CN109636828A - Object tracking methods and device based on video image - Google Patents
Object tracking methods and device based on video image Download PDFInfo
- Publication number
- CN109636828A CN109636828A CN201811386399.6A CN201811386399A CN109636828A CN 109636828 A CN109636828 A CN 109636828A CN 201811386399 A CN201811386399 A CN 201811386399A CN 109636828 A CN109636828 A CN 109636828A
- Authority
- CN
- China
- Prior art keywords
- area
- target following
- video image
- following object
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
Abstract
The present invention provides a kind of object tracking methods and device based on video image, when being tracked to the target following object in continuous video image, when position of the target following object in video image has been determined, only by target following object after the second area that the position in the first video image determines in the second video image in subsequent calculating, position of the target following object in the second video image is determined according to the position of target following object in the second area.Therefore it does not need during subsequent object tracking, identification calculating is carried out to entire video image, and then reduce calculation amount when tracking for object, improve computational efficiency.
Description
Technical field
The present invention relates to technical field of intelligence more particularly to a kind of object tracking methods and device based on video image.
Background technique
Object tracking is essential technology in the application such as monitoring security protection in technical field of intelligence.In general, object
Its performance of volume tracing is to the most important of most subsequent development.For example, needing to track user in the application shops such as unmanned supermarket
Track analyzed with the behavior to user, and pedestrian positioning no doubt be unable to do without object tracking.Therefore one kind is just needed
Efficiently, accurately object tracking methods that can monitor in real time to the position of object, active orientation generate structuring
Data are supplied to high-rise analysis decision.
In the prior art, object tracking technology is usually using video image as target, for example, by Image Acquisition such as cameras
After equipment obtains continuous video image, the position of target following object in each frame video image is analyzed frame by frame, and will own
Continuously get up to obtain the motion track of target following object in the position of continuous target in video image tracking object.
But the prior art is used, when frame by frame to video image analysis, require using video image integrally as identification
Object, and position of the target following object in every frame video image is determined respectively.And target following object is generally in video figure
As in exist only in the lesser region in part, and then result in identification target following object when can to entire video image carry out compared with
More invalid computations, calculation amount when causing object tracking is larger, computational efficiency is lower.
Summary of the invention
The present invention provides a kind of object tracking methods and device based on video image, reduces and object is tracked
When calculation amount, improve computational efficiency.
First aspect present invention provides a kind of object tracking methods based on video image, comprising:
According to position of the target following object in the first video image, the target following object corresponding first is determined
Region；
The corresponding second area of the target following object in the second video image is determined according to the first area；
The area of the second area is greater than the area of the first area, and the center of the second area is in the second video figure
Position as in is identical as position of the center of the first area in first video image；
According to the position of the target following object in the second region, determine the target following object described
Position in second video image.
In one embodiment of first aspect present invention,
The position according to the target following object in the second region determines that the target following object exists
After position in second video image, further includes:
According to the target following object in the position in first video image and in second video image
Position, determine the motion track of the target following object.
In one embodiment of first aspect present invention,
The position according to the target following object in first video image and in the second video figure
Position as in, determines the motion track of the target following object, comprising:
Third region corresponding with the target following object is determined from the second area；The first area and institute
State the area equation in third region, relative position and the target following object of the target following object in the first area
Body is identical in the relative position in the third region；
According to position of the first area in first video image and the third region in second view
Position in frequency image determines the motion track of the target following object.
In one embodiment of first aspect present invention,
The position according to the target following object in first video image and in the second video figure
Position as in, before the motion track for determining the target following object, further includes:
Judge in the second area whether to include the target following object；
If so, according to position of the target following object in first video image and in second video
Position in image determines the motion track of the target following object.
In one embodiment of first aspect present invention,
It is described to judge in the second area whether to include the target following object, comprising:
The similarity between the first area and the second area is calculated by Hungary Algorithm；
If similarity is greater than or equal to preset threshold, it is determined that include the target following object in the second area；
If similarity is less than preset threshold, it is determined that do not include the target following object in the second area.
In one embodiment of first aspect present invention, the position according to target following object in the first video image
It sets, before determining the corresponding first area of the target following object, further includes:
Position of the target following object in first video image is calculated by deep learning algorithm；
The position according to the target following object in the second region determines that the target following object exists
Before position in second video image, further includes:
The position of the target following object in the second region is calculated by deep learning algorithm.
In one embodiment of first aspect present invention, further includes:
By feature of the multiple target following objects of deep learning algorithm training in multiple third video images, described the
Three video images are identical as the attribute information of first video image；
Pass through feature of the multiple target following objects of deep learning algorithm training in multiple the fourth regions, the 4th area
Domain includes the target following object and the area of the fourth region is identical as the area of the second area.
