CN104008371A - Regional suspicious target tracking and recognizing method based on multiple cameras - Google Patents
Regional suspicious target tracking and recognizing method based on multiple cameras Download PDFInfo
- Publication number