CN108345885A - A kind of method and device of target occlusion detection - Google Patents
A kind of method and device of target occlusion detection Download PDFInfo
- Publication number
- CN108345885A CN108345885A CN201810047872.1A CN201810047872A CN108345885A CN 108345885 A CN108345885 A CN 108345885A CN 201810047872 A CN201810047872 A CN 201810047872A CN 108345885 A CN108345885 A CN 108345885A
- Authority
- CN
- China
- Prior art keywords
- target
- detected
- moment
- response diagram
- diagram matrix
- 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
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
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
- G06V10/32—Normalisation of the pattern dimensions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present invention provides a kind of method and device of target occlusion detection, and to solve existing target following technology due to lacking reliable target occlusion detection method, there are the relatively low technical problems of target following technology accuracy rate.Method includes L response diagram matrix of the determining target to be detected in the forward direction time span L of preset time period, based on empirical equation and L response diagram matrix, determine that the degree of stability value of L response diagram matrix within a preset period of time, empirical equation are used to determine the degree of stability of the corresponding L response diagram of L response diagram matrix;If degree of stability value is less than predetermined threshold value, it is determined that target to be detected is blocked, and predetermined threshold value is related to target to be detected.
Description
Technical field
The present invention relates to the method and devices that technical field of image processing more particularly to a kind of target occlusion detect.
Background technology
Target occlusion is always that video sequence image handles one of the problem faced.Intelligent transportation class product, such as electricity are alert, block
Mouthful etc., it needs to be detected tracking and attributive analysis to targets such as pedestrian, vehicles, tracking is combined that can be excluded with detection algorithm
It is a little that there are the non-target for triggering identification function for the first time in monitored picture, the loads of reduction attributive analysis module, due to person to person, vehicle
With vehicle, people is mutually blocked with vehicle etc., and prodigious difficulty is caused to the continuous analysis of video sequence.In order to make tracking system exist
Target remains to reliably track target in the case of being blocked, it is necessary to have reliable shadowing standard, be blocked with basis
Severity dynamic select determines follow-up mechanism.
It is typically that the model obtained calculating the learn according to the 0-t periods in the existing target following technology based on correlation filtering
The response diagram at t+1 moment, then the location determination institute tracking position of object where the maximum value in figure according to response.And due to mould
Plate can be all updated per frame, only by the data training pattern of 0-t periods, interval Ns frames then be taken to update a mould
The mode of type is difficult to ensure the reliability of sample, is easy so that target is drifted about when blocking.Occur within nearly 2 years by
The method that representative sample joint training model and multiple features mutually merge is chosen in the 0-t periods to improve model
Accuracy, but still suffer from sample comprising background due to lacking reliable target occlusion judgment module and be inserted into model, and
And the mode of Fusion Features is also to sacrifice tracking velocity as cost, as engineering field real time handling requirement needs reach processing
Time is less than or equal to 2ms so that the accuracy rate of target following technology is relatively low.
In summary, existing target following technology is due to lacking reliable target occlusion detection method, there are target with
The relatively low technical problem of track technology accuracy rate.
Invention content
The embodiment of the present invention provides a kind of method and device of target occlusion detection, to solve existing target following skill
Art is due to lacking reliable target occlusion detection method, and there are the relatively low technical problems of target following technology accuracy rate.
In a first aspect, the embodiment of the present invention provides a kind of method of target occlusion detection, including:
Determine L response diagram matrix of the target to be detected in the forward direction time span L of preset time period, wherein described
Preset time period is the corresponding period at the first moment to T moment, and first moment is initial time, the forward direction time
A moment in length L corresponds to a response diagram matrix, and the response diagram matrix at each moment includes N number of pixel and each picture
The corresponding response of vegetarian refreshments, T, N are the integer more than or equal to 1, and L is the positive integer less than or equal to T;
Based on empirical equation and the L response diagram matrix, determine the L response diagram matrix in the preset time period
Interior degree of stability value, the empirical equation are used to determine the stabilization journey of the corresponding L response diagram of the L response diagram matrix
Degree;
If the degree of stability value is less than predetermined threshold value, it is determined that the target to be detected is blocked, the predetermined threshold value
It is related to the target to be detected.
In one possible implementation, the empirical equation is specially:
Wherein, RES indicates the degree of stability value, Fmax|tIndicate maximum response in t-th of response diagram matrix, Fmin|t
Indicate minimum response value in t-th of response diagram matrix, Fw,h|tIndicate that any response in t-th of response diagram matrix, t take successively
The integer of T-L to T-1.
In one possible implementation, determining target to be detected in the forward direction time span L of preset time period
L response diagram matrix before, further include:
Based on the original image at first moment, target area of the target to be detected in the original image is determined
Domain, and initialize the target component to be detected of the target to be detected, wherein the target component to be detected includes region of interest
Domain sizes, be used to indicate the target to be detected renewal speed learning rate and expand target area ratio at least one
;
The target area is adjusted based on the target component to be detected after initialization, it is interested after being adjusted
Region;
The area-of-interest is handled based on the target tracking algorism of correlation filtering, obtains initialization model, institute
State response diagram matrix of the initialization model for determining the target to be detected.