Second aspect of the present invention provides a kind of object tracking apparatus based on video image, comprising:
First determining module determines the target for the position according to target following object in the first video image
Track the corresponding first area of object；
Second determining module, for determining the target following object in the second video image according to the first area
The corresponding second area of body；The area of the second area is greater than the area of the first area, and in the second area
Center position phase in first video image of the heart in position and the first area in second video image
Together；
Third determining module, for the position according to the target following object in the second region, determine described in
Position of the target following object in second video image.
In one embodiment of second aspect of the present invention, further includes:
4th determining module, for according to position of the target following object in first video image and in institute
The position in the second video image is stated, determines the motion track of the target following object.
In one embodiment of second aspect of the present invention, the 4th determining module is specifically used for,
Third region corresponding with the target following object is determined from the second area；The first area and institute
State the area equation in third region, relative position and the target following object of the target following object in the first area
Body is identical in the relative position in the third region；
According to position of the first area in first video image and the third region in second view
Position in frequency image determines the motion track of the target following object.It is described in one embodiment of second aspect of the present invention
4th determining module is also used to,
Judge in the second area whether to include the target following object；
If so, according to position of the target following object in first video image and in second video
Position in image determines the motion track of the target following object.
In one embodiment of second aspect of the present invention, the 4th determining module is specifically used for,
The similarity between the first area and the second area is calculated by Hungary Algorithm；
If similarity is greater than or equal to preset threshold, it is determined that include the target following object in the second area；
If similarity is less than preset threshold, it is determined that do not include the target following object in the second area.
In one embodiment of second aspect of the present invention, first determining module is also used to, and passes through deep learning algorithm meter
Calculate position of the target following object in first video image；
The third determining module is also used to, and calculates the target following object described second by deep learning algorithm
Position in region.
In one embodiment of second aspect of the present invention, further includes: training module, for more by the training of deep learning algorithm
Feature of a target following object in multiple third video images, the third video image and first video image
Attribute information is identical；
Pass through feature of the multiple target following objects of deep learning algorithm training in multiple the fourth regions, the 4th area
Domain includes the target following object and the area of the fourth region is identical as the area of the second area.
Third aspect present invention provides a kind of electronic equipment, comprising: processor, the processor are coupled with memory；Institute
It states memory to be used for, stores computer program；The processor is used for, and calls the computer program stored in the memory,
To realize method described in aforementioned first aspect any embodiment.
Fourth aspect present invention provides a kind of electronic equipment readable storage medium storing program for executing, comprising: described program is worked as in program or instruction
Or instruction realizes method described in aforementioned first aspect any embodiment when running on an electronic device.
To sum up, the present invention provides a kind of object tracking methods and device based on video image, and wherein method includes: basis
Position of the target following object in the first video image determines the corresponding first area of target following object；According to the firstth area
Domain determines the corresponding second area of target following object in the second video image；The area of second area is greater than first area
Area, and the center of second area position and first area in the second video image center in the first video image
Position it is identical；According to the position of target following object in the second area, determine target following object in the second video image
In position.Object tracking methods provided by the invention based on video image carry out the object in continuous video image
When tracking, when position of the target following object in video image has been determined, only pass through target following object in subsequent calculating
Body is after the second area that the position in the first video image determines in the second video image, according to target following object second
Position in region determines position of the target following object in the second video image.Therefore it does not need in subsequent target following
During object tracking, identification calculating is carried out to entire video image, and then reduce calculating when tracking for object
It measures, improve computational efficiency.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention without any creative labor, may be used also for those of ordinary skill in the art
To obtain other drawings based on these drawings.
Fig. 1 is that the present invention is based on the flow diagrams of one embodiment of object tracking methods of video image；
Fig. 2A is that the present invention is based on the application schematic diagrams of one embodiment of object tracking methods of video image；
Fig. 2 B is that the present invention is based on the application schematic diagrams of one embodiment of object tracking methods of video image；
Fig. 2 C is that the present invention is based on the application schematic diagrams of one embodiment of object tracking methods of video image；
Fig. 2 D is that the present invention is based on the application schematic diagrams of one embodiment of object tracking methods of video image；
Fig. 3 is that the present invention is based on the flow diagrams of one embodiment of object tracking methods of video image；
Fig. 4 is that the present invention is based on the application schematic diagrams of one embodiment of object tracking methods of video image；
Fig. 5 is that the present invention is based on the structural schematic diagrams of one embodiment of object tracking apparatus of video image；
Fig. 6 is that the present invention is based on the structural schematic diagrams of one embodiment of object tracking apparatus of video image.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Description and claims of this specification and term " first ", " second ", " third ", " in above-mentioned attached drawing
The (if present)s such as four " are to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should manage
The data that solution uses in this way are interchangeable under appropriate circumstances, so that the embodiment of the present invention described herein for example can be to remove
Sequence other than those of illustrating or describe herein is implemented.In addition, term " includes " and " having " and theirs is any
Deformation, it is intended that cover it is non-exclusive include, for example, containing the process, method of a series of steps or units, system, production
Product or equipment those of are not necessarily limited to be clearly listed step or unit, but may include be not clearly listed or for this
A little process, methods, the other step or units of product or equipment inherently.
