, n-1, n-2, n-3, n-4) is compared respectively, and Fig. 2 is threshold correction method schematic diagram provided by the invention, such as Fig. 2
It is shown, if existing target area in four consecutive frames, and do not have in current image frame, then it is assumed that go out in current image frame
Show target all to lack;It then needs through adaptive mode to threshold value th1It is corrected, specific bearing calibration is as follows:
S301, by current image frame n and adjacent image frame n-1, n-2, n-3, n-4 is compared respectively, according to target area number
With the presence or absence of missing target in amount analysis current image frame, for example, if mesh present in adjacent image frame n-1, n-2, n-3, n-4
The target number that areal is greater than in current image frame is marked, then can determine that there are target missings in current image frame;
S302, the location information by comparing target area, find the target area of missing, and record in adjacent image frame
The size and location information of the target area;
The noise-removed threshold value th of S303, adjustment (timing is to reduce for the first time) current image frame1, for example, being arranged according to step-length
The amplitude adjusted every time, and re-execute the steps S20 and obtain the current image frame comprising multiple target areas through denoising;
The current image frame that S304, analytical procedure S303 are obtained, in the position for the desired target area that step S302 is obtained
The size of target area is compared at information;
If the target area area in current image frame is greater than the target area area in consecutive frame, illustrate step S303
The threshold value th reset1It is less than normal, it needs to increase, for example, according to the default amplitude for needing to increase of step-length;
If the target area area in current image frame is less than the target area area in consecutive frame or believes in the position
There is no above-mentioned target area (i.e. area is 0) at breath, then illustrate the threshold value th that step S303 is reset1It is bigger than normal, need into
One step reduces, for example, needing reduced amplitude according to step-length is default;
After completing above-mentioned adjusting thresholds, repeat step S303 and execute above-mentioned comparison again, until obtain meet it is excellent
The appropriate threshold th of change standard1。
In one embodiment of the invention, in order to reduce the extraneous factors such as illumination and noise to above-mentioned object detection results
Influence, execute above-mentioned steps S10 difference processing before, can for obtain background image and foreground image be located in advance
Reason operation, the garbage in image is reduced by the operation as far as possible, restores true information, and it is outer to reduce illumination, noise etc.
Influence of boundary's factor to object detection results.It simultaneously include that the bianry image for making to obtain after difference to present frame and background frames does form
Processing, eliminates the isolated point in bianry image, improves the accuracy of Objective extraction result.
In one embodiment of the invention, above-mentioned optimisation criteria can be pre-set error range, for example, different
It is determined as that the center position deviation of identical target area should meet certain numberical range in picture frame.
In one embodiment of the invention, when selecting the picture frame adjacent with current image frame, in addition to above-mentioned optional
N-1, n-2, n-3, the n-4 before current image frame timeline, if video data not acquire in real time, e.g. record
The video of system, it is also an option that the adjacent image frame after current image frame timeline, for example, n+1, n+2 etc..
In one embodiment of the invention, inventor has found through many experiments, is obtained using Background difference interested
Target area there is also the phenomenon that overlapping, and it is completely overlapped to be embodied in two target areas, that is, there is a region and include
The case where another region;Either two target areas partly overlap, that is, there is a region partly belongs to another area
The case where domain.For above-mentioned two situations, following optimizations can be executed to the interesting target region of acquisition.
Firstly, the positioning to interesting target region is realized using clustering algorithm, it, will by taking k-means clustering algorithm as an example
For the central point of each target area as point set, randomly selecting several target's center's points is initial cluster center μ1, μ2...,
μk∈Rn;
Secondly, calculating separately remaining all target area central points to the cluster centre to each above-mentioned cluster centre
Distance, and therefrom select the several samples nearest apart from the cluster centre as a kind of, specifically:
c(i)=argmin | | x(i)-μj||2
Wherein, x(i)For the center point set in current goal region.
Then, the center of the class of above-mentioned acquisition is recalculated:
Wherein, m represents the target numbers tentatively obtained, and 1 represents true value judgement.
Finally, repeating the above process until restraining, if can be by the interesting target region division obtained in current scene
Dry cluster, for example, L cluster, then L is the number in interesting target region in present image, while also can get mesh interested therein
Mark the position coordinates in region.