In one possible implementation, forward direction time span L of the determination target to be detected in preset time period
L interior response diagram matrix, including:
Based on the sound at the initialization model each moment that determines the target to be detected in the first moment to T moment
It should figure matrix;
Determine that the L moment corresponding response diagram matrix before the T moment is institute using the T moment as starting point
State L response diagram matrix of the target to be detected in the forward direction time span L.
In one possible implementation, if the degree of stability value is more than the predetermined threshold value, it is determined that described to wait for
Detection target is not blocked, and the method further includes:
Based on the original image at the second moment after first moment, second moment corresponding first mould is determined
Type, first model are used to determine the response diagram matrix of the target to be detected, first model and the initialization mould
Type is different;
The initialization model is updated based on first model.
Second aspect, the embodiment of the present invention provide a kind of detection device, and the detection device includes:
Determining module, for determining L response diagram square of the target to be detected in the forward direction time span L of preset time period
Battle array, wherein the preset time period is the corresponding period at the first moment to T moment, and first moment is initial time,
A moment in the forward direction time span L corresponds to a response diagram matrix, and the response diagram matrix at each moment includes N number of picture
Vegetarian refreshments and the corresponding response of each pixel, T, N are the integer more than or equal to 1, and L is the positive integer less than or equal to T;
Processing module determines that the L response diagram matrix exists for being based on empirical equation and the L response diagram matrix
Degree of stability value in the preset time period, the empirical equation is for determining the corresponding L sound of the L response diagram matrix
Should figure degree of stability;
Judgment module, if being less than predetermined threshold value for the degree of stability value, it is determined that the target to be detected is blocked,
The predetermined threshold value is related to the target to be detected.
In one possible implementation, the empirical equation is specially:
Wherein, RES indicates the degree of stability value, Fmax|tIndicate maximum response in t-th of response diagram matrix, Fmin|t
Indicate minimum response value in t-th of response diagram matrix, Fw,h|tIndicate that any response in t-th of response diagram matrix, t take successively
The integer of T-L to T-1.
In one possible implementation, the determining module is additionally operable to:
Before determining L response diagram matrix of the target to be detected in the forward direction time span L of preset time period, it is based on
The original image at first moment determines target area of the target to be detected in the original image, and initialization
The target component to be detected of the target to be detected, wherein the target component to be detected includes region of interest domain sizes, is used for
It indicates the learning rate of the renewal speed of the target to be detected and expands at least one in the ratio of target area;
The target area is adjusted based on the target component to be detected after initialization, it is interested after being adjusted
Region;
The area-of-interest is handled based on the target tracking algorism of correlation filtering, obtains initialization model, institute
State response diagram matrix of the initialization model for determining the target to be detected.
In one possible implementation, the determining module is used for:
Based on the sound at the initialization model each moment that determines the target to be detected in the first moment to T moment
It should figure matrix;
Determine that the L moment corresponding response diagram matrix before the T moment is institute using the T moment as starting point
State L response diagram matrix of the target to be detected in the forward direction time span L.
In one possible implementation, the judgment module is used for:
If the degree of stability value is more than the predetermined threshold value, it is determined that the target to be detected is not blocked;
Based on the original image at the second moment after first moment, second moment corresponding first mould is determined
Type, first model are used to determine the response diagram matrix of the target to be detected, first model and the initialization mould
Type is different;
The initialization model is updated based on first model.
The third aspect, the embodiment of the present invention provide a kind of computer installation, and the computer installation includes:
At least one processor, and
The memory that is connect at least one processor communication, communication interface;
Wherein, the memory is stored with the instruction that can be executed by least one processor, at least one place
The instruction that device is stored by executing the memory is managed, the method for communication interface execution as described in relation to the first aspect is utilized.
Fourth aspect, the embodiment of the present invention provide a kind of computer readable storage medium, the computer-readable storage medium
Matter is stored with computer instruction, when the computer instruction is run on computers so that computer executes such as first aspect
The method.
It being capable of the rule of thumb forward direction time span L of formula and target to be detected in preset time period in the embodiment of the present invention
L interior response diagram matrix determines the degree of stability value of L response diagram matrix within a preset period of time, by degree of stability
Value carries out whether threshold decision is blocked with determination target to be detected, i.e., if degree of stability value is less than predetermined threshold value, really
Fixed target to be detected is blocked, and provides a kind of reliable target occlusion detection method for target following technology and then improves
The accuracy of target following technology.
Description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, will make below to required in the embodiment of the present invention
Attached drawing is briefly described, it should be apparent that, attached drawing described below is only some embodiments of the present invention, for
For those of ordinary skill in the art, without creative efforts, other are can also be obtained according to these attached drawings
Attached drawing.
Fig. 1 is a kind of flow diagram of the method for target occlusion detection provided in the embodiment of the present invention;
Fig. 2 is a kind of module diagram of the detection device provided in the embodiment of the present invention;
Fig. 3 is the schematic diagram of Computer device of the embodiment of the present invention.
Specific implementation mode
In order to keep the purpose, technical scheme and advantage of the embodiment of the present invention clearer, implement below in conjunction with the present invention
Attached drawing in example, technical scheme in the embodiment of the invention is clearly and completely described.