Technical solution of the present invention is described in detail with specifically embodiment below.These specific implementations below
Example can be combined with each other, and the same or similar concept or process may be repeated no more in some embodiments.
Fig. 1 is that the present invention is based on the flow diagrams of one embodiment of object tracking methods of video image.As shown in Figure 1,
Object tracking methods provided in this embodiment based on video image include:
S101: according to position of the target following object in the first video image, target following object corresponding is determined
One region；
S102: the corresponding second area of target following object in the second video image is determined according to first area；The
The area in two regions is greater than the area of first area, and position of the center of second area in the second video image and the firstth area
Position of the center in domain in the first video image is identical；
S103: according to the position of target following object in the second area, determine target following object in the second video figure
Position as in.
The executing subject of the present embodiment can be any electronic equipment for having dependent image data processing function, such as:
Perhaps the executing subject of server etc. or the present embodiment can also be the chip of electronic equipment for mobile phone, computer, tablet computer,
Such as CPU or GPU etc..
Below with reference to Fig. 2A -2D, the embodiment in above-mentioned Fig. 1 is illustrated.Wherein, Fig. 2A is that the present invention is based on videos
The application schematic diagram of one embodiment of object tracking methods of image；Fig. 2 B is that the present invention is based on the object tracking methods of video image
The application schematic diagram of one embodiment；Fig. 2 C is the application signal the present invention is based on one embodiment of object tracking methods of video image
Figure；Fig. 2 D is that the present invention is based on the application schematic diagrams of one embodiment of object tracking methods of video image.
It specifically, include at least one target following in the first video image 1 as Fig. 2A shows the first video image 1
Object.Wherein, at least one target following object can be a target following object or multiple target following objects.This reality
It applies in example and is illustrated so that at least one target following object is two as an example, wherein at least one target following object includes:
First object tracks object 11 and the second target following object 21.First object tracks object 11 and the second target following object 21
The different location being in the first video image 1 respectively.Similarly, as Fig. 2 C shows the second video image 2, the second video figure
As equally including first object tracking object 11 and the second target following object 12 in 2.And first object tracking object 11 exists
First video image 1 and the difference the location of in the second video image 2, the second target following object 12 is in the first video figure
It is also different the location of in picture 1 and the second video image 2.
It should be noted that each first video image as described in the examples of the application and the second video image should be same
Image in one video capture device video flowing collected, and video capture device acquires the first video image and the second view
It needs to be kept fixed when frequency image, so that the field in the first video image and the second video image in addition to target following object
Scape is fixed, so as to track the rail of target following object by relative position of the target following object in different video image
Mark.Optionally, the first video image and the second video image can be the continuous image of any two in video flowing, alternatively, the
Two video images are after the first video image, and the two interval preset quantity video image.
In the S101 of the present embodiment, it is necessary first to determine at least one target following object in the first video image
Position, and determine according to the position of at least one target following object at least one first area in the first video image.Such as
Shown in Fig. 2 B, it needs to be determined that including the first area of target following object in S101, due to including the in the first video image 1
One target following object 11 and the second target following object 21, therefore in this example it needs to be determined that two first areas.Wherein,
First area 101 in such as figure is corresponding with first object tracking object 11 and tracks object 11, first area including first object
201 is corresponding with the second target following object 21 and including the second target following object 21.
Optionally, first area here refers to the region where target following object, and due to target following object
It is shape, of different sizes, therefore the first area in the present embodiment can be set to accommodate the minimum rectangle of target following object
Region.Such as in Fig. 2 B, due to first object tracking object 11 be it is round, first area 101 is and circular first mesh
The tangent rectangle of mark tracking object 11, also due to the circle of the second target following object 21, therefore second area 201 is and circle
The tangent rectangle of second target following object 21 of shape.When the target following object in video image be other shapes when, square
The size in shape region can be adjusted according to the profile of target following object, and implementation is identical, repeats no more.