After completing the above-mentioned positioning to interesting target region, it is assumed that i-th of target coordinate position is xiAnd yi, wide and high
Respectively wiAnd hi;J-th of target coordinate position is xjAnd yj, wide and high respectively wjAnd hj。
The location information in above-mentioned two target area is traversed, and is judged according to following regular two:
If i-th of the target and j-th of goal satisfaction above-mentioned regular two obtained, can determine that target j is wrapped completely by target i
Contain, i.e., completely overlapped situation occurs in the interesting target region obtained using Background difference, then target j can be deleted
It removes.
Using above-mentioned regular two, area completely overlapped in the interesting target region of Background difference acquisition can be effectively removed
Domain, so that the recognition result obtained to Background difference optimizes.
In addition to this, when traversing the location information in above-mentioned two target area, can also according to following regular three into
Row judgement:
Or
If i-th of the target and j-th of goal satisfaction above-mentioned regular three obtained, can determine that region and the target i of target j
Region have a partial intersection, i.e., there is the case where partly overlapping between the interesting target region obtained using Background difference,
The location information in above-mentioned two target area then can be further analyzed, and is clustered or directly closes two regions again
And.
Using above-mentioned regular three, partly overlapping area in the interesting target region of Background difference acquisition can be effectively removed
Domain, so that the recognition result obtained to Background difference optimizes.
In one embodiment of the invention, occur in the interesting target region in order to solve to utilize Background difference acquisition
Unreasonable segmentation the phenomenon that, inventor it is found through experiment that, can by the characteristic of HVS visual attention location mechanism applied to target know
Not.
In general, the space of vision system and frequency characteristic are complementary, for moving image, there are it is a kind of when
Between resolution ratio and spatial resolution exchange.For example, be difficult to see its details clearly when moving object quickly passes through before human eye,
It can only see that rough profile, i.e. human eye " can track " resolution capability of moving object with human eye related.Therefore, according to
The movement of target should have a successional characteristic, inventor think can will to be regarded as occurring between adjacent image frame compared with
Big mutation.
According to this principle, following optimizations can be executed to the interesting target region obtained using Background difference.
After the location information for obtaining interesting target region using the localization method in above-described embodiment, it is assumed that i-th of mesh
Mark coordinate position is xiAnd yi, wide and high respectively wiAnd hi;J-th of target coordinate position is xjAnd yj, wide and high respectively wj
And hj.By the location information of erroneous segmentation target in analysis target area, threshold value th is set2, and carried out according to following regular four
Judgement:
xi+wi-xj≤th2;Or yi+hi-yj≤th2;
If i-th of the target and j-th of goal satisfaction above-mentioned regular four obtained, can determine that the region target j and mesh
The region i is marked by unreasonable segmentation, therefore, can analyze the location information in above-mentioned two region, to wherein unreasonable portion
Divide and re-starts cluster, the final reasonable segmentation for realizing area-of-interest.
Using above-mentioned regular four, the area of erroneous segmentation in the interesting target region of Background difference acquisition can be effectively removed
Domain, so that the recognition result obtained to Background difference optimizes.
In one embodiment of the invention, occur in the interesting target region in order to solve to utilize Background difference acquisition
Jump problem, inventor through experiments, it was found that, can by the characteristic of HVS visual attention location mechanism be applied to target identification.It is general next
It says, for human eye when observing picture frame, attention can concentrate the close quarters in scape of showing up, while whether the mankind can accurately identify
There are much relations in target area in scene out with whether human eye may be implemented tracking, in conjunction with the feature of video monitoring scene, In
The target area moved in the scene generally has continuity, i.e., the position of target area should be continuous between consecutive frame,
There is no biggish jumps.
According to above-mentioned principle, inventor has found through many experiments, and the jump that target area occurs is embodied in following two
Kind of situation: one to will lead to the variation of consecutive frame target area when being target area partial loss excessive;The other is target area
It will lead to target area between consecutive frame when losing completely to jump, that is, the interesting target region that detection obtains occur and do not connect
It passes through.To solve the above-mentioned problems, following methods are inventors herein proposed.
Firstly, judging that current image frame is using the relationship between the interesting target region area in adjacent image frame
It is no that there are target area jumps.