It illustrates below and the application scenarios in the embodiment of the present invention is simply introduced, so as to those skilled in the art's
Understand.
Such as intelligent-tracking ball machine has the function of far distance controlled from motion tracking, after finding target to be detected, video camera
Program can be stopped " going on patrol " and target to be detected analyze while zoom amplifies object region to be detected, automatic identification
Objective attribute target attribute to be detected and state in visual range, and to target to be detected into line trace, while controlling holder and triggering is followed to ring
The target to be detected movement answered, keeps target to be detected always situated in monitored picture central area, monitoring image, which is able to record, to be waited for
Detect the overall process of target movement.And the target to be detected moved in this process is highly susceptible to flowering shrubs, trees, other pedestrians
With blocking for vehicle etc., the performance of intelligent-tracking ball machine is substantially affected.
For another example, face tracking product is usually applied to scene of the stream of people than comparatively dense, at the same face also belong to it is typical non-
Rigidity target, deformation and occlusion issue are quite serious, and acid test is proposed to capturing to complete effectively face.
According to the video analysis under multiple business scene as a result, target to be detected may be hidden by the stationary object in background
Gear, it is also possible to mutually blocking between multiple targets to be detected, or caused by target to be detected itself rotational deformation
From blocking;The degree blocked also is not quite similar, and such as partial occlusion, seriously blocks and all blocks;The length of time is not blocked also not
Equally, have and block and blocked when long in short-term.
To sum up, the process that target to be detected is blocked can be roughly divided into three phases:One, target to be detected enters screening
Gear, information of target to be detected is gradually lost during this;Two, for target to be detected among blocking, this is to be detected in the process
The information of target keeps lost condition;Three, target to be detected leaves occlusion area, and information of target to be detected is gradual during this
Restore.It blocks the unstable of the information for causing target to be detected or even loses, therefore, one kind is provided in the embodiment of the present invention
The method of target occlusion detection, can accurately detect whether target to be detected is blocked, for subsequently to mesh to be detected
Target tracking provides safeguard.
The preferred embodiment of the invention is described in detail below in conjunction with the accompanying drawings.
Embodiment one
Refer to Fig. 1, the embodiment of the present invention provides a kind of method of target occlusion detection, and the realization process of this method can be with
It is described as follows:
S101:Determine L response diagram matrix of the target to be detected in the forward direction time span L of preset time period, wherein
Preset time period is the corresponding period at the first moment to T moment, and the first moment was initial time, in forward direction time span L
A moment correspond to a response diagram matrix, the response diagram matrix at each moment includes N number of pixel and each pixel pair
The response answered, T, N are the integer more than or equal to 1, and L is the positive integer less than or equal to T;
S102:Based on empirical equation and L response diagram matrix, L response diagram matrix within a preset period of time steady is determined
Determine degree value, empirical equation is used to determine the degree of stability of the corresponding L response diagram of L response diagram matrix;
S103:If degree of stability value be less than predetermined threshold value, it is determined that target to be detected is blocked, predetermined threshold value with it is to be detected
Target is related.
In the embodiment of the present invention, it can determine target to be detected in original image according to the original image at the first moment
Target area, and initialization target to be detected target component to be detected, target component to be detected may include interested
Area size, can serve to indicate that target to be detected renewal speed learning rate and expand target area ratio at least
One.
Wherein, the first moment can be initial time, such as T0The original image at moment, the first moment can be the first moment
Initial video frame original image;Such as pedestrian, vehicle target to be detected can be determined from original image, it then can be to wait for
It detects centered on target and its target area of the ratio-dependent target to be detected shared on the original image in original image
Domain, the target area can be made of regularly or irregularly regional frame.
For example, however, it is determined that target to be detected be pedestrian, then constitute target area regional frame can be ellipse,
The rule regional frame such as rectangle or square, or it can also be just with pedestrian in original graph to constitute the regional frame of target area
The irregular area frame etc. constituted as upper body's border line.
When determining target area of the target to be detected in original image, the items of target to be detected can also be joined
Number, such as region of interest domain sizes, i.e. template size template_sz;It can serve to indicate that the renewal speed of target to be detected
Learning rate lr;At least one target component to be detected expanded in target area ratio padding etc. is initialized, and each
The initialization value of parameter can be set according to target to be detected, or for ease of the tracking and detection of target to be detected,
For different targets to be detected, the initialization value of parameters can also be identical.
For example, when carrying out tracing detection to target A to be detected, the initialization value of template size could be provided as 96 ×
96, the initialization value of learning rate could be provided as 0.002, and the initialization value for expanding target area ratio could be provided as
1.25;When carrying out tracing detection to target B to be detected, mesh to be detected may be used in the initialization value of corresponding parameters
The initialization value of the corresponding parameters of A is marked, or can also be reset according to target B to be detected, what is specifically used
Kind mode, the embodiment of the present invention are not restricted.
Then target area can be adjusted according to the target component to be detected after initialization, the sense after being adjusted
Target area such as can be expanded 1.25 times according to the initialization value for expanding target area ratio, then will be enlarged by by interest region
Target area shaping after 1.25 times zooms to fixed dimension, i.e. template size 96 × 96, obtains area-of-interest, according to
Target component to be detected after initialization is adjusted target area, can improve the accurate of the detection to target to be detected
Property.