Then, in S102, according to position of at least one first area in the first video image identified in S101
It sets, determines at least one second area in the second video image.For example, being had determined that in fig. 2b by S101 first
Corresponding two first areas 101 and 201 of two target following objects in video image 1, then the second video shown in Fig. 2 D
Corresponding two second areas of two target following objects, label 102 and 202 as shown in the figure in image 2.Wherein, the first mesh
Mark tracking object 11, first area 101 and second area 102 are corresponding, the second target following object 21,201 and of second area
Second area 202 is corresponding.Specifically, due to first area 101 fixed in the first video image 1, can by coordinate,
The modes such as pixel mark position of the first area 101 in the first video image 1, and equally mark in the second video image 2
First area 101 as shown by dashed lines out.Then, it is assumed that first area 101 is square a length of s1 of regional edge, then in figure 2d
Second area 102 is obtained after the side length of first area being extended 2 times, the side length of second area 102 is denoted as s3.Same side
Formula obtains second area 202, and 202 side length of second area is denoted as s4.
Optionally, second area here be on the basis of first area, it is outside as starting point using the center of first area
Expanding obtained region, the mode that second area is obtained after doubling the side length of first area in above-described embodiment is merely illustrative,
It can be adjusted according to actual needs in the application, such as area expands as the modes such as 2 times also in the present embodiment protection scope,
The area of second area is only needed to be greater than the area of first area.
Finally, existing in S103 according to second area identified in S102, and according at least one target following object
Position at least one second area determines at least one position of target following object in the second video image.Specifically
Ground, as shown in Figure 2 D, due in the second video image 2, first object tracks object 11 and the second target following object 21
Position is all different compared in the first video image 1.That is, first object tracks object 11 not in the range of first area 101
It is interior but in the range of second area 102, the second target following object 21 not in the range of first area 201 but exist
In the range of second area 202.Simultaneously as the position of second area 102 and second area 202 in the second video image 2
Determined by S102, therefore in order to calculate position of the first object tracking object 11 in the second video image 2, it is only necessary to
First object tracking object 11 is calculated behind the relative position in second area 102, calculates first again depending on the relative position
Position of the target following object 11 in the second video image 2；For the purposes of calculating the second target following object 21 in the second view
Position in frequency image 2, it is only necessary to the second target following object 21 is calculated behind the relative position in second area 202, according to
The relative position calculates position of the second target following object 21 in the second video image 2 again.
Such as: the first video image in Fig. 2 B is expressed as the matrix form of long wide 50 pixel of 100 pixels with pixel, and first
Coordinate of the target following object 11 in the first video image is (20,30).In the square of the second area 102 as obtained in Fig. 2 D
The coordinate at four angles of shape is respectively (5,10), (5,40), (35,10) and (35,40), and the rectangular area of second area 102 is 30*
30 matrix.Then in S103, it is only necessary to determine first object track coordinate of the object 11 in second area 102 be (10,
15) after, according to coordinate relationship of the second area 102 in the second video image, it can determine that first object tracks object 11
Coordinate in the second video image 2 is (15,25).
Therefore, in the object tracking methods provided in this embodiment based on video image, in continuous video image
When target following object is tracked, when position of the target following object in the first video image has been determined, in subsequent meter
Only by target following object after the second area that the position in the first video image determines in the second video image in calculation, root
Position of the target following object in the second video image is determined according to the position of target following object in the second area.Therefore exist
During subsequent target following object tracking, it is only necessary to identification calculating is carried out to second area, without to entire second
Video image carries out identification calculating, and then reduces calculation amount when tracking for object, improves computational efficiency.
Optionally, the target following object in above-described embodiment can be pedestrian, or the more specifically head of pedestrian
Portion and shoulder, above-described embodiment can be applied to the scene of video monitoring, unmanned supermarket and other fields, be shot by fixed monitoring device
After video image, video image is handled to realize the tracking to pedestrians multiple in video image.
Further, Fig. 3 is that the present invention is based on the flow diagrams of one embodiment of object tracking methods of video image.Such as
On the basis of embodiment embodiment shown in Fig. 2 shown in Fig. 3, after S103 further include:
S104: according at least one target following object in the position in the first video image and in the second video image
Position, determine the motion track of at least one target following object.
Specifically, in order to realize the variation of the position of at least one target following object in continuous video image
Rule, to monitor the motion track of at least one target following object.Then when process S103 has determined at least one target following
Object is after the position in the second video image, by itself and at least one position of the target following object in the first video image
The changing rule for obtaining the position of at least one target following object jointly is set, such as the first object in above-mentioned example tracks object
Body is moved to the coordinate (15,25) in the second video image from the coordinate (20,30) in the first video image.It can be by coordinate
Variation as first object tracking object motion track and recorded, and can be further by the variation of coordinate mapping
First object tracking the object mobile direction and distance into actual scene, such as obtain user and move 2 meters or class to the north
Like the motion track of the target following object of format.