After the location information for obtaining interesting target region using the localization method in above-described embodiment, it is assumed that present image
I-th of target coordinate position in frame n is xiAnd yi, wide and high respectively wiAnd hi;Some image adjacent with current image frame
Corresponding i-th of target coordinate position is x in framei' and yi', wide and high respectively wi' and hi', according to object variations rule
Threshold value th is set3, and judged according to following regular five:
Or
If the above-mentioned rule of i-th of goal satisfaction in i-th of target and adjacent n frame in the current image frame obtained
Five, then it can determine that mutation occurs in i-th of target in present frame.
Fig. 3 is the method schematic diagram provided by the invention for solving target area partial loss, as shown in figure 3, according to above-mentioned
Rule five all compares corresponding target between targets all in present frame n and multiple consecutive frames, record mutation number p
With total comparison number m;Suitable threshold value th is arranged according to object variations rule simultaneously4, and judged according to following regular six:
p/m≥th4
If the comparing result of the current image frame and consecutive frame that obtain meets above-mentioned regular six, current image frame can determine that
In the case where there are target area excalations;
Secondly, after completing above-mentioned judgement, if it is determined that there is excalation in the target in current image frame as the result is shown,
It can use the correspondence target area location information obtained in the previous frame of current image frame to carry out the testing result of present frame
Assignment, to guarantee that the target area detected shows coherence and continuity.
Using above-mentioned regular five and rule six, it is possible to prevente effectively from occurring in the interesting target region that Background difference obtains
The case where excalation, so that the recognition result obtained to Background difference optimizes.
Although real to video object region provided by the invention in the above-described embodiments, using minimum circumscribed rectangle
When extract optimization method and be illustrated, but it will be recognized by one of ordinary skill in the art that above-mentioned optimization method can also be by making
It is realized with the external regular figure of other minimums, for example, minimum circumscribed circle shape etc..
Compared with the existing technology, video object extract real-time optimization method provided by the present invention is closed according to the vision of HVS
Principle is infused, is optimized for the interesting target region obtained using Background difference, above-mentioned target area can be effectively removed
In noise and repeat cluster;Simultaneously, moreover it is possible to the problems such as avoiding unreasonable segmentation and target from lacking.
Although the present invention has been described by means of preferred embodiments, the present invention is not limited to described here
Embodiment, without departing from the present invention further include made various changes and variation.
Claims (10)
1. a kind of video object extract real-time optimization method, the optimization method include:
Interesting target region in step 1) traversal search current image frame;
Step 2) obtains the external regular figure of minimum in the interesting target region;
The target area that the area of the external regular figure of minimum is less than preset first threshold value is determined as noise simultaneously by step 3)
It deletes, obtains the target area by denoising;
If occurring target area in the step 4) current image frame all to lack, the first threshold is corrected, and hold again
The row step 3).
2. optimization method according to claim 1, which is characterized in that the step 4) further comprises:
The current image frame is compared step 41) with adjacent image frame, and analyzing whether there is in the current image frame
Lack target area;
Step 42) the missing target area if it exists, identifies and records the missing target area in the adjacent image frame
In location information and size;
The location information and size that step 43) is obtained using the step 42) adjust the first threshold, and re-execute
The step 3) obtains the target area for meeting the size of the principle of optimality at the position in the current image frame
Domain.
3. optimization method according to claim 2, which is characterized in that the step 41) further comprises: working as described in comparison
Target area quantity in preceding picture frame and the adjacent image frame, if the target area number of the current image frame is less than institute
The target area number in adjacent image frame is stated, then determines there is missing target area in the current image frame.
4. optimization method according to claim 2, which is characterized in that the step 43) further comprises:
Step 431) reduces the first threshold, re-executes the step 3), and obtain the target area by denoising;
Target area in the current image frame at the location information that the step 42) obtains is identified as newly by step 432)
Increase target area, and by the newly-increased target area area compared with the area that the step 42) obtains to a successive step of going forward side by side
The first threshold.
5. optimization method according to claim 4, which is characterized in that the step 432) further comprises:
If the area of the newly-increased target area is greater than the area that the step 42) obtains, illustrate the step 431) again
The threshold value th being arranged1It is less than normal, it needs to increase;
If the area of the newly-increased target area is less than the area or the newly-increased target area that the step 42) obtains
Area is zero, then illustrates the threshold value th that step 431) is reset1It is bigger than normal, it needs to reduce again.