After obtaining area-of-interest, can according to the target tracking algorism of correlation filtering to area-of-interest at
Reason, obtains initialization model, which is determined for the response diagram matrix of target to be detected.According to correlation filtering
Target area-of-interest is handled according to algorithm, obtain initialization model process can be as follows, wherein initialization model
Including initialization model model_xf and initialization model model_alphaf.
For example, dimensional Gaussian label y, dimensional Gaussian mark can be calculated according to the target component to be detected after initialization
Label y can obtain Fourier label matrix yf after two dimensional discrete Fourier transform;Set area-of-interest as x, then
The 31 dimension histogram of gradients features of x are extracted, two-dimentional Fourier is after the Cosine Window of the same size of characteristic pattern dot product of each dimension
Transformation, obtains xf.In turn, model_xf points of the initialization model that can be used for more new feature is following two situations:
The first, is if the original image at the first moment is first frame, initialization template model_xf=xf;
Second, if not first frame, then model_xf=(1-lr) * model_xf+lr*xf, wherein lr is study
Rate.
And the process for obtaining initialization model model_alphaf can be as follows:
Xf with itself do linearly related operation, obtain kxf=sum (xf.*conj (xf), 3)/numel (xf), then into
Row quickly training, obtains alphaf=yf./(kxf+lambda), wherein and sum () indicates that summation, " .* " indicate dot product,
Congj () indicates that conjugation, numel (xf) indicate to calculate the pixel number of xf, and "/" indicates that point removes, and lambda is regularization coefficient.
And the initialization model model_alphaf for being used to quickly detect can also be divided into following two situations:
The first, is if the original image at the first moment is first frame, initialization model model_alphaf=
alphaf;
Second, if not first frame, then model_alphaf=(1-lr) * model_alphaf+lr*alphaf.
The specific calculating process of the target tracking algorism of correlation filtering may refer to the prior art, and the embodiment of the present invention is not made
It repeats.
After the target tracking algorism based on correlation filtering obtains initialization model, S101 can be entered, i.e. determination waits for
Detect L response diagram matrix of the target in the forward direction time span L of preset time period, wherein when preset time period is first
It carves to the corresponding period at T moment, the first moment was initial time, and a moment in forward direction time span L corresponds to one
The response diagram matrix of response diagram matrix, each moment includes N number of pixel and the corresponding response of each pixel, T, N are big
In the integer equal to 1, L is the positive integer less than or equal to T.
Forward direction time span L can be long using L moment corresponding time of the T moment as starting point, before the T moment
Degree, it is 5, T 10 such as to take L, and the first moment to the 10th moment is followed successively by T0、T1、T2、T3、……、T7、T8、T9, wherein when the 10th
Carve corresponding T9, then forward direction time span can correspond to T8、T7、T6、T5、T4。
In one possible implementation, target to be detected can be determined according to the initialization model of aforementioned acquisition
The response diagram matrix at each moment in one moment to T moment.
Below to determine T of the target to be detected in the first moment to T moment according to initialization model1The response at moment
For figure matrix.
Assuming that the first moment, i.e. initial time are T0, according to T0After the original image at moment obtains initialization model, really
T after fixed first moment1Moment video frame, i.e. T1Moment corresponding original image, can be according to the mesh to be detected after initialization
Parameter is marked, extracts target to be detected in T1In T in moment video frame, with target to be detected0Same position on the original image at moment
The area-of-interest of same size, is set as X1。
Then, X is extracted131 dimension histogram of gradients features, the Cosine Window of the same size of characteristic pattern dot product of each dimension
After do two-dimensional Fourier transform, obtain X1f;X1F and initialization model model_xf does dot product initialization mould after linearly related operation
Then type model_alphaf carries out two-dimensional inverse Fourier transform and real part is taken to obtain T1The corresponding response diagram square of response diagram at moment
Battle array.
The response diagram square at other moment of the target to be detected in the first moment to T moment is determined according to initialization model
Battle array and above-mentioned determining T1The process of the response diagram matrix at moment is similar, and the embodiment of the present invention does not repeat.
It is then possible to determine using the T moment as starting point, the L moment corresponding response diagram matrix before the T moment is to wait for
L response diagram matrix of the target in forward direction time span L is detected, the response diagram matrix at each moment includes N number of pixel, and
Each pixel, which corresponds to, has response, and has maximum response and minimum response value in the response diagram matrix at each moment.
For example, the first moment was T0, T takes 10, L to take 5, then being followed successively by T from the first moment to the 10th moment0、T1、T2、
T3、……、T7、T8、T9, wherein the 10th moment corresponded to T9.Each moment corresponds to corresponding response diagram matrix, here 5 response diagrams
Matrix can be T8、T7、T6、T5、T4Corresponding response diagram matrix.
After determining L response diagram matrix of target to be detected, S102 can be entered, that is, be based on empirical equation and L
Response diagram matrix determines the degree of stability value of L response diagram matrix within a preset period of time, and empirical equation is for determining L sound
Should the corresponding L response diagram of figure matrix degree of stability.