Optionally, determine that the mode of the motion track of at least one target following object may is that from least one in S104
At least one third region is determined in a second area；Wherein, third region includes target following object, target following object with
Third region corresponds, the area equation of first area and third region, opposite position of the target following object in first area
It sets identical in the relative position in third region as target following object；According at least one first area in the first video image
Position and at least one position of third region in the second video image, determine the movement of at least one target following object
Track.Wherein, by this present embodiment including target following object by first area, therefore first area can be passed through
Motion track embodies the motion track of target following object, and third region refers to can accommodate mesh in the second video image
The minimum rectangular area of mark tracking object, principle is identical as first area in the first video image, repeats no more.That is, passing through
It include the third of target following object in first area, the second video image in first video image including target following object
Region, relative positional relationship between the two regions determine the relative positional relationship of target following object, and further really
It sets the goal and tracks the motion track of object.
Still optionally further, in another embodiment of the present embodiment, in order to realize the tracking for target following object, also
Can provide it is a kind of the position to judge target following object is matched similarity in a manner of.Specifically, pass through breast tooth
Sharp algorithm calculates the similarity between first area corresponding with same target following object and second area；If similarity is greater than
Or it is equal to preset threshold, it is determined that include corresponding target following object in second area；And it is wrapped in determining second area
When including target following object, S103 is executed according to position of at least one target following object in the first video image and the
Position in two video images determines the motion track of at least one target following object.If similarity is less than preset threshold,
Determine in second area do not include corresponding target following object.And determine that in second area do not include target following object again
When, judge that target following object has had been moved out the range that video image can be shot, does not continue to track the target following object, and
Continue to monitor next frame video image.
Such as: Fig. 4 shows a kind of application schematic diagram of above-described embodiment.Wherein, there are two virtual containers, monitorings
Target container and tracking target container.Wherein, detection target container is for keeping at least one target following object in the first view
Position in frequency image indicates the position of 3 target following objects as shown in the figure with the position of 3 first areas；Tracking
Target container is used to store the position of at least one target following object in the second video image, as shown a mesh in Fig. 2
The change in location of the multi-frame video image of mark tracking object.Here after the second video image can be regarded as the first video image
All video images.Then in whether matching module, by Hungary Algorithm in target following object and the second video image
Second area matched, this meeting is so that all position frames in detection target container are divided into matches and do not match two
Class.The position frame not matched can be considered the target for having just enter into the visual field, be added into tracking target container；At this point, matching
Position frame be then considered as be present in tracking target container in, do not deal with.
Optionally, using Hungary Algorithm to calculating first area corresponding with same target following object and second area
Between similarity used by mathematical form are as follows:Wherein,detector
Refer to detection target container at least one target following object number, trackers refer to tracking target container in comprising target with
The number of the second area of track object, sim (Oi,Oj) indicate to detect the overlapping rate in i-th, j region in two target containers, table
Show the measurement of the two similarity.S indicates the optimization of the higher tracking Container elements spatial match of similarity in target container.
Subsequently, for a new frame video image, the position of all target following objects in target container is tracked in order to obtain
It sets.What first area represented in tracking target container is target following object in the position at t-1 frame moment, firstly, to detect mesh
It marks in container centered on all first areas, respectively expands the bounding box of 2 times of ranges outward as second area；Then, own
The video image of second area by partial model trained in advance, obtains the firstth area of peak response in corresponding region respectively
Domain；Finally, updating tracking container using the first area of peak response, and then realize the purpose of target following.
Optionally, in the above embodiments, before S101 further include: calculate at least one mesh by deep learning algorithm
Position of the mark tracking object in the first video image；Before S103 further include: calculate at least one by deep learning algorithm
Position of the target following object at least one corresponding second area.That is, determining target following object in the present embodiment
When position, using the deep learning algorithm (such as convolutional neural networks) based on machine learning, by advance to different targets
After tracking object is trained, its study is set to arrive the feature of target following object, and in identification to video image to be determined
Feature matched with the feature of the target following object learnt, if successful match illustrates that there are targets in video image
Track object.It is therefore desirable to through the multiple target following objects of deep learning algorithm training in multiple third video images
Feature, third video image is identical as the attribute information of the first video image.Pass through the multiple targets of deep learning algorithm training
Track feature of the object in multiple the fourth regions, the fourth region includes the area and the of target following object and the fourth region
The area in two regions is identical.
