CN106355603A - Method and device for human tracking - Google Patents
Method and device for human tracking Download PDFInfo
- Publication number
- CN106355603A CN106355603A CN201610752578.1A CN201610752578A CN106355603A CN 106355603 A CN106355603 A CN 106355603A CN 201610752578 A CN201610752578 A CN 201610752578A CN 106355603 A CN106355603 A CN 106355603A
- Authority
- CN
- China
- Prior art keywords
- human body
- tracked
- human
- tracking
- list
- 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.)
- Granted
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
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
-
- 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
- G06T2207/30232—Surveillance
Abstract
The invention discloses a method and a device for human tracking. The human tracking method comprises the steps of determining a human to be tracked; initializing the accumulated missing frequency of the human into zero; recording the tracking information of the human in a tracking list; detecting the human continuously based on images collected by cameras; updating the tracking information of the currently detected human into the tracking list and clearing the accumulated missing frequency of the human if the the human is detected and the currently detected human is the tracked one; when no human is detected or the currently detected human is not the one which is tracked, and the accumulated missing frequency of the tracked human is not up to the missing frequency threshold value, adding one to the accumulated missing frequency of the tracked human and returning to the aforementioned continuous human detection and subsequent steps which are based on images collected by cameras. The technical scheme provided by the invention can implement continuous tracking of a lost human.
Description
Technical field
The present invention relates to intelligent monitoring technology field is and in particular to a kind of human body tracing method and human body tracking device.
Background technology
Human detection and tracking technique refer to carry out human detection to current scene, when human body has been detected, automatically right
Detection human body is tracked.Human detection and tracking technique can be widely applied to video monitoring, security protection, indoor and outdoor robot, different
In the often application such as behavior analysiss and people's stream statistics, it is a technology with very big researching value and wide application prospect.
At present, conventional human body detecting method mainly has by camera collection image, then passes through feature extraction and identification
Position human body etc. detection technique.Wherein, conventional detection technique has based on histogram of gradients (hog, histogram of ori
Ented gradients) method that combines with support vector machine (svm, support vector machine) and deformable part
Method of part model (dpm, deformable part model) etc..
The above-mentioned human body detecting method based on camera collection image, after human body is detected, can pass through human body tracking technology
Realize human body from motion tracking.At present, conventional human body tracking technology has method based on contour feature, camshift to follow the tracks of to calculate
The filter related with coring of method, mean shift algorithm (i.e. mean-shift), particle filter algorithm, optical flow method, Kalman filtering algorithm
Ripple device (kcf, kernelized correlation filters) algorithm etc..These traditional human body tracking technology above-mentioned are one
Can obtain good tracking effect in simple scene a bit, but work as human body and background to have higher similarity, human motion fast
When speed, human body are blocked or human body attitude changes greatly, just easily the situation with losing in above-mentioned human body tracking technology.Further
Ground, when human body be blocked or because other reasons by with occur in after losing again photographic head monitoring sight line when, above-mentioned human body tracking
Technology will be unable to continue to follow the tracks of by the human body with losing.
Content of the invention
The present invention provides a kind of human body tracing method and human body tracking device, for realizing to by the continuation of the human body with losing
Follow the tracks of.
First aspect present invention provides a kind of human body tracing method, comprising:
Determine tracked human body；
The accumulative loss number of times of above-mentioned tracked human body is initialized as zero；
The tracking information of above-mentioned tracked human body is recorded in following the tracks of list；
Human detection is persistently carried out based on the image that photographic head collects；
When human body is detected and when the secondary human body detecting is above-mentioned tracked human body, will be when the human body time detecting
Tracking information updates in above-mentioned tracking list, and when the accumulative loss number of times of above-mentioned tracked human body is not zero by above-mentioned quilt
The accumulative loss number of times following the tracks of human body resets；
When can't detect human body or when the secondary human body detecting not is above-mentioned tracked human body, detect above-mentioned tracked people
Whether the accumulative loss number of times of body reaches default loss frequency threshold value；
If the accumulative loss number of times of above-mentioned tracked human body reaches above-mentioned loss frequency threshold value, return and execute above-mentioned determination
The step of tracked human body and subsequent step；
If the accumulative loss number of times of above-mentioned tracked human body is not up to above-mentioned loss frequency threshold value, by above-mentioned tracked people
The accumulative loss number of times of body adds one, and returns the step that the above-mentioned image collecting based on photographic head of execution persistently carries out human detection
Rapid and subsequent step.
Second aspect present invention provides a kind of human body tracking device, comprising:
Determining unit, for determining tracked human body；
Initialization unit, for being initialized as zero by the accumulative loss number of times of above-mentioned tracked human body；
Follow the tracks of updating block, for following the tracks of the tracking information recording above-mentioned tracked human body in list；
Human detection unit, the image for being collected based on photographic head persistently carries out human detection；
Lose number of times detector unit, for can't detect human body or above-mentioned human detection unit when above-mentioned human detection unit
When the secondary human body detecting not is above-mentioned tracked human body, detect whether the accumulative loss number of times of above-mentioned tracked human body reaches
Default loss frequency threshold value；
Lose accumulated unit, for the accumulative loss of above-mentioned tracked human body is detected when above-mentioned loss number of times detector unit
When number of times is not up to above-mentioned loss frequency threshold value, the accumulative loss number of times of above-mentioned tracked human body is added one；
Above-mentioned determining unit detects the accumulative loss number of times of above-mentioned tracked human body in above-mentioned loss number of times detector unit
Trigger again when reaching above-mentioned loss frequency threshold value；
Above-mentioned tracking updating block is additionally operable to: when above-mentioned human detection unit detects human body and above-mentioned human detection unit
When the secondary human body detecting is above-mentioned tracked human body, by above-mentioned human detection unit when the tracking letter of the secondary human body detecting
Breath updates in above-mentioned tracking list；
Above-mentioned initialization unit is additionally operable to: works as when above-mentioned human detection unit detects human body and above-mentioned human detection unit
The secondary human body detecting is above-mentioned tracked human body, and when the accumulative loss number of times of above-mentioned tracked human body is not zero, will be above-mentioned
The accumulative loss number of times of tracked human body resets.
Therefore, the present invention, after determining tracked human body, initializes the accumulative loss number of times of this tracked human body
And the image being collected based on photographic head persistently carries out human detection, when can't detect human body or when the human body that time detect is not
During above-mentioned tracked human body (when tracked human body is with losing), detect whether this is reached by the accumulative loss number of times with body of loseing face pre-
If loss frequency threshold value, if not up to, the loss number of times of this tracked human body is added continue in the lump to keep to this by with
The tracking of track human body, if reached, re-starts the determination of tracked human body.On the one hand, by introducing tracked human body
Accumulative loss number of times, can realize following the tracks of to by the continuation of the human body with losing in the case of with losing tracked human body；The opposing party
Face, loses frequency threshold value it is also possible to avoid not appearing in the figure of this photographic head collection for a long time because of tracked human body by setting
As interior and lead to cannot be carried out for a long time the problem of human body tracking, improve the motility of human body tracking.
Brief description
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
The accompanying drawing having required use in technology description is briefly described.It should be evident that drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, without having to pay creative labor, also may be used
So that other accompanying drawings are obtained according to these accompanying drawings.
One embodiment schematic flow sheet of a kind of human body tracing method that Fig. 1-a provides for the present invention；
Position coverage rate explanation schematic diagram under a kind of application scenarios that Fig. 1-b provides for the present invention；
Another embodiment schematic flow sheet of a kind of human body tracing method that Fig. 2 provides for the present invention；
One example structure schematic diagram of a kind of human body tracking device that Fig. 3 provides for the present invention.
Specific embodiment
For enabling the goal of the invention of the present invention, feature, advantage more obvious and understandable, below in conjunction with the present invention
Accompanying drawing in embodiment, is clearly and completely described to the technical scheme in the embodiment of the present invention.Obviously, described reality
Applying example is only a part of embodiment of the present invention, and not all embodiments.Based on the embodiment in the present invention, the common skill in this area
The every other embodiment that art personnel are obtained under the premise of not making creative work, broadly falls into the model of present invention protection
Enclose.