In one possible implementation, empirical equation is specifically as follows:
In formula (1), RES indicates degree of stability value, Fmax|tIndicate maximum response in t-th of response diagram matrix, Fmin|t
Indicate minimum response value in t-th of response diagram matrix, Fw,h|tIndicate that any response in t-th of response diagram matrix, t take successively
The integer of T-L to T-1.In practical applications, formula (1) can reflect the degree of stability of response diagram, and RES values are bigger, response diagram
It is more stable, and RES values are smaller, response diagram more shakes and target to be detected is blocked.
In practical applications, to when being blocked completely before target to be detected is blocked, RES can show downward trend,
Ascendant trend is will present corresponding to first derivative.When not blocking, due to the possible slight displacement state in target area,
RES is still within the state fluctuated up and down, and such case embodies particularly evident in the tracking of the non-rigid targets such as face.Cause
Then this does occlusion detection and undesirable, error is larger on this basis if directly calculating the RES at each moment.And this hair
In bright embodiment, may be used it is preceding acquire RES to the method for smothing filtering, i.e. experience acquires RES using formula (1), and herein
On the basis of carry out target occlusion detection, improve target occlusion detection accuracy.
After obtaining degree of stability value RES according to above-mentioned empirical equation (1), S103 can be entered, you can with to RES into
Row threshold decision can determine that target to be detected is blocked, wherein the setting of predetermined threshold value can if RES is less than predetermined threshold value
To be specially which kind of target is set with target to be detected.And target to be detected can be generally divided into but be not limited only to following two
Type:
The first, the non-rigid target to be detected of such as pedestrian, the target to be detected of this type is such as time, position
Apparent variation occurs in posture or in shape for variation;
Second, such as vehicle rigidity target to be detected, the target to be detected of this type generally will not be with time, position
Variation occur in shape significantly change.
Also, the RES that the RES predetermined threshold values of non-rigid target to be detected will be generally less than rigidity target to be detected presets threshold
Value.
If degree of stability value is more than predetermined threshold value, it can determine that target to be detected is not blocked, at this moment, can count again
Calculate the second moment corresponding first model after the first moment, and to can be used for updating the first moment corresponding just for the first model
Beginningization model, with calculate the T+1 moment and its later at the time of corresponding response diagram matrix.Wherein, the first model includes first
Model model_X1F and the first model model_alphaf1.
The process for obtaining the first model is referred to the aforementioned process for obtaining initialization model.
For example, setting for the second moment as T1, area-of-interest of the target to be detected on the original image at moment be set as X1,
Then X is extracted131 dimension histogram of gradients features, be two-dimentional Fu after the Cosine Window of the same size of characteristic pattern dot product of each dimension
In leaf transformation, obtain X1f.In turn, if first frame, then model_X1F=X1f;Otherwise, model_X1F=(1-lr) *
model_X1f+lr*X1F, wherein lr is learning rate.
The process for obtaining the first model model_alphaf1 can be as follows:
X1F with itself do linearly related operation, obtain kX1F=sum (X1f.*conj(X1f),3)/numel(X1F), then
It is quickly trained, obtains alphaf1=yf./(kX1F+lambda), if first frame, model_alphaf1=
alphaf1;Otherwise, model_alphaf1=(1-lr) * model_alphaf1+lr*alphaf1.
And then initialization model can be updated according to the first model, initialization model drift is greatly reduced in this way
The case where, while the newer number of initialization model is reduced, achieve the effect that accelerate.
In conclusion one or more technical solution of the embodiment of the present invention, at least have the following technical effect that or
Advantage:
It the first, being capable of the rule of thumb forward direction time of formula and target to be detected in preset time period in the embodiment of the present invention
L response diagram matrix in length L determines the degree of stability value of L response diagram matrix within a preset period of time, by stabilization
Degree value carries out whether threshold decision is blocked with determination target to be detected, i.e., if degree of stability value is less than predetermined threshold value,
It then determines that target to be detected is blocked, a kind of reliable target occlusion detection method, Jin Erti is provided for target following technology
The high accuracy of target following technology.
The second, due to that, when determining that target to be detected is not blocked, can be carried out to initialization model in the embodiment of the present invention
Update obtains the response diagram matrix for more accurately being used to determine target to be detected, further ensures target occlusion detection
Accuracy.
Embodiment two
Fig. 2 is referred to, same inventive concept is based on, the embodiment of the present invention provides a kind of detection device, the detection device
Including:
Determining module 21, for determining L response diagram of the target to be detected in the forward direction time span L of preset time period
Matrix, wherein the preset time period is the corresponding period at the first moment to T moment, when first moment is starting
It carves, a moment in the forward direction time span L corresponds to a response diagram matrix, and the response diagram matrix at each moment includes N
A pixel and the corresponding response of each pixel, T, N are the integer more than or equal to 1, and L is the positive integer less than or equal to T;
Processing module 22 determines the L response diagram matrix for being based on empirical equation and the L response diagram matrix
Degree of stability value in the preset time period, the empirical equation is for determining that the L response diagram matrix is L corresponding
The degree of stability of response diagram;
Judgment module 23, if being less than predetermined threshold value for the degree of stability value, it is determined that the target to be detected is hidden
Gear, the predetermined threshold value are related to the target to be detected.