Specifically, in deep learning algorithm provided in this embodiment there are the detection model of two kinds of target following objects,
In one is the world model of the target following object based on video image, another kind be mesh based on second area
The partial model of mark tracking object.It is illustrated separately below:
Is acquired by different multiple target following objects and is existed with the generally calculating main body of video image for world model
Multiple video images used in training can be denoted as third video figure here by the feature in different multiple video images
Picture, third video image is identical as the first video image, the attribute information of the second video image in previous embodiment, i.e. third
The attribute informations such as video image and the first video image, the resolution ratio of the second video image, shooting angle, shooting focal length are identical,
In order to subsequent identification.Therefore, using the third video image comprising multiple target following objects as in deep learning algorithm
The input of convolutional neural networks, and after determining video image related data such as attribute information, boundary information, so that convolutional Neural net
After network learning training obtains the feature comprising target following object.If need to judge at this time in the first video image be target with
The position of track object, so that it may be identified by convolutional neural networks.
It is the model being specially arranged for the second area realized in previous embodiment for partial model, which is
This multiple specific region is denoted as the fourth region by multiple specific regions comprising multiple target following objects in video image.
The fourth region is identical as the area of aforementioned second area, and the attribute information of the video image is also and where second area
The attribute information of second video image is identical.Therefore, before S103 determine target following object at the position of second area,
It can be using second area as the input of neural network in deep learning algorithm, so that neural network includes according to what is learnt
The feature of the fourth region of target following object determines the position of target following object in second area.
Further, above two model is based on SSD frame, in frame foundation, has merged basic network
The discrete convolution layer of MobileNet and different resolution.Wherein, target following object to be checked is counted in advance in global visual field institute
Area distributions and length-width-ratio example are accounted for, the hyper parameter of responsible recurrence coordinate position in network is adjusted with this；Meanwhile according to
Different Detection tasks selects the characteristic pattern of corresponding resolution, enables to adapt to target following object dimensional variation and raising
The Reasoning Efficiency of model.
Fig. 5 is that the present invention is based on the structural schematic diagrams of one embodiment of object tracking apparatus of video image.As shown in figure 5,
Object tracking apparatus provided in this embodiment based on video image includes: the first determining module 501, the second determining module 502
With third determining module 503.Wherein, the first determining module 501 is used for according to target following object in the first video image
Position determines the corresponding first area of target following object；Second determining module 502 is used to be determined according to first area second
The corresponding second area of target following object in video image；The area of second area is greater than the area of first area, and the
The center in two regions is identical as position of the center of first area in the first video image in the position in the second video image；
Third determining module 503 determines target following object second for the position according to target following object in the second area
Position in video image.
Object tracking apparatus provided in this embodiment based on video image can be used for realizing as shown in Figure 1 based on video
The object tracking methods of image, its implementation is identical as principle, repeats no more.
Fig. 6 is that the present invention is based on the structural schematic diagrams of one embodiment of object tracking apparatus of video image.It is as shown in FIG. 6
Device is on the basis of device as shown in Figure 5, further includes: the 4th determining module 601.Wherein, the 4th determining module 601 is used for
According to target following object in the position in the first video image and the position in the second video image, target following object is determined
The motion track of body.
Object tracking apparatus provided in this embodiment based on video image can be used for realizing as shown in Figure 3 based on video
The object tracking methods of image, its implementation is identical as principle, repeats no more.
Optionally, the 4th determining module 601 is specifically used for, and corresponding with target following object the is determined from second area
Three regions；The area equation of first area and third region, target following object first area relative position and target with
Track object is identical in the relative position in third region；Existed according to position of the first area in the first video image and third region
Position in second video image determines the motion track of target following object.
Optionally, whether the 4th determining module 601 is also used to, judge in second area to include target following object；If so,
Target following is then determined in the position in the first video image and the position in the second video image according to target following object
The motion track of object.
Optionally, the 4th determining module 601 is specifically used for, by Hungary Algorithm calculate first area and second area it
Between similarity；If similarity is greater than or equal to preset threshold, it is determined that include target following object in second area；If similar
Degree is less than preset threshold, it is determined that does not include target following object in second area.
Optionally, the first determining module 501 is also used to, and calculates target following object in the first view by deep learning algorithm
Position in frequency image；Third determining module 503 is also used to calculate target following object in the secondth area by deep learning algorithm
Position in domain.
Optionally, further includes: training module, for training multiple target following objects multiple by deep learning algorithm
Feature in third video image, third video image are identical as the attribute information of the first video image；It is calculated by deep learning
Feature of the multiple target following objects of method training in multiple the fourth regions, the fourth region includes target following object and the 4th
The area in region and the area of second area are identical.
The present invention also provides a kind of electronic equipment, comprising: processor, processor are coupled with memory；Memory is used for, and is deposited
Store up computer program；Processor is used for, and the computer program stored in memory is called, to realize the side of aforementioned any embodiment
Method.
The present invention also provides a kind of electronic equipment readable storage medium storing program for executing, comprising: program or instruction when program or are instructed in electricity
When being run in sub- equipment, the method for realizing aforementioned any embodiment.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above-mentioned each method embodiment can lead to
The relevant hardware of program instruction is crossed to complete.Program above-mentioned can be stored in a computer readable storage medium.The journey
When being executed, execution includes the steps that above-mentioned each method embodiment to sequence；And storage medium above-mentioned include: ROM, RAM, magnetic disk or
The various media that can store program code such as person's CD.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (10)
1. a kind of object tracking methods based on video image characterized by comprising
According to position of the target following object in the first video image, corresponding firstth area of the target following object is determined
Domain；
The corresponding second area of the target following object in the second video image is determined according to the first area；It is described
The area of second area is greater than the area of the first area, and the center of the second area is in second video image
Position it is identical as position of the center of the first area in first video image；
According to the position of the target following object in the second region, determine the target following object described second
Position in video image.
2. the method according to claim 1, wherein it is described according to the target following object in secondth area
Position in domain determines the target following object after the position in second video image, further includes:
According to the target following object in the position in first video image and the position in second video image
It sets, determines the motion track of the target following object.
3. according to the method described in claim 2, it is characterized in that, described regard according to the target following object described first
Position in frequency image and the position in second video image determine the motion track of the target following object, packet
It includes:
Third region corresponding with the target following object is determined from the second area；The first area and described
The area equation in three regions, the target following object exist in the relative position of the first area and the target following object
The relative position in the third region is identical；
According to position of the first area in first video image and the third region in the second video figure
Position as in, determines the motion track of the target following object.
4. according to the method described in claim 2, it is characterized in that, described regard according to the target following object described first
Position in frequency image and the position in second video image, determine the target following object motion track it
Before, further includes:
Judge in the second area whether to include the target following object；
If so, according to position of the target following object in first video image and in second video image
In position, determine the motion track of the target following object.
5. according to the method described in claim 4, it is characterized in that, whether described judge in the second area to include the mesh
Mark tracking object, comprising:
The similarity between the first area and the second area is calculated by Hungary Algorithm；
If similarity is greater than or equal to preset threshold, it is determined that include the target following object in the second area；
If similarity is less than preset threshold, it is determined that do not include the target following object in the second area.
6. method according to claim 1-5, which is characterized in that described to be regarded according to target following object first
Position in frequency image, before determining the corresponding first area of the target following object, further includes:
Position of the target following object in first video image is calculated by deep learning algorithm；
The position according to the target following object in the second region determines the target following object described
Before position in second video image, further includes:
The position of the target following object in the second region is calculated by deep learning algorithm.
7. according to the method described in claim 6, it is characterized by further comprising:
Pass through feature of the multiple target following objects of deep learning algorithm training in multiple third video images, the third view
Frequency image is identical as the attribute information of first video image；
Pass through feature of the multiple target following objects of deep learning algorithm training in multiple the fourth regions, the fourth region packet
It includes the target following object and the area of the fourth region is identical as the area of the second area.
8. a kind of object tracking apparatus based on video image characterized by comprising
First determining module determines the target following for the position according to target following object in the first video image
The corresponding first area of object；
Second determining module, for determining the target following object pair in the second video image according to the first area
The second area answered；The area of the second area is greater than the area of the first area, and the center of the second area exists
Position in second video image is identical as position of the center of the first area in first video image；
Third determining module determines the target for the position according to the target following object in the second region
Track position of the object in second video image.
9. a kind of electronic equipment characterized by comprising processor, the processor are coupled with memory；The memory is used
In storage computer program；The processor is used for, and calls the computer program stored in the memory, to realize right
It is required that any method of 1-7.