Embodiment one
Present example provides a kind of human body tracing method.Human body tracking side as shown in Fig. 1-a, in the embodiment of the present invention
Method includes:
Step 101, determine tracked human body；
In a step 101, can persistently carrying out human detection until human body is detected, afterwards the human body detecting being determined
For tracked human body.Or it is also possible to the human body image (comprising the image of human body) of receives input or other characteristics of human body's letter
Breath, determines tracked human body by terminal according to the human body image of input or other characteristics of human body's information.For example, user can be to end
End inputs the human body image that the arbitrary human body image chosen from image library or captured in real-time obtain, and is based on triggering this terminal
The human body that the human body image of input carries out human detection and comprises this human body image is defined as tracked human body.
Optionally, only 1 human body is defined as tracked human body every time.Step 101 is: persistently carry out human detection until
Human body is detected；If the number of the human body detecting is 1, tracked human body will be defined as when the secondary human body detecting；If inspection
The number of the human body measuring is not 1, then will be defined as tracked human body when one of the secondary multiple human bodies detecting human body.Tool
Body, when the number of the human body detecting not is 1, can calculate when the area of secondary each human body detecting respectively, by face
Long-pending maximum human body is defined as tracked human body.Below with the image being collected based on photographic head persistently carry out human detection until
Illustrate as a example human body is detected.The image collecting for photographic head, can adopt more rapid region convolutional neural networks
(i.e. faster r-cnn, wherein, the English full name of r-cnn is region-convolutional neural network) or
Other detection based on the human detection algorithm of image whether there is human body in this image, when the human body number in this image is detected
During for 1, the human body detecting is defined as tracked human body；When the human body number in this image is detected more than 1, can basis
Formula si=wi*hiCalculate each human body area in the picture, afterwards the maximum human body of area is defined as tracked human body.?
In above-mentioned formula, siRepresent human body i area in the picture, wiRepresent the human body rectangle that the human body i detecting in the picture is located
The width of frame, hiRepresent the height of the human body rectangle frame that the human body i detecting in the picture is located.
It should be noted that 1 human body is defined as tracked human body by such scheme every time, certainly, in other embodiments
In it is also possible to every time n human body is defined as tracked human body, wherein, n is the preset value more than 2.In a step 101, work as inspection
When measuring human body and the number no more than n of human body, the human body detecting all is defined as tracked human body；When human body is detected
And the number of human body be more than n when, n human body in the human body detecting is defined as tracked human body.The embodiment of the present invention is not
To determining in step 101 that the specific implementation that tracked human body is adopted is defined.
Step 102, the accumulative loss number of times of above-mentioned tracked human body is initialized as zero；
In a step 102, the accumulative loss number of times of tracked human body step 101 being determined is initialized as zero.
Specifically, the accumulative loss number of times of the tracked human body that can be determined by tabular form recording step 101.Example
As by losing the accumulative loss number of times of the tracked human body that list recording step 101 determines, then step 102 is embodied in
Initialize above-mentioned loss list, wherein, the loss list after initialization is sky.
Further, this loss list, can also be in tracked human body except recording the accumulative loss number of times of tracked human body
After losing, store the identification information of tracked human body, after in tracked human body with losing, can be by this loss list
The identification information of the tracked human body of storage, judges that the human body newly detecting with the tracked human body of record in this loss list is
No for same human body.
Step 103, record the tracking information of above-mentioned tracked human body in following the tracks of list；
In step 103, follow the tracks of list in recording step 101 determine tracked human body tracking information.This tracking
Information includes but is not limited to: the identification information of tracked human body, the positional information of tracked human body.Wherein, this tracked human body
Identification information namely the information being based on during this tracked human body is being detected, the positional information instruction of this tracked human body should
Tracked human body location when being detected.For example, when the image being gathered based on photographic head persistently carries out human detection
When, if tracked human body is detected from image a1, what the identification information of this tracked human body namely image a1 were comprised should
The image section (image section of the human body rectangle inframe that tracked human body is located in such as image a1) of tracked human body, this quilt
The positional information following the tracks of human body indicates that the residing coordinate position in image a1 of this tracked human body (for example can be by image a1
The left upper apex of human body rectangle frame that the tracked human body detecting is located and the coordinate of bottom right vertex are considered as this tracked human body
Residing coordinate position in image a1).
Step 104, human detection is persistently carried out based on the image that photographic head collects；
At step 104, human detection is persistently carried out based on the image that photographic head collects, this photographic head can be terminal
The photographic head of (such as monitor terminal, robot etc.) upper configuration, or or can be with other shooting of this terminal called
Head instrument, is not construed as limiting herein.
Optionally, at step 104, the image being collected based on photographic head, persistently enters pedestrian using faster r-cnn
Health check-up survey, using convolutional neural networks powerful extraction image further feature ability, realize more accurate human detection with
Follow the tracks of.Certainly, in the embodiment of the present invention, it would however also be possible to employ other is collected to this photographic head based on the human detection algorithm of image
Image carry out human detection, do not limit herein.
Optionally, when human body is detected, judge when whether the human body that time detect is above-mentioned tracking list or above-mentioned loses
Lose the tracked human body of list records；If when the secondary human body detecting is above-mentioned tracking list or the quilt of above-mentioned loss list records
Follow the tracks of human body, then judge to work as the secondary human body detecting as above-mentioned tracked human body；If when the human body that time detect be not above-mentioned with
The tracked human body of track list records and be not above-mentioned loss list records tracked human body, then judge as the people that time detect
Body is not above-mentioned tracked human body.In a kind of application scenarios, whether above-mentioned judgement is above-mentioned tracking when the secondary human body detecting
The tracked human body of list or above-mentioned loss list records may include that and judges whether when the secondary human body detecting be above-mentioned tracking
The tracked human body of list records；If when the secondary human body detecting is not the tracked human body of above-mentioned tracking list records, sentencing
Disconnected when the human body that time detect be whether the tracked human body of above-mentioned loss list records.That is, first judge as the secondary people detecting
Whether body is the tracked human body of above-mentioned tracking list records, be not above-mentioned tracking list records when the human body that time detect
Tracked human body, then judge that whether when the human body that time detect be the tracked human body of above-mentioned loss list records.
Specifically, when above-mentioned tracking information includes the image of above-mentioned tracked human body, the embodiment of the present invention can be passed through
Following manner judges to be whether the tracked human body of above-mentioned tracking list records when the human body that time detect: based on above-mentioned tracking letter
Cease the image of above-mentioned tracked human body comprising, judged using default human bioequivalence model when the human body that time detect be whether
The tracked human body of above-mentioned tracking list records.Wherein, above-mentioned human bioequivalence model is based on convolutional neural networks training and obtains, on
State human bioequivalence model for judging whether two human bodies belong to same human body.By substantial amounts of training data and based on convolution
This human bioequivalence model of neural metwork training, enables to carry out human bioequivalence using this human bioequivalence model (judge two
Whether human body belongs to is same human body) identification accuracy be better than traditional recognition methodss (side for example based on hog feature
Method and the method based on contour feature etc.).In the same manner, it would however also be possible to employ said method judges on whether the human body that time detect is
State the tracked human body losing list records.