In one possible implementation, the empirical equation is specially:
Wherein, RES indicates the degree of stability value, Fmax|tIndicate maximum response in t-th of response diagram matrix, Fmin|t
Indicate minimum response value in t-th of response diagram matrix, Fw,h|tIndicate that any response in t-th of response diagram matrix, t take successively
The integer of T-L to T-1.
In one possible implementation, the determining module 21 is additionally operable to:
Before determining L response diagram matrix of the target to be detected in the forward direction time span L of preset time period, it is based on
The original image at first moment determines target area of the target to be detected in the original image, and initialization
The target component to be detected of the target to be detected, wherein the target component to be detected includes region of interest domain sizes, is used for
It indicates the learning rate of the renewal speed of the target to be detected and expands at least one in the ratio of target area;
The target area is adjusted based on the target component to be detected after initialization, it is interested after being adjusted
Region;
The area-of-interest is handled based on the target tracking algorism of correlation filtering, obtains initialization model, institute
State response diagram matrix of the initialization model for determining the target to be detected.
In one possible implementation, the determining module 21 is used for:
Based on the sound at the initialization model each moment that determines the target to be detected in the first moment to T moment
It should figure matrix;
Determine that the L moment corresponding response diagram matrix before the T moment is institute using the T moment as starting point
State L response diagram matrix of the target to be detected in the forward direction time span L.
In one possible implementation, the judgment module 23 is used for:
If the degree of stability value is more than the predetermined threshold value, it is determined that the target to be detected is not blocked;
Based on the original image at the second moment after first moment, second moment corresponding first mould is determined
Type, first model are used to determine the response diagram matrix of the target to be detected, first model and the initialization mould
Type is different;
The initialization model is updated based on first model.
Embodiment three
Fig. 3 is referred to, same inventive concept is based on, provides a kind of computer installation in the embodiment of the present invention, including at least
One processor 31, and memory 32 and communication interface 33 at least one processor 31 communication connection, in Fig. 3 with
For one processor 31 is shown.
Wherein, the memory 32 is stored with the instruction that can be executed by least one processor 31, and described at least one
The instruction that a processor 31 is stored by executing the memory 32, is executed using the communication interface 33 such as institute in embodiment one
The method stated.
Example IV
Based on same inventive concept, the embodiment of the present invention provides a kind of computer readable storage medium, and the computer can
It reads storage medium and is stored with computer instruction, when the computer instruction is run on computers so that computer executes such as
Method described in embodiment one.
In specific implementation process, computer readable storage medium includes:General serial bus USB
(Universal Serial Bus flash drive, USB), mobile hard disk, read-only memory (Read-Only Memory,
ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc. are various can store program
The storage medium of code.
The apparatus embodiments described above are merely exemplary, wherein the units/modules illustrated as separating component
It may or may not be physically separated, the component shown as units/modules may or may not be
Physical unit/module, you can be located at a place, or may be distributed in multiple network element/modules.It can basis
It is actual to need that some or all of module therein is selected to achieve the purpose of the solution of this embodiment.Ordinary skill people
Member is not in the case where paying performing creative labour, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It is realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be expressed in the form of software products in other words, should
Computer software product can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
Finally it should be noted that:The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
Present invention has been described in detail with reference to the aforementioned embodiments, it will be understood by those of ordinary skill in the art that:It still may be used
With technical scheme described in the above embodiments is modified or equivalent replacement of some of the technical features;
And these modifications or replacements, various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (12)
1. a kind of method of target occlusion detection, which is characterized in that including:
Determine L response diagram matrix of the target to be detected in the forward direction time span L of preset time period, wherein described default
Period is the corresponding period at the first moment to T moment, and first moment is initial time, the forward direction time span
A moment in L corresponds to a response diagram matrix, and the response diagram matrix at each moment includes N number of pixel and each pixel
Corresponding response, T, N are the integer more than or equal to 1, and L is the positive integer less than or equal to T;
Based on empirical equation and the L response diagram matrix, determine the L response diagram matrix in the preset time period
Degree of stability value, the empirical equation are used to determine the degree of stability of the corresponding L response diagram of the L response diagram matrix;
If the degree of stability value is less than predetermined threshold value, it is determined that the target to be detected is blocked, the predetermined threshold value and institute
It is related to state target to be detected.
2. the method as described in claim 1, which is characterized in that the empirical equation is specially:
Wherein, RES indicates the degree of stability value, Fmax|tIndicate maximum response in t-th of response diagram matrix, Fmin|tIndicate the
Minimum response value in t response diagram matrix, Fw,h|tIndicate that any response in t-th of response diagram matrix, t take T-L to arrive successively
The integer of T-1.
3. method as claimed in claim 2, which is characterized in that determining forward direction time of the target to be detected in preset time period
Before L response diagram matrix in length L, further include:
Based on the original image at first moment, target area of the target to be detected in the original image is determined,
And the target component to be detected of the initialization target to be detected, wherein the target component to be detected includes area-of-interest
Size, be used to indicate the target to be detected renewal speed learning rate and expand target area ratio at least one
;
The target area is adjusted based on the target component to be detected after initialization, the region of interest after being adjusted
Domain;
The area-of-interest is handled based on the target tracking algorism of correlation filtering, obtains initialization model, it is described first
Beginningization model is used to determine the response diagram matrix of the target to be detected.