10. a kind of electronic equipment readable storage medium storing program for executing characterized by comprising program or instruction, when described program or instruction exist
When running on electronic equipment, method as claimed in claim 1 to 7 is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811386399.6A CN109636828A (en) | 2018-11-20 | 2018-11-20 | Object tracking methods and device based on video image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811386399.6A CN109636828A (en) | 2018-11-20 | 2018-11-20 | Object tracking methods and device based on video image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109636828A true CN109636828A (en) | 2019-04-16 |
Family
ID=66068580
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811386399.6A Pending CN109636828A (en) | 2018-11-20 | 2018-11-20 | Object tracking methods and device based on video image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109636828A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110188719A (en) * | 2019-06-04 | 2019-08-30 | 北京字节跳动网络技术有限公司 | Method for tracking target and device |
| CN110189364A (en) * | 2019-06-04 | 2019-08-30 | 北京字节跳动网络技术有限公司 | For generating the method and apparatus and method for tracking target and device of information |
| CN110390261A (en) * | 2019-06-13 | 2019-10-29 | 北京汽车集团有限公司 | Object detection method, device, computer readable storage medium and electronic equipment |
| CN110977950A (en) * | 2019-11-12 | 2020-04-10 | 长沙长泰机器人有限公司 | Robot grabbing and positioning method |
| CN111583660A (en) * | 2020-05-22 | 2020-08-25 | 济南博观智能科技有限公司 | Vehicle steering behavior detection method, device, equipment and storage medium |
| WO2022056924A1 (en) * | 2020-09-21 | 2022-03-24 | 西门子（中国）有限公司 | Target positioning method and device, and computer-readable medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102521580A (en) * | 2011-12-21 | 2012-06-27 | 华平信息技术（南昌）有限公司 | Real-time target matching tracking method and system |
| CN107240120A (en) * | 2017-04-18 | 2017-10-10 | 上海体育学院 | The tracking and device of moving target in video |
-
2018
- 2018-11-20 CN CN201811386399.6A patent/CN109636828A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102521580A (en) * | 2011-12-21 | 2012-06-27 | 华平信息技术（南昌）有限公司 | Real-time target matching tracking method and system |
| CN107240120A (en) * | 2017-04-18 | 2017-10-10 | 上海体育学院 | The tracking and device of moving target in video |
Non-Patent Citations (2)
| Title |
|---|
| BO LI 等: "High Performance Visual Tracking with Siamese Region Proposal Network", 《2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| LUCA BERTINETTO 等: "Fully-Convolutional Siamese Networks for Object Tracking", 《ARXIV》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110188719A (en) * | 2019-06-04 | 2019-08-30 | 北京字节跳动网络技术有限公司 | Method for tracking target and device |
| CN110189364A (en) * | 2019-06-04 | 2019-08-30 | 北京字节跳动网络技术有限公司 | For generating the method and apparatus and method for tracking target and device of information |
| CN110189364B (en) * | 2019-06-04 | 2022-04-01 | 北京字节跳动网络技术有限公司 | Method and device for generating information, and target tracking method and device |
| CN110390261A (en) * | 2019-06-13 | 2019-10-29 | 北京汽车集团有限公司 | Object detection method, device, computer readable storage medium and electronic equipment |
| CN110977950A (en) * | 2019-11-12 | 2020-04-10 | 长沙长泰机器人有限公司 | Robot grabbing and positioning method |
| CN110977950B (en) * | 2019-11-12 | 2021-05-25 | 长沙长泰机器人有限公司 | Robot grabbing and positioning method |
| CN111583660A (en) * | 2020-05-22 | 2020-08-25 | 济南博观智能科技有限公司 | Vehicle steering behavior detection method, device, equipment and storage medium |
| WO2022056924A1 (en) * | 2020-09-21 | 2022-03-24 | 西门子（中国）有限公司 | Target positioning method and device, and computer-readable medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109636828A (en) | Object tracking methods and device based on video image | |
| Luo et al. | Fast and furious: Real time end-to-end 3d detection, tracking and motion forecasting with a single convolutional net | |
| US20190188533A1 (en) | Pose estimation | |
| Wojek et al. | Monocular 3d scene understanding with explicit occlusion reasoning | |
| CN110706259B (en) | Space constraint-based cross-shot tracking method and device for suspicious people | |
| US20190130583A1 (en) | Still and slow object tracking in a hybrid video analytics system | |
| CN109190508B (en) | Multi-camera data fusion method based on space coordinate system | |
| CN105405150B (en) | Anomaly detection method and device based on fusion feature | |
| KR20200040665A (en) | Systems and methods for detecting a point of interest change using a convolutional neural network | |
| CN111985385B (en) | Behavior detection method, device and equipment | |
| Kumar et al. | Motion trajectory for human action recognition using fourier temporal features of skeleton joints | |
| CN104484868B (en) | The moving target of a kind of combination template matches and image outline is taken photo by plane tracking | |
| CN101167086A (en) | Human detection and tracking for security applications | |
| CN102831382A (en) | Face tracking apparatus and method | |
| CN111047626A (en) | Target tracking method and device, electronic equipment and storage medium | |
| CN106033601A (en) | Method and apparatus for detecting abnormal situation | |
| CN112633282B (en) | Real-time tracking method for vehicle and computer readable storage medium | |
| CN104217192B (en) | Hand localization method and equipment based on depth image | |
| CN110827320B (en) | Target tracking method and device based on time sequence prediction | |
| CN111091025B (en) | Image processing method, device and equipment | |
| CN105975923A (en) | Method and system for tracking human object | |
| CN102521842A (en) | Method and device for detecting fast movement | |
| CN112541938A (en) | Pedestrian speed measuring method, system, medium and computing device | |
| CN111899279A (en) | Method and device for detecting motion speed of target object | |
| CN114067428A (en) | Multi-view multi-target tracking method and device, computer equipment and storage medium |
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: 20190416 |