Further, also include the positional information of this tracked human body in above-mentioned tracking information in the case of, above-mentioned judgement is worked as
Whether the secondary human body detecting is that the tracked human body of above-mentioned tracking list records may include that and remembers according in above-mentioned tracking list
The positional information of this tracked human body of record, calculates this tracked human body and covers with when the position of the secondary all human bodies detecting
Rate；If existence position coverage rate is not less than the human body of default coverage rate threshold value, the position with this tracked human body is covered
What rate was maximum works as the tracked human body that the secondary human body detecting is defined as this tracking list records；If existence position coverage rate is not
Less than the human body of above-mentioned coverage rate threshold value, then the image of the above-mentioned tracked human body being comprised based on above-mentioned tracking information, using upper
State human bioequivalence model and judge that whether when the human body that time detect be the tracked human body of above-mentioned tracking list records.Wherein, on
Rheme puts coverage rate namely human body shared region (the human body rectangle that for example human body is located in different images in different images
Frame) overlapping area and non-overlapped area ratio.Because step 104 is persistently to enter pedestrian based on the image that photographic head collects
Health check-up survey, therefore, tracked human body continuous acquisition to two field picture in position should vary less, accordingly, tracked
The overlapping area of human body region in two continuous frames image should be larger, and accordingly, tracked human body is in two continuous frames figure
Position coverage rate in picture also can be larger.Therefore in such scheme, before being identified using above-mentioned human bioequivalence model judging,
First pass through the position coverage rate calculating this tracked human body and working as the secondary all human bodies detecting, not little in existence position coverage rate
When the human body of above-mentioned coverage rate threshold value, will be true for the human body that ought secondary detect maximum with the position coverage rate of this tracked human body
It is set to the tracked human body of this tracking list records, be identified judging to need using above-mentioned human bioequivalence model such that it is able to save
Time to be expended, improve human body tracking efficiency.Specifically, in conjunction with Fig. 1-b, above-mentioned position coverage rate is illustrated, if Fig. 1-
C1 in b is the people that the human body (for ease of description, subsequently this human body being described as human body s1) detecting in image a1 is located
Body rectangle frame, c2 is that the human body (for ease of description, subsequently this human body being described as human body s2) detecting in image a2 is located
Human body rectangle frame, take the left upper apex of human body rectangle frame and the coordinate of bottom right vertex as the coordinate position of corresponding human body.Then
Shown in Fig. 1-b, the coordinate position of human body s1 is: left upper apex coordinate (xs11,ys11) and bottom right vertex coordinate (xs12,ys12),
The coordinate position of human body s2 is: left upper apex coordinate (xs21,ys21) and bottom right vertex coordinate (xs22,ys22), w1And h1Represent respectively
The width of c1 and height, w2And h2Represent the width of c2 and height it is clear that w respectively1=| xs12-xs11|,h1=| ys12-ys11|, w2
=| xs22-xs21|,h2=| ys22-ys21|.Then the overlapping area (subsequently representing this overlapping area with intersec) of c1 and c2 is i.e.
The area in the region of oblique dotted line filling in Fig. 1-b, the non-overlapped area of c1 and c2 is that the area sum of c1 and c2 deducts this overlap
Area (subsequently represents this non-overlapped area with union), then the ratio of intersec and union is human body s1's and human body s2
Position coverage rate.Represent the position coverage rate of human body s1 and human body s2 with overlap, then exist: overlap=intersec/
Union, wherein, intersec=(x1+w1-x2)*(y1+h1-y2), union=w1*h1+w2*h21-intersec.
Step 105, when human body is detected and when the secondary human body detecting is above-mentioned tracked human body, detect when secondary
The tracking information of human body update in above-mentioned tracking list, and when the accumulative loss number of times of above-mentioned tracked human body is not zero
The accumulative loss number of times of above-mentioned tracked human body is reset；
In step 105, when detect human body and when the secondary human body detecting be above-mentioned tracked human body when, when time examine
The tracking information of the human body measuring updates in above-mentioned tracking list, return to step 104 afterwards.Illustrate, if above-mentioned tracking
Information includes: the identification information of tracked human body and positional information.When step 104 detects tracked human body from image a1,
The then image section of this tracked human body that the identification information of this tracked human body namely image a1 are comprised, this tracked human body
Positional information indicate the residing coordinate position in image a1 of this tracked human body, now in above-mentioned tracking list update quilt
Follow the tracks of the tracking information of human body, this tracked human body that the identification information of the tracked human body after renewal is comprised by image a1
Image section, the positional information of the tracked human body after renewal is the residing coordinate position in image a1 of tracked human body.Tool
Body, the left upper apex of human body rectangle frame this tracked human body detecting in image a1 being located and bottom right vertex
Coordinate is considered as the residing coordinate position in image a1 of this tracked human body, or it is also possible to will detect in image a1
The bottom left vertex of human body rectangle frame that this tracked human body is located and the coordinate of right vertices are considered as this tracked human body in image
In a1, residing coordinate position, is not construed as limiting herein.
Further, if the accumulative loss number of times of above-mentioned tracked human body is not zero, show the tracked people before this with losing
Body is traced into again, now resets the accumulative loss number of times of above-mentioned tracked human body.Specifically, if by losing list note
Record the accumulative loss number of times of tracked human body, then the above-mentioned accumulative loss number of times clearing concrete manifestation by above-mentioned tracked human body
For: initialize above-mentioned loss list.
Step 106, when can't detect human body or when the secondary human body detecting not is above-mentioned tracked human body, detection is above-mentioned
Whether the accumulative loss number of times of tracked human body reaches default loss frequency threshold value；
In the embodiment of the present invention, when tracked human body does not appear in the image that above-mentioned photographic head collects, will appear from
Can't detect human body or the human body detecting be not above-mentioned tracked human body situation.Now, detect this tracked people further
Whether the accumulative loss number of times of body reaches default loss frequency threshold value, if the accumulative loss time of this tracked human body is detected
Number reaches default loss frequency threshold value, then return execution step 101, to redefine tracked human body, and based on newly true
Step after fixed tracked human body execution step 101.If the accumulative loss number of times this tracked human body is detected does not reach
To above-mentioned loss frequency threshold value, then enter step 107.
Further, when the human body that can't detect human body or ought secondary detect is not above-mentioned tracked human body, and this is tracked
The accumulative loss number of times of human body is 0, then show that this tracked human body the situation with losing first, now can arrange in above-mentioned loss
The current image following the tracks of this tracked human body of record in list (is for example stored in by the image recording above-mentioned tracked human body in table
In above-mentioned loss list), and empty related to this tracked human body record in above-mentioned tracking list.That is, empty above-mentioned tracking
The tracking information of this tracking human body of record in list.
Step 107, the accumulative loss number of times of above-mentioned tracked human body is added one, and return to step 104；
In the embodiment of the present invention, when the accumulative loss number of times that step 106 detects this tracked human body is not up to above-mentioned losing
When losing frequency threshold value, keep the continuation to this tracking human body to follow the tracks of, the accumulative loss number of times of this tracked human body is added one.Enter one
Step, if by the accumulative loss number of times losing this tracked human body of list records, step 107 is embodied in: this is lost
In list, the accumulative loss number of times of this tracked human body of record adds one.Further, if also record has this quilt in this loss list
Follow the tracks of the image of human body, then the image in this loss list can be kept constant.
It should be noted that the human body tracing method in the embodiment of the present invention can be executed by human body tracking device.Above-mentioned
Human body tracking device can be integrated in robot, monitor terminal or other terminal, be not construed as limiting herein.
Therefore, the embodiment of the present invention, after determining tracked human body, initializes the accumulative of this tracked human body and loses
Lose number of times and the image that collected based on photographic head persistently carries out human detection, when can't detect human body or as the people that time detect
By the accumulative loss number of times with body of loseing face whether body is not during above-mentioned tracked human body (when tracked human body is with losing), detect this
Reach default loss frequency threshold value, if not up to, the loss number of times of this tracked human body is added and continues in the lump to keep right
The tracking of this tracked human body, if reached, re-starts the determination of tracked human body.On the one hand, tracked by introducing
The accumulative loss number of times of human body, can realize following the tracks of to by the continuation of the human body with losing in the case of with losing tracked human body；
On the other hand, by setting loss frequency threshold value it is also possible to avoid adopting because tracked human body does not appear in this photographic head for a long time
Lead to cannot be carried out the problem of human body tracking for a long time in the image of collection, improve the motility of human body tracking.