4. method as claimed in claim 3, which is characterized in that the determination target to be detected is in the forward direction of preset time period
Between L response diagram matrix in length L, including:
Based on the response diagram at the initialization model each moment that determines the target to be detected in the first moment to T moment
Matrix;
It determines using the T moment as starting point, the L moment corresponding response diagram matrix before the T moment is described waits for
Detect L response diagram matrix of the target in the forward direction time span L.
5. method as described in claim 3 or 4, which is characterized in that if the degree of stability value is more than the predetermined threshold value,
Determine that the target to be detected is not blocked, the method further includes:
Based on the original image at the second moment after first moment, second moment corresponding first model is determined,
First model is used to determine the response diagram matrix of the target to be detected, and first model and the initialization model are not
Together;
The initialization model is updated based on first model.
6. a kind of detection device, which is characterized in that the detection device includes:
Determining module, for determining L response diagram matrix of the target to be detected in the forward direction time span L of preset time period,
Wherein, the preset time period is the corresponding period at the first moment to T moment, and first moment is initial time, institute
A moment before stating into time span L corresponds to a response diagram matrix, and the response diagram matrix at each moment includes N number of pixel
Point and the corresponding response of each pixel, T, N are the integer more than or equal to 1, and L is the positive integer less than or equal to T;
Processing module determines the L response diagram matrix described for being based on empirical equation and the L response diagram matrix
Degree of stability value in preset time period, the empirical equation is for determining the corresponding L response diagram of the L response diagram matrix
Degree of stability;
Judgment module, if being less than predetermined threshold value for the degree of stability value, it is determined that the target to be detected is blocked, described
Predetermined threshold value is related to the target to be detected.
7. detection device as claimed in claim 6, which is characterized in that the empirical equation is specially:
Wherein, RES indicates the degree of stability value, Fmax|tIndicate maximum response in t-th of response diagram matrix, Fmin|tIndicate the
Minimum response value in t response diagram matrix, Fw,h|tIndicate that any response in t-th of response diagram matrix, t take T-L to arrive successively
The integer of T-1.
8. detection device as claimed in claim 7, which is characterized in that the determining module is additionally operable to:
Before determining L response diagram matrix of the target to be detected in the forward direction time span L of preset time period, based on described
The original image at the first moment determines target area of the target to be detected in the original image, and described in initialization
The target component to be detected of target to be detected, wherein the target component to be detected includes region of interest domain sizes, is used to indicate
At least one of in the learning rate and expansion target area ratio of the renewal speed of the target to be detected;
The target area is adjusted based on the target component to be detected after initialization, the region of interest after being adjusted
Domain;
The area-of-interest is handled based on the target tracking algorism of correlation filtering, obtains initialization model, it is described first
Beginningization model is used to determine the response diagram matrix of the target to be detected.
9. detection device as claimed in claim 8, which is characterized in that the determining module is used for:
Based on the response diagram at the initialization model each moment that determines the target to be detected in the first moment to T moment
Matrix;
It determines using the T moment as starting point, the L moment corresponding response diagram matrix before the T moment is described waits for
Detect L response diagram matrix of the target in the forward direction time span L.
10. method as claimed in claim 8 or 9, which is characterized in that the judgment module is used for:
If the degree of stability value is more than the predetermined threshold value, it is determined that the target to be detected is not blocked;
Based on the original image at the second moment after first moment, second moment corresponding first model is determined,
First model is used to determine the response diagram matrix of the target to be detected, and first model and the initialization model are not
Together;
The initialization model is updated based on first model.
11. a kind of computer installation, the computer installation include:
At least one processor, and
The memory that is connect at least one processor communication, communication interface;
Wherein, the memory is stored with the instruction that can be executed by least one processor, at least one processor
By executing the instruction of the memory storage, executed as described in any one of claim 1-5 using the communication interface
Method.
12. a kind of computer readable storage medium, the computer-readable recording medium storage has computer instruction, when the meter
When the instruction of calculation machine is run on computers so that computer executes the method as described in any one of claim 1-5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810047872.1A CN108345885A (en) | 2018-01-18 | 2018-01-18 | A kind of method and device of target occlusion detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810047872.1A CN108345885A (en) | 2018-01-18 | 2018-01-18 | A kind of method and device of target occlusion detection |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108345885A true CN108345885A (en) | 2018-07-31 |
Family
ID=62961537
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810047872.1A Pending CN108345885A (en) | 2018-01-18 | 2018-01-18 | A kind of method and device of target occlusion detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108345885A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110290351A (en) * | 2019-06-26 | 2019-09-27 | 广东康云科技有限公司 | A kind of video target tracking method, system, device and storage medium |
CN111160242A (en) * | 2019-12-27 | 2020-05-15 | 上海眼控科技股份有限公司 | Image target detection method, system, electronic terminal and storage medium |
CN111604888A (en) * | 2020-05-29 | 2020-09-01 | 珠海格力电器股份有限公司 | Inspection robot control method, inspection system, storage medium and electronic device |
CN113963428A (en) * | 2021-12-23 | 2022-01-21 | 北京的卢深视科技有限公司 | Model training method, occlusion detection method, system, electronic device, and medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160174895A1 (en) * | 2008-02-29 | 2016-06-23 | The Regents Of The University Of Michigan | Systems and methods for imaging changes in tissue |
CN106570478A (en) * | 2016-11-04 | 2017-04-19 | 北京智能管家科技有限公司 | Object loss determine method and device in visual tracking |
CN107122741A (en) * | 2017-04-27 | 2017-09-01 | 中国科学院长春光学精密机械与物理研究所 | A kind of method for tracking target and system |
CN107154024A (en) * | 2017-05-19 | 2017-09-12 | 南京理工大学 | Dimension self-adaption method for tracking target based on depth characteristic core correlation filter |
CN107197199A (en) * | 2017-05-22 | 2017-09-22 | 哈尔滨工程大学 | A kind of intelligent monitoring and controlling device and method for tracking target |
-
2018
- 2018-01-18 CN CN201810047872.1A patent/CN108345885A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160174895A1 (en) * | 2008-02-29 | 2016-06-23 | The Regents Of The University Of Michigan | Systems and methods for imaging changes in tissue |
CN106570478A (en) * | 2016-11-04 | 2017-04-19 | 北京智能管家科技有限公司 | Object loss determine method and device in visual tracking |
CN107122741A (en) * | 2017-04-27 | 2017-09-01 | 中国科学院长春光学精密机械与物理研究所 | A kind of method for tracking target and system |
CN107154024A (en) * | 2017-05-19 | 2017-09-12 | 南京理工大学 | Dimension self-adaption method for tracking target based on depth characteristic core correlation filter |
CN107197199A (en) * | 2017-05-22 | 2017-09-22 | 哈尔滨工程大学 | A kind of intelligent monitoring and controlling device and method for tracking target |
Non-Patent Citations (1)
Title |
---|
王萌萌等: ""基于循环特征的大边际目标跟踪算法"", 《2017年IEEE计算机视觉与模式识别会议》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110290351A (en) * | 2019-06-26 | 2019-09-27 | 广东康云科技有限公司 | A kind of video target tracking method, system, device and storage medium |
CN110290351B (en) * | 2019-06-26 | 2021-03-23 | 广东康云科技有限公司 | Video target tracking method, system, device and storage medium |
CN111160242A (en) * | 2019-12-27 | 2020-05-15 | 上海眼控科技股份有限公司 | Image target detection method, system, electronic terminal and storage medium |
CN111604888A (en) * | 2020-05-29 | 2020-09-01 | 珠海格力电器股份有限公司 | Inspection robot control method, inspection system, storage medium and electronic device |
CN111604888B (en) * | 2020-05-29 | 2021-09-14 | 珠海格力电器股份有限公司 | Inspection robot control method, inspection system, storage medium and electronic device |
CN113963428A (en) * | 2021-12-23 | 2022-01-21 | 北京的卢深视科技有限公司 | Model training method, occlusion detection method, system, electronic device, and medium |
CN113963428B (en) * | 2021-12-23 | 2022-03-25 | 北京的卢深视科技有限公司 | Model training method, occlusion detection method, system, electronic device, and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Xu et al. | Learn to scale: Generating multipolar normalized density maps for crowd counting | |
CN108345885A (en) | A kind of method and device of target occlusion detection | |
CN110837856B (en) | Neural network training and target detection method, device, equipment and storage medium | |
EP2693300A2 (en) | Device and method for recognizing gesture based on direction of gesture | |
CN103093198B (en) | A kind of crowd density monitoring method and device | |
US9105101B2 (en) | Image tracking device and image tracking method thereof | |
CN110706247A (en) | Target tracking method, device and system | |
CN112036381B (en) | Visual tracking method, video monitoring method and terminal equipment | |
CN108776974A (en) | A kind of real-time modeling method method suitable for public transport scene | |
CN112949512B (en) | Dynamic gesture recognition method, gesture interaction method and interaction system | |
Lim et al. | River flow lane detection and Kalman filtering‐based B‐spline lane tracking | |
CN111627049A (en) | High-altitude parabola determination method and device, storage medium and processor | |
Ramirez-Alonso et al. | Temporal weighted learning model for background estimation with an automatic re-initialization stage and adaptive parameters update | |
EP3675041B1 (en) | Method and apparatus for determining road information, and vehicle | |
CN113168520A (en) | Method of tracking objects in a scene | |
CN105469054A (en) | Model construction method of normal behaviors and detection method of abnormal behaviors | |
Zhang et al. | New mixed adaptive detection algorithm for moving target with big data | |
WO2023016182A1 (en) | Pose determination method and apparatus, electronic device, and readable storage medium | |
CN108876824B (en) | Target tracking method, device and system and dome camera | |
CN112489077A (en) | Target tracking method and device and computer system | |
CN108932465B (en) | Method and device for reducing false detection rate of face detection and electronic equipment | |
Xu et al. | Visual tracking of partially observable targets with suboptimal filtering | |
Baba et al. | Stray dogs behavior detection in urban area video surveillance streams | |
KR20150081797A (en) | Apparatus and method for tracking object | |
CN116012421A (en) | Target tracking method and device |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180731 |