6. optimization method according to claim 1, which is characterized in that the optimization method further comprises:
Step 5) judges whether the target area in the current image frame partial loss situation occurs, will if occurring
The part of the loss is restored, and is specifically included:
If the area of the external regular figure of minimum in presently described picture frame, with corresponding institute in the adjacent image frame
The area ratio for stating minimum external regular figure is less than preset third threshold value, then determines that the external regular figure of minimum is corresponding
There is mutation in the interesting target region;
Compare all external regular figures of minimum in presently described picture frame and Record Comparison number and mutation number;
If the mutation number accounts for the ratio of the comparison number not less than preset 4th threshold value, presently described image is determined
There is partial loss in interesting target region in frame;
Utilize target area location information pair corresponding with the external regular figure of minimum of loss in the adjacent image frame
Presently described picture frame assignment.
7. optimization method according to claim 6, which is characterized in that the optimization method further comprises:
Step 6) obtains the location information of the current all described external regular figure of minimum;
Step 7) judges whether the presently described external regular figure of minimum is Chong Die with other external regular figures of minimum;If having
Overlapping then merges the corresponding external regular figure of minimum;
Step 8) judges unreasonable segmentation whether occur between the adjacent external regular figure of minimum;It, will be opposite if occurring
The external regular figure of minimum answered clusters again.
8. optimization method according to claim 7, which is characterized in that the step 8) further comprises: will be with current institute
State the adjacent external rule schema of minimum that the target position distance in minimum external regular figure is less than preset second threshold
Shape is determined as unreasonable segmentation.
9. optimization method according to claim 1, which is characterized in that the step 1) further comprises to the current figure
As frame is pre-processed.
10. a kind of video object extract real-time optimization system, including memory, processor and storage on a memory and can located
The computer program run on reason device, wherein execute when the processor operation described program such as any one of claim 1 to 9
The step.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810413067.6A CN110443097A (en) | 2018-05-03 | 2018-05-03 | A kind of video object extract real-time optimization method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810413067.6A CN110443097A (en) | 2018-05-03 | 2018-05-03 | A kind of video object extract real-time optimization method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110443097A true CN110443097A (en) | 2019-11-12 |
Family
ID=68427971
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810413067.6A Pending CN110443097A (en) | 2018-05-03 | 2018-05-03 | A kind of video object extract real-time optimization method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110443097A (en) |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101783076A (en) * | 2010-02-04 | 2010-07-21 | 西安理工大学 | Method for quick vehicle type recognition under video monitoring mode |
CN102289672A (en) * | 2011-06-03 | 2011-12-21 | 天津大学 | Infrared gait identification method adopting double-channel feature fusion |
CN102509086A (en) * | 2011-11-22 | 2012-06-20 | 西安理工大学 | Pedestrian object detection method based on object posture projection and multi-features fusion |
CN105868708A (en) * | 2016-03-28 | 2016-08-17 | 锐捷网络股份有限公司 | Image object identifying method and apparatus |
CN106339690A (en) * | 2016-08-30 | 2017-01-18 | 上海交通大学 | Video object flow detecting method and system based on noise elimination and auxiliary determination line |
US20170083753A1 (en) * | 2015-09-22 | 2017-03-23 | ImageSleuth, Inc. | Automated methods and systems for identifying and characterizing face tracks in video |
US20170154247A1 (en) * | 2015-11-30 | 2017-06-01 | Sony Interactive Entertainment Inc. | Information processing device, information processing method, light-emitting device regulating apparatus, and drive current regulating method |
CN107301388A (en) * | 2017-06-16 | 2017-10-27 | 重庆交通大学 | A kind of automatic vehicle identification method and device |
CN107330465A (en) * | 2017-06-30 | 2017-11-07 | 清华大学深圳研究生院 | A kind of images steganalysis method and device |
CN107480585A (en) * | 2017-07-06 | 2017-12-15 | 西安电子科技大学 | Object detection method based on DPM algorithms |
-
2018
- 2018-05-03 CN CN201810413067.6A patent/CN110443097A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101783076A (en) * | 2010-02-04 | 2010-07-21 | 西安理工大学 | Method for quick vehicle type recognition under video monitoring mode |
CN102289672A (en) * | 2011-06-03 | 2011-12-21 | 天津大学 | Infrared gait identification method adopting double-channel feature fusion |
CN102509086A (en) * | 2011-11-22 | 2012-06-20 | 西安理工大学 | Pedestrian object detection method based on object posture projection and multi-features fusion |
US20170083753A1 (en) * | 2015-09-22 | 2017-03-23 | ImageSleuth, Inc. | Automated methods and systems for identifying and characterizing face tracks in video |
US20170154247A1 (en) * | 2015-11-30 | 2017-06-01 | Sony Interactive Entertainment Inc. | Information processing device, information processing method, light-emitting device regulating apparatus, and drive current regulating method |
CN105868708A (en) * | 2016-03-28 | 2016-08-17 | 锐捷网络股份有限公司 | Image object identifying method and apparatus |
CN106339690A (en) * | 2016-08-30 | 2017-01-18 | 上海交通大学 | Video object flow detecting method and system based on noise elimination and auxiliary determination line |
CN107301388A (en) * | 2017-06-16 | 2017-10-27 | 重庆交通大学 | A kind of automatic vehicle identification method and device |
CN107330465A (en) * | 2017-06-30 | 2017-11-07 | 清华大学深圳研究生院 | A kind of images steganalysis method and device |
CN107480585A (en) * | 2017-07-06 | 2017-12-15 | 西安电子科技大学 | Object detection method based on DPM algorithms |
Non-Patent Citations (4)
Title |
---|
吴思等: "动态场景视频序列中的前景区域自动提取", 《计算机辅助设计与图形学学报》 * |
徐佩等: "视频中提取车辆的最小外接矩形框的改进算法", 《网友世界》 * |
王琼等: "基于监控视频的运动车辆检测优化方法", 《计算机工程与设计》 * |
赵潇等: ""基于HVS的视频监控目标提取方法研究"", 《计算机应用研究》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108985186B (en) | Improved YOLOv 2-based method for detecting pedestrians in unmanned driving | |
CN109145708B (en) | Pedestrian flow statistical method based on RGB and D information fusion | |
US10696938B2 (en) | Method and system for automated microbial colony counting from streaked sample on plated media | |
CN112926410B (en) | Target tracking method, device, storage medium and intelligent video system | |
US9075026B2 (en) | Defect inspection device and defect inspection method | |
CN113592845A (en) | Defect detection method and device for battery coating and storage medium | |
CN108810620A (en) | Identify method, computer equipment and the storage medium of the material time point in video | |
KR100612858B1 (en) | Method and apparatus for tracking human using robot | |
CN107644418B (en) | Optic disk detection method and system based on convolutional neural networks | |
CN101782526B (en) | Method and device for automatically restoring, measuring and classifying steel dimple images | |
CN114241548A (en) | Small target detection algorithm based on improved YOLOv5 | |
CN107230203A (en) | Casting defect recognition methods based on human eye vision attention mechanism | |
CN111383244B (en) | Target detection tracking method | |
CN101094413A (en) | Real time movement detection method in use for video monitoring | |
CN106447701A (en) | Methods and devices for image similarity determining, object detecting and object tracking | |
CN102346854A (en) | Method and device for carrying out detection on foreground objects | |
CN113763424B (en) | Real-time intelligent target detection method and system based on embedded platform | |
CN111161313A (en) | Multi-target tracking method and device in video stream | |
CN111199556A (en) | Indoor pedestrian detection and tracking method based on camera | |
CN113066064A (en) | Cone beam CT image biological structure identification and three-dimensional reconstruction system based on artificial intelligence | |
CN110647836A (en) | Robust single-target tracking method based on deep learning | |
CN110197494A (en) | A kind of pantograph contact point real time detection algorithm based on monocular infrared image | |
CN116665095B (en) | Method and system for detecting motion ship, storage medium and electronic equipment | |
CN109492647A (en) | A kind of power grid robot barrier object recognition methods | |
Zhang Sr et al. | A ship target tracking algorithm based on deep learning and multiple features |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191112 |
|
WD01 | Invention patent application deemed withdrawn after publication |