Embodiment two
Present example records the tracking information of tracked human body and tracked people respectively to follow the tracks of list and loss list
The accumulative loss number of times of body, as shown in Fig. 2 the human body tracing method in the embodiment of the present invention includes:
Step 201, determine tracked human body；
Specifically, step 201 is referred to the description of step 101 in Fig. 1-a, and here is omitted.
List is lost in step 202, initialization；
Wherein, the loss list after initialization is sky.
Step 203, record the tracking information of above-mentioned tracked human body in following the tracks of list；
In step 203, follow the tracks of list in recording step 201 determine tracked human body tracking information.This tracking
Information includes but is not limited to: the identification information of tracked human body, the positional information of tracked human body.
Specifically, step 203 is referred to the description of step 103 in Fig. 1-a, and here is omitted.
Step 204, human detection is persistently carried out based on the image that photographic head collects；
In step 204, human detection is persistently carried out based on the image that photographic head collects, this photographic head can be terminal
The photographic head of (such as monitor terminal, robot etc.) upper configuration, or or can be with other shooting of this terminal called
Head instrument, is not construed as limiting herein.
Optionally, in step 204, the image being collected based on photographic head, persistently enters pedestrian using faster r-cnn
Health check-up survey, using convolutional neural networks powerful extraction image further feature ability, realize more accurate human detection with
Follow the tracks of.Certainly, in the embodiment of the present invention, it would however also be possible to employ other is collected to this photographic head based on the human detection algorithm of image
Image carry out human detection, do not limit herein.
When human body is detected, entering step 205, when can't detect human body, entering step 208.
Step 205, judge that whether when the human body that time detect be the tracked human body of above-mentioned tracking list records；
In the embodiment of the present invention, judge that whether this human body is the tracked human body of above-mentioned tracking list records, if so, then enter
Enter step 207, if it is not, then entering step 206.
Specifically, when above-mentioned tracking information includes: during the image of above-mentioned tracked human body, step 205 can be passed through as follows
Mode judges to be whether the tracked human body of above-mentioned tracking list records when the human body that time detect: based on above-mentioned tracking information bag
Using default human bioequivalence model, the image of the above-mentioned tracked human body containing, judges whether when the secondary human body detecting be above-mentioned
Follow the tracks of the tracked human body of list records.Wherein, above-mentioned human bioequivalence model is based on convolutional neural networks training and obtains, above-mentioned people
Body identification model is used for judging whether two human bodies belong to same human body.By substantial amounts of training data and based on convolutional Neural
This human bioequivalence model of network training, enables to carry out human bioequivalence using this human bioequivalence model (judge two human bodies
Whether belong to is same human body) identification accuracy be better than traditional recognition methodss (method for example based on hog feature and
Method based on contour feature etc.).
Further, also include the positional information of this tracked human body in above-mentioned tracking information in the case of, step 205 is permissible
Including: when human body is detected, according in above-mentioned tracking list record this tracked human body positional information, calculate this by with
Track human body and the position coverage rate working as the secondary all human bodies detecting；If existence position coverage rate is not less than default coverage rate threshold
The human body of value, then be defined as this tracking list by the secondary human body detecting of working as maximum with the position coverage rate of this tracked human body
The tracked human body of record；If existence position coverage rate is not less than the human body of above-mentioned coverage rate threshold value, it is based on above-mentioned tracking
The image of the above-mentioned tracked human body that packet contains, judged using above-mentioned human bioequivalence model when the human body that time detect be whether
The tracked human body of above-mentioned tracking list records.Wherein, above-mentioned position coverage rate namely human body shared region in different images
The overlapping area of (the human body rectangle frame that for example human body is located in different images) and the ratio of non-overlapped area.Due to step
204 is persistently to carry out human detection based on the image that photographic head collects, therefore, the frame figure that tracked human body arrives in continuous acquisition
Position in picture should vary less, and accordingly, the overlapping area of tracked human body region in two continuous frames image should
This is larger, and accordingly, position coverage rate in two continuous frames image for the tracked human body also can be larger.Therefore in such scheme,
Before being identified judging using above-mentioned human bioequivalence model, first pass through that to calculate this tracked human body all with when time detect
The position coverage rate of human body, when existence position coverage rate is not less than the human body of above-mentioned coverage rate threshold value, will be with this tracked people
The human body that ought secondary detect of the position coverage rate maximum of body is defined as the tracked human body of this tracking list records, such that it is able to
Save and be identified judging the time that needs expend using above-mentioned human bioequivalence model, improve human body tracking efficiency.Specifically, close
The description being referred in Fig. 1-a step 104 in the explanation of above-mentioned position coverage rate, here is omitted.
Step 206, judge that whether when the human body that time detect be the tracked human body of above-mentioned loss list records；
In the embodiment of the present invention, when step 205 judges that this human body is not the tracked human body of above-mentioned tracking list records
When, determine whether that whether when the human body that time detect be the tracked human body of above-mentioned loss list records, if so, then enter step
Rapid 207, if it is not, then entering step 208.
Further, in the embodiment of the present invention when tracked human body is first with losing, can't detect this first in step 204
During tracked human body, the image that can record above-mentioned tracked human body in above-mentioned loss list (for example will currently follow the tracks of list
The image of this tracked human body of middle record is stored in above-mentioned loss list), and empty tracked with this in above-mentioned tracking list
The related record of human body.That is, empty the tracking information of this tracking human body of record in above-mentioned tracking list.Specifically, when above-mentioned
When tracking information includes the image of above-mentioned tracked human body, the above-mentioned quilt that step 206 can also be comprised based on above-mentioned tracking information
Follow the tracks of the image of human body, judge whether when the secondary human body detecting be above-mentioned loss list records using above-mentioned human bioequivalence model
Tracked human body.
Step 207, will update in above-mentioned tracking list when the tracking information of the human body time detecting, and in above-mentioned loss
In list, the accumulative loss number of times of the tracked human body of record initializes this loss list when being not zero；
In the embodiment of the present invention, when detect human body and when the secondary human body detecting be above-mentioned tracked human body when, ought
The tracking information of the secondary human body detecting updates in above-mentioned tracking list, return to step 204 afterwards.Illustrate, if above-mentioned
Tracking information includes: the identification information of tracked human body and positional information.When step 204 detects tracked people from image a1
Body, then image section (the such as image of this tracked human body that the identification information of this tracked human body namely image a1 are comprised
The image section of the human body rectangle inframe that tracked human body is located in a1), the positional information of this tracked human body indicate this by with
The residing coordinate position in image a1 of track human body, now updates the tracking information of tracked human body in above-mentioned tracking list,
The image section of this tracked human body that the identification information of the tracked human body after renewal is comprised by image a1, the quilt after renewal
The positional information following the tracks of human body is the residing coordinate position in image a1 of tracked human body.Specifically, can be by image a1
The left upper apex of human body rectangle frame that this tracked human body detecting is located and the coordinate of bottom right vertex are considered as this tracked people
The residing coordinate position in image a1 of body, or it is also possible to this tracked human body detecting in image a1 is located
The coordinate of the bottom left vertex of human body rectangle frame and right vertices is considered as the residing coordinate position in image a1 of this tracked human body,
It is not construed as limiting herein.
Further, if the accumulative loss number of times of above-mentioned tracked human body is not zero, show the tracked people before this with losing
Body is traced into again, now initializes above-mentioned loss list.
Whether step 208, the accumulative loss number of times of the above-mentioned tracked human body of detection reach default loss frequency threshold value；
In the embodiment of the present invention, when tracked human body does not appear in the image that above-mentioned photographic head collects, will appear from
Can't detect human body or the human body detecting be not above-mentioned tracked human body situation, now, detect above-mentioned loss row further
In table, whether the accumulative loss number of times of this tracked human body of record reaches default loss frequency threshold value, if this quilt is detected
Follow the tracks of human body accumulative loss number of times reach default loss frequency threshold value, then return execution step 201 so that redefine by
Follow the tracks of human body, and based on the step after the new tracked human body execution step 201 determining.If this tracked human body is detected
Accumulative loss number of times be not up to above-mentioned loss frequency threshold value, then enter step 209.
Further, when the accumulative loss number of times 0 of above-mentioned tracked human body, then show that this tracked human body occurs first with losing
Situation, the image that now can record above-mentioned tracked human body in above-mentioned loss list (for example will currently follow the tracks of in list note
The image of this tracked human body of record is stored in above-mentioned loss list), and empty in above-mentioned tracking list with this tracked human body
Related record.That is, empty the tracking information of this tracking human body of record in above-mentioned tracking list.
Step 209, the accumulative loss number of times of above-mentioned tracked human body is added one, and return to step 204；
In the embodiment of the present invention, when the accumulative loss number of times that step 208 detects this tracked human body is not up to above-mentioned losing
When losing frequency threshold value, the continuation to this tracking human body is kept to follow the tracks of, and this tracked human body by record in above-mentioned loss list
Accumulative loss number of times add one.Further, if also record has the image of this tracked human body in above-mentioned loss list, can keep
Image in this loss list is constant.
It should be noted that the human body tracing method in the embodiment of the present invention can be executed by human body tracking device, above-mentioned
Human body tracking device can be integrated in robot, monitor terminal or other terminal, be not construed as limiting herein.
Therefore, the embodiment of the present invention, after determining tracked human body, initializes the accumulative of this tracked human body and loses
Lose number of times and the image that collected based on photographic head persistently carries out human detection, when can't detect human body or as the people that time detect
By the accumulative loss number of times with body of loseing face whether body is not during above-mentioned tracked human body (when tracked human body is with losing), detect this
Reach default loss frequency threshold value, if not up to, the loss number of times of this tracked human body is added and continues in the lump to keep right
The tracking of this tracked human body, if reached, re-starts the determination of tracked human body.On the one hand, tracked by introducing
The accumulative loss number of times of human body, can realize following the tracks of to by the continuation of the human body with losing in the case of with losing tracked human body；
On the other hand, by setting loss frequency threshold value it is also possible to avoid adopting because tracked human body does not appear in this photographic head for a long time
Lead to cannot be carried out the problem of human body tracking for a long time in the image of collection, improve the motility of human body tracking.
Embodiment three
The embodiment of the present invention provides a kind of human body tracking device, as shown in figure 3, the human body tracking dress in the embodiment of the present invention
Put 300 to include:
Determining unit 301, for determining tracked human body；
Initialization unit 302, for being initialized as zero by the accumulative loss number of times of above-mentioned tracked human body；
Follow the tracks of updating block 303, for following the tracks of the tracking information recording above-mentioned tracked human body in list；
Human detection unit 304, the image for being collected based on photographic head persistently carries out human detection；
Lose number of times detector unit 305, for can't detect human body or human detection unit when human detection unit 304
304 when the secondary human body detecting not is above-mentioned tracked human body, and whether the accumulative loss number of times of the above-mentioned tracked human body of detection
Reach default loss frequency threshold value；
Lose accumulated unit 306, for losing when loss number of times detector unit 305 detects above-mentioned the accumulative of tracked human body
When mistake number of times is not up to above-mentioned loss frequency threshold value, the accumulative loss number of times of above-mentioned tracked human body is added one；
The accumulative loss number of times that determining unit 301 detects above-mentioned tracked human body in loss number of times detector unit 305 reaches
Trigger again during to above-mentioned loss frequency threshold value；
Follow the tracks of updating block 303 to be additionally operable to: work as when human detection unit 304 detects human body and human detection unit 304
When the secondary human body detecting is above-mentioned tracked human body, by human detection unit 304 when the tracking information of the secondary human body detecting
Update in above-mentioned tracking list；
Initialization unit 302 is additionally operable to: when human detection unit 304 detect human body and human detection unit 304 when time
The human body detecting is above-mentioned tracked human body, and when the accumulative loss number of times of above-mentioned tracked human body is not zero, by above-mentioned quilt
The accumulative loss number of times following the tracks of human body resets.
Optionally, determining unit 301 includes:
Follow the tracks of object detection unit, for persistently carrying out human detection until human body is detected；
Sub- determining unit, for when the human body that above-mentioned tracking object detection unit detects number be 1 when, by above-mentioned with
Track object detection unit is defined as tracked human body when the secondary human body detecting；When above-mentioned tracking object detection unit detects
When the number of human body is not 1, by above-mentioned tracking object detection unit when one of the secondary multiple human bodies detecting human body determines
For tracked human body.
Optionally, above-mentioned sub- determining unit specifically for: the human body detecting when above-mentioned tracking object detection unit
When number is not 1, calculate above-mentioned tracking object detection unit respectively when the area of secondary each human body detecting, by area maximum
Human body is defined as tracked human body.
Optionally, by losing the accumulative loss number of times of the above-mentioned tracked human body of list records；Initialization unit 302 is concrete
For the accumulative loss number of times of above-mentioned tracked human body being reset in the way of initializing above-mentioned loss list；Wherein, initialize
Loss list afterwards is sky.
Optionally, the human body tracking device in the embodiment of the present invention also includes: loses recording unit, for working as human detection
Unit 304 can't detect human body or human detection unit 304 when the human body that time detect be not above-mentioned tracked human body, and above-mentioned
When the accumulative loss number of times of tracked human body is 0, above-mentioned loss list records the image of above-mentioned tracked human body；Initialization
Unit 302 is additionally operable to: when human detection unit 304 can't detect human body or human detection unit 304 when the secondary human body detecting
Be not above-mentioned tracked human body, and the accumulative loss number of times of above-mentioned tracked human body be 0 when, empty in above-mentioned tracking list with upper
State the related record of tracked human body.
Optionally, human detection unit 304 specifically for: when human body is detected, judge human detection unit 304 when time
Whether the human body detecting is above-mentioned tracking list or the tracked human body of above-mentioned loss list records；If human detection unit
304 is above-mentioned tracking list or the tracked human body of above-mentioned loss list records when the secondary human body detecting, then judge people's health check-up
Survey unit 304 to work as the secondary human body detecting is above-mentioned tracked human body；If human detection unit 304 works as the secondary human body detecting not
Tracked human body for above-mentioned tracking list records and be not above-mentioned loss list records tracked human body, then judge people's health check-up
Survey unit 304 to work as the secondary human body detecting is not above-mentioned tracked human body.
Optionally, human detection unit 304 specifically for: when human body is detected, judge human detection unit 304 when time
Whether the human body detecting is the tracked human body of above-mentioned tracking list records；If human detection unit 304 ought secondary detect
Human body is not the tracked human body of above-mentioned tracking list records, then judge that human detection unit 304 when the secondary human body detecting is
The no tracked human body for above-mentioned loss list records.
Optionally, above-mentioned tracking information includes: the figure of the positional information of above-mentioned tracked human body and above-mentioned tracked human body
Picture；Human detection unit 304 specifically for: when human body is detected, according in above-mentioned tracking list record above-mentioned tracked
The positional information of human body, calculates above-mentioned tracked human body with human detection unit 304 when the position of the secondary all human bodies detecting
Coverage rate；If existence position coverage rate is not less than the human body of default coverage rate threshold value, by the position with above-mentioned tracked human body
That puts coverage rate maximum works as the tracked human body that the secondary human body detecting is defined as above-mentioned tracking list records；If not existence position
Coverage rate is not less than the human body of default coverage rate threshold value, the then figure of the described tracked human body comprising based on above-mentioned tracking information
Using default human bioequivalence model, picture, judges whether human detection unit 304 is above-mentioned tracking row when the secondary human body detecting
The tracked human body of table record, wherein, above-mentioned human bioequivalence model is based on convolutional neural networks training and obtains.
Optionally, human detection unit 304 is specifically for the image being collected based on photographic head, using faster r-
Cnn persistently carries out human detection.
It should be noted that the human body tracking device in the embodiment of the present invention can be integrated in robot, monitor terminal or
In other terminals.The function of each functional module of this human body tracking device is referred to the description in said method embodiment,
It implements the associated description that process can refer in said method embodiment, and here is omitted.
Therefore, the human body tracking device in the embodiment of the present invention, after determining tracked human body, initializes this quilt
The image followed the tracks of the accumulative loss number of times of human body and collected based on photographic head persistently carries out human detection, when can't detect human body
Or when the secondary human body detecting not is described tracked human body (when tracked human body is with losing), detect this by with body of loseing face
Whether accumulative loss number of times reaches default loss frequency threshold value, if not up to, by the loss number of times of this tracked human body
Plus continue to keep the tracking to this tracked human body in the lump, if reached, re-start the determination of tracked human body.One side
Face, by introducing the accumulative loss number of times of tracked human body, can realize to by with losing in the case of with losing tracked human body
Human body continuation follow the tracks of；On the other hand, by setting loss frequency threshold value it is also possible to avoid because of tracked human body for a long time not
Occur in the image of this photographic head collection and lead to cannot be carried out the problem of human body tracking for a long time, improve human body tracking
Motility.
It should be understood that disclosed apparatus and method in several embodiments provided herein, can be passed through it
Its mode is realized.
It should be noted that for aforesaid each method embodiment, for easy description, therefore it is all expressed as a series of
Combination of actions, but those skilled in the art should know, the present invention is not limited by described sequence of movement because
According to the present invention, some steps can be carried out using other orders or simultaneously.Secondly, those skilled in the art also should know
Know, embodiment described in this description belongs to preferred embodiment, and involved action and module might not be all these
Bright necessary.
In the above-described embodiments, the description to each embodiment all emphasizes particularly on different fields, and does not have the portion described in detail in certain embodiment
Point, may refer to the associated description of other embodiments.
It is more than the description to a kind of human body tracing method provided by the present invention and human body tracking device, for this area
Those skilled in the art, according to the embodiment of the present invention thought, all have change in specific embodiments and applications
Place, to sum up, this specification content should not be construed as limitation of the present invention.
Claims (18)
1. a kind of human body tracing method is it is characterised in that include:
Determine tracked human body；
The accumulative loss number of times of described tracked human body is initialized as zero；
The tracking information of described tracked human body is recorded in following the tracks of list；
Human detection is persistently carried out based on the image that photographic head collects；
When human body is detected and when the secondary human body detecting is described tracked human body, will be when the tracking of the human body time detecting
Information updating is followed the tracks of in list to described, and will be described tracked when the accumulative loss number of times of described tracked human body is not zero
The accumulative loss number of times of human body resets；
When can't detect human body or when the secondary human body detecting not is described tracked human body, detect described tracked human body
Whether accumulative loss number of times reaches default loss frequency threshold value；
If the accumulative loss number of times of described tracked human body reaches described loss frequency threshold value, return execution described determine by with
The step of track human body and subsequent step；
If the accumulative loss number of times of described tracked human body is not up to described loss frequency threshold value, by described tracked human body
Accumulative number of times of losing adds one, and return execution described based on the image that photographic head collects persistently carry out the step of human detection with
And subsequent step.
2. method according to claim 1 is it is characterised in that the tracked human body of described determination includes:
Persistently carry out human detection until human body is detected；
If the number of the human body detecting is 1, tracked human body will be defined as when the secondary human body detecting；
If the number of the human body detecting is not 1, by when one of the multiple human bodies that time detect human body be defined as by with
Track human body.
3. method according to claim 2 is it is characterised in that described will work as one of the secondary multiple human bodies detecting people
Body is defined as tracked human body and includes:
Calculate respectively when the area of secondary each human body detecting；
The maximum human body of area is defined as tracked human body.
4. method according to claim 1 is it is characterised in that pass through to lose the accumulative of tracked human body described in list records
Lose number of times；
Described the accumulative loss number of times of described tracked human body is initialized as zero, particularly as follows: initialize described loss list；
Described when human body is detected and when the secondary human body detecting is described tracked human body, will be when the human body time detecting
Tracking information updates in described tracking list, and when the accumulative loss number of times of described tracked human body is not zero by described quilt
The accumulative loss number of times following the tracks of human body resets.Particularly as follows: work as human body is detected and when the secondary human body detecting is described tracked
During human body, will update in described tracking list when the tracking information of the secondary human body detecting, and in described tracked human body
Accumulative loss when number of times is not zero initializes described loss list；
Wherein, the loss list after initialization is sky.
5. method according to claim 4 can't detect human body or when the secondary human body detecting is not it is characterised in that working as
Described tracked human body, described by the accumulative loss number of times of described tracked human body add one before also include:
If the accumulative loss number of times of described tracked human body is 0, in the list of described loss, record described tracked human body
Image, and empty related to described tracked human body record in described tracking list.
6. method according to claim 5 is it is characterised in that the described image being collected based on photographic head persistently enters pedestrian
Health check-up is surveyed, comprising:
When human body is detected, judge whether when the secondary human body detecting be described tracking list or described loss list records
Tracked human body；
If working as the secondary human body detecting is the described tracked human body following the tracks of list or described loss list records, judge when secondary
The human body detecting is described tracked human body；
If the human body that ought secondary detect is not the tracked human body of described tracking list records and is not described loss list records
Tracked human body, then judge when the human body that time detect is not as described tracked human body.
7. method according to claim 6 it is characterised in that described judgement when the human body that time detect be whether described with
Track list or the tracked human body of described loss list records, comprising:
Judge whether when the secondary human body detecting be the described tracked human body following the tracks of list records；
If working as the secondary human body detecting is not the described tracked human body following the tracks of list records, judge when the secondary human body detecting
Whether it is the described tracked human body losing list records.
8. method according to claim 7 is it is characterised in that described tracking information includes: the position of described tracked human body
Confidence breath and the image of described tracked human body；
Whether described judgement is the described tracked human body following the tracks of list records when the secondary human body detecting, comprising:
According to the positional information of the described tracked human body of record in described tracking list, calculate described tracked human body and when secondary
The position coverage rate of all human bodies detecting；
If existence position coverage rate is not less than the human body of default coverage rate threshold value, the position with described tracked human body is covered
What lid rate was maximum works as the tracked human body that the secondary human body detecting is defined as described tracking list records；
If existence position coverage rate is not less than the human body of default coverage rate threshold value, the institute comprising based on described tracking information
State the image of tracked human body, judge whether when the secondary human body detecting be that described tracking arranges using default human bioequivalence model
The tracked human body of table record, wherein, described human bioequivalence model is based on convolutional neural networks training and obtains.
9. the method according to any one of claim 1 to 8 is it is characterised in that the described image being collected based on photographic head
Persistently carry out human detection, particularly as follows:
The image being collected based on photographic head, persistently carries out human detection using more rapid region convolutional neural networks.
10. a kind of human body tracking device is it is characterised in that include:
Determining unit, for determining tracked human body；
Initialization unit, for being initialized as zero by the accumulative loss number of times of described tracked human body；
Follow the tracks of updating block, for following the tracks of the tracking information recording described tracked human body in list；
Human detection unit, the image for being collected based on photographic head persistently carries out human detection；
Lose number of times detector unit, for can't detect human body or described human detection unit when secondary when described human detection unit
When the human body detecting not is described tracked human body, detect whether the accumulative loss number of times of described tracked human body reaches default
Loss frequency threshold value；
Lose accumulated unit, for losing, when described, the accumulative loss number of times that number of times detector unit detects described tracked human body
When being not up to described loss frequency threshold value, the accumulative loss number of times of described tracked human body is added one；
The accumulative loss number of times that described determining unit detects described tracked human body in described loss number of times detector unit reaches
Trigger again during described loss frequency threshold value；
Described tracking updating block is additionally operable to: when described human detection unit detect human body and described human detection unit when time
When the human body detecting is described tracked human body, the tracking information of the human body that described human detection unit ought secondary be detected is more
Newly in described tracking list；
Described initialization unit is additionally operable to: when described human detection unit detects human body and described human detection unit when time inspection
The human body measuring is described tracked human body, and when the accumulative loss number of times of described tracked human body is not zero, by described by with
The accumulative loss number of times of track human body resets.
11. human body tracking devices according to claim 10 are it is characterised in that described determining unit includes:
Follow the tracks of object detection unit, for persistently carrying out human detection until human body is detected；
Sub- determining unit, for when the described number following the tracks of the human body that object detection unit detects is 1, by described tracking mesh
Mark detector unit is defined as tracked human body when the secondary human body detecting；Follow the tracks of the human body that object detection unit detects when described
Number when being not 1, by described follow the tracks of object detection unit when one of the multiple human bodies that time detect human body be defined as by
Follow the tracks of human body.
12. human body tracking devices according to claim 11 it is characterised in that
Described sub- determining unit specifically for: when described follow the tracks of the number of human body that object detection unit detects and not be 1 when, point
Do not calculate described object detection unit of following the tracks of when the area of each human body that time detect, by the maximum human body of area be defined as by
Follow the tracks of human body.
13. human body tracking devices according to claim 10 are it is characterised in that pass through tracked described in loss list records
The accumulative loss number of times of human body；
Described initialization unit is specifically for being lost described the accumulative of tracked human body in the way of initializing the list of described loss
Lose number of times to reset；
Wherein, the loss list after initialization is sky.
14. human body tracking devices according to claim 13 are it is characterised in that described human body tracking device also includes:
Lose recording unit, for can't detect human body or described human detection unit when time detection when described human detection unit
To human body be not described tracked human body, and the accumulative loss number of times of described tracked human body be 0 when, in described loss list
The middle image recording described tracked human body；
Described initialization unit is additionally operable to: when described human detection unit can't detect human body or described human detection unit when secondary
The human body detecting is not described tracked human body, and when the accumulative loss number of times of described tracked human body is 0, empty described with
The record related to described tracked human body in track list.
15. human body tracking devices according to claim 14 it is characterised in that described human detection unit specifically for:
When human body is detected, judge described human detection unit when whether the human body that time detect is described tracking list or described loses
Lose the tracked human body of list records；If described human detection unit is described tracking list or described when the secondary human body detecting
Lose the tracked human body of list records, then judge that described human detection unit works as the secondary human body detecting as described tracked people
Body；If described human detection unit is worked as the tracked human body that the secondary human body detecting is not described tracking list records and is not institute
State the tracked human body losing list records, then judge described human detection unit when the human body that time detect not as described by with
Track human body.
16. human body tracking devices according to claim 15 it is characterised in that described human detection unit specifically for:
When human body is detected, judge whether described human detection unit is the described quilt following the tracks of list records when the secondary human body detecting
Follow the tracks of human body；If described human detection unit is not the described tracked human body following the tracks of list records when the secondary human body detecting,
Then judge whether described human detection unit is the described tracked human body losing list records when the secondary human body detecting.
17. human body tracking devices according to claim 16 are it is characterised in that described tracking information includes: described by with
The positional information of track human body and the image of described tracked human body；
Described human detection unit specifically for: when human body is detected, according to described follow the tracks of list in record described by with
The positional information of track human body, calculates described tracked human body with described human detection unit when the position of the secondary all human bodies detecting
Put coverage rate；If existence position coverage rate be not less than default coverage rate threshold value human body, by with described tracked human body
What position coverage rate was maximum works as the tracked human body that the secondary human body detecting is defined as described tracking list records；If there is not position
Put the human body that coverage rate is not less than default coverage rate threshold value, then the described tracked human body being comprised based on described tracking information
Using default human bioequivalence model, image, judges whether described human detection unit is described tracking when the secondary human body detecting
The tracked human body of list records, wherein, described human bioequivalence model is based on convolutional neural networks training and obtains.
18. human body detection devices according to any one of claim 10 to 17 are it is characterised in that described human detection unit
Specifically for: the image being collected based on photographic head, persistently carry out human detection using more rapid region convolutional neural networks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610752578.1A CN106355603B (en) | 2016-08-29 | 2016-08-29 | Human body tracing method and human body tracking device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610752578.1A CN106355603B (en) | 2016-08-29 | 2016-08-29 | Human body tracing method and human body tracking device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106355603A true CN106355603A (en) | 2017-01-25 |
| CN106355603B CN106355603B (en) | 2019-10-22 |
Family
ID=57854633
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610752578.1A Active CN106355603B (en) | 2016-08-29 | 2016-08-29 | Human body tracing method and human body tracking device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106355603B (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273836A (en) * | 2017-06-07 | 2017-10-20 | 深圳市深网视界科技有限公司 | A kind of pedestrian detection recognition methods, device, model and medium |
| CN108762309A (en) * | 2018-05-03 | 2018-11-06 | 浙江工业大学 | It is a kind of based on the assumption that Kalman filtering human body target follower method |
| CN108921008A (en) * | 2018-05-14 | 2018-11-30 | 深圳市商汤科技有限公司 | Portrait identification method, device and electronic equipment |
| CN109145564A (en) * | 2018-08-22 | 2019-01-04 | 中国工商银行股份有限公司 | Control the method and device of mobile terminal |
| CN109271942A (en) * | 2018-09-26 | 2019-01-25 | 上海七牛信息技术有限公司 | A kind of stream of people's statistical method and system |
| CN110176025A (en) * | 2019-04-25 | 2019-08-27 | 沈阳航空航天大学 | A kind of proctor's tracking based on posture |
| CN110264498A (en) * | 2019-06-26 | 2019-09-20 | 北京深醒科技有限公司 | A kind of human body tracing method under video monitoring scene |
| CN110363787A (en) * | 2018-03-26 | 2019-10-22 | 北京市商汤科技开发有限公司 | Information acquisition method and system, electronic equipment, program and medium |
| CN110378931A (en) * | 2019-07-10 | 2019-10-25 | 成都数之联科技有限公司 | A kind of pedestrian target motion track acquisition methods and system based on multi-cam |
| CN110675426A (en) * | 2018-07-02 | 2020-01-10 | 百度在线网络技术（北京）有限公司 | Human body tracking method, device, equipment and storage medium |
| CN111241995A (en) * | 2020-01-09 | 2020-06-05 | 苏宁云计算有限公司 | Loss compensation method and system for tracked object and computer equipment |
| CN111539283A (en) * | 2020-04-15 | 2020-08-14 | 上海摩象网络科技有限公司 | Face tracking method and face tracking equipment |
| CN112200841A (en) * | 2020-09-30 | 2021-01-08 | 杭州海宴科技有限公司 | Cross-domain multi-camera tracking method and device based on pedestrian posture |
| CN112287846A (en) * | 2020-10-30 | 2021-01-29 | 深圳市优必选科技股份有限公司 | Target person following method, device, mobile robot and readable storage medium |
| CN112894810A (en) * | 2021-01-19 | 2021-06-04 | 四川阿泰因机器人智能装备有限公司 | KCF algorithm-based mobile robot target loss prevention following method |
| CN113272752A (en) * | 2019-01-12 | 2021-08-17 | 株式会社Zmp | Autonomous traveling vehicle, distribution system, and program |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101339608A (en) * | 2008-08-15 | 2009-01-07 | 北京中星微电子有限公司 | Object tracking method and system based on detection |
| CN101783015A (en) * | 2009-01-19 | 2010-07-21 | 北京中星微电子有限公司 | Equipment and method for tracking video |
| CN102831378A (en) * | 2011-06-14 | 2012-12-19 | 株式会社理光 | Method and system for detecting and tracking human object |
| CN103064086A (en) * | 2012-11-04 | 2013-04-24 | 北京工业大学 | Vehicle tracking method based on depth information |
| CN103838258A (en) * | 2014-02-26 | 2014-06-04 | 上海微小卫星工程中心 | Automatic tracking method and system applied to space-based space target |
| CN103927508A (en) * | 2013-01-11 | 2014-07-16 | 浙江大华技术股份有限公司 | Target vehicle tracking method and device |
| JP2014149716A (en) * | 2013-02-01 | 2014-08-21 | Panasonic Corp | Object tracking apparatus and method thereof |
| CN105261035A (en) * | 2015-09-15 | 2016-01-20 | 杭州中威电子股份有限公司 | Method and device for tracking moving objects on highway |
| CN105825524A (en) * | 2016-03-10 | 2016-08-03 | 浙江生辉照明有限公司 | Target tracking method and apparatus |
-
2016
- 2016-08-29 CN CN201610752578.1A patent/CN106355603B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101339608A (en) * | 2008-08-15 | 2009-01-07 | 北京中星微电子有限公司 | Object tracking method and system based on detection |
| CN101783015A (en) * | 2009-01-19 | 2010-07-21 | 北京中星微电子有限公司 | Equipment and method for tracking video |
| CN102831378A (en) * | 2011-06-14 | 2012-12-19 | 株式会社理光 | Method and system for detecting and tracking human object |
| CN103064086A (en) * | 2012-11-04 | 2013-04-24 | 北京工业大学 | Vehicle tracking method based on depth information |
| CN103927508A (en) * | 2013-01-11 | 2014-07-16 | 浙江大华技术股份有限公司 | Target vehicle tracking method and device |
| JP2014149716A (en) * | 2013-02-01 | 2014-08-21 | Panasonic Corp | Object tracking apparatus and method thereof |
| CN103838258A (en) * | 2014-02-26 | 2014-06-04 | 上海微小卫星工程中心 | Automatic tracking method and system applied to space-based space target |
| CN105261035A (en) * | 2015-09-15 | 2016-01-20 | 杭州中威电子股份有限公司 | Method and device for tracking moving objects on highway |
| CN105825524A (en) * | 2016-03-10 | 2016-08-03 | 浙江生辉照明有限公司 | Target tracking method and apparatus |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273836A (en) * | 2017-06-07 | 2017-10-20 | 深圳市深网视界科技有限公司 | A kind of pedestrian detection recognition methods, device, model and medium |
| CN110363787A (en) * | 2018-03-26 | 2019-10-22 | 北京市商汤科技开发有限公司 | Information acquisition method and system, electronic equipment, program and medium |
| CN108762309A (en) * | 2018-05-03 | 2018-11-06 | 浙江工业大学 | It is a kind of based on the assumption that Kalman filtering human body target follower method |
| CN108762309B (en) * | 2018-05-03 | 2021-05-18 | 浙江工业大学 | Human body target following method based on hypothesis Kalman filtering |
| CN108921008A (en) * | 2018-05-14 | 2018-11-30 | 深圳市商汤科技有限公司 | Portrait identification method, device and electronic equipment |
| CN110675426A (en) * | 2018-07-02 | 2020-01-10 | 百度在线网络技术（北京）有限公司 | Human body tracking method, device, equipment and storage medium |
| US11348354B2 (en) | 2018-07-02 | 2022-05-31 | Baidu Online Network Technology (Beijing) Co., Ltd. | Human body tracing method, apparatus and device, and storage medium |
| CN110675426B (en) * | 2018-07-02 | 2022-11-22 | 百度在线网络技术（北京）有限公司 | Human body tracking method, device, equipment and storage medium |
| CN109145564A (en) * | 2018-08-22 | 2019-01-04 | 中国工商银行股份有限公司 | Control the method and device of mobile terminal |
| CN109271942A (en) * | 2018-09-26 | 2019-01-25 | 上海七牛信息技术有限公司 | A kind of stream of people's statistical method and system |
| CN113272752B (en) * | 2019-01-12 | 2023-11-17 | 株式会社Zmp | Autonomous traveling vehicle, distribution system, and program |
| CN113272752A (en) * | 2019-01-12 | 2021-08-17 | 株式会社Zmp | Autonomous traveling vehicle, distribution system, and program |
| CN110176025A (en) * | 2019-04-25 | 2019-08-27 | 沈阳航空航天大学 | A kind of proctor's tracking based on posture |
| CN110176025B (en) * | 2019-04-25 | 2021-06-18 | 沈阳图为科技有限公司 | Invigilator tracking method based on posture |
| CN110264498A (en) * | 2019-06-26 | 2019-09-20 | 北京深醒科技有限公司 | A kind of human body tracing method under video monitoring scene |
| CN110378931A (en) * | 2019-07-10 | 2019-10-25 | 成都数之联科技有限公司 | A kind of pedestrian target motion track acquisition methods and system based on multi-cam |
| CN111241995A (en) * | 2020-01-09 | 2020-06-05 | 苏宁云计算有限公司 | Loss compensation method and system for tracked object and computer equipment |
| CN111241995B (en) * | 2020-01-09 | 2022-12-30 | 苏宁云计算有限公司 | Loss compensation method and system for tracked object and computer equipment |
| CN111539283B (en) * | 2020-04-15 | 2023-08-11 | 上海摩象网络科技有限公司 | Face tracking method and face tracking equipment |
| CN111539283A (en) * | 2020-04-15 | 2020-08-14 | 上海摩象网络科技有限公司 | Face tracking method and face tracking equipment |
| CN112200841A (en) * | 2020-09-30 | 2021-01-08 | 杭州海宴科技有限公司 | Cross-domain multi-camera tracking method and device based on pedestrian posture |
| CN112287846A (en) * | 2020-10-30 | 2021-01-29 | 深圳市优必选科技股份有限公司 | Target person following method, device, mobile robot and readable storage medium |
| CN112894810A (en) * | 2021-01-19 | 2021-06-04 | 四川阿泰因机器人智能装备有限公司 | KCF algorithm-based mobile robot target loss prevention following method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106355603B (en) | 2019-10-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106355603A (en) | Method and device for human tracking | |
| CN110866480B (en) | Object tracking method and device, storage medium and electronic device | |
| CN106446815B (en) | A kind of simultaneous localization and mapping method | |
| TWI420906B (en) | Tracking system and method for regions of interest and computer program product thereof | |
| CN106447680B (en) | The object detecting and tracking method that radar is merged with vision under dynamic background environment | |
| CN105678288B (en) | Method for tracking target and device | |
| CN111210477B (en) | Method and system for positioning moving object | |
| CN109084732A (en) | Positioning and air navigation aid, device and processing equipment | |
| JP6624877B2 (en) | Information processing apparatus, information processing method and program | |
| CN108038415B (en) | Unmanned aerial vehicle automatic detection and tracking method based on machine vision | |
| JP5484184B2 (en) | Image processing apparatus, image processing method, and program | |
| CN104361327A (en) | Pedestrian detection method and system | |
| CN110553648A (en) | method and system for indoor navigation | |
| JP2009510541A (en) | Object tracking method and object tracking apparatus | |
| CN101883261A (en) | Method and system for abnormal target detection and relay tracking under large-range monitoring scene | |
| CN103778645B (en) | Circular target real-time tracking method based on images | |
| KR101787542B1 (en) | Estimation system and method of slope stability using 3d model and soil classification | |
| CN112880687A (en) | Indoor positioning method, device, equipment and computer readable storage medium | |
| CN108520559A (en) | A method of the unmanned plane location navigation based on binocular vision | |
| CN113066129A (en) | Visual positioning and mapping system based on target detection in dynamic environment | |
| CN113065397B (en) | Pedestrian detection method and device | |
| CN106530328A (en) | Method for detecting and smoothly following moving object based on video images | |
| Kang et al. | Persistent objects tracking across multiple non overlapping cameras | |
| CN105809108B (en) | Pedestrian's localization method and system based on distributed vision | |
| CN114898307A (en) | Object tracking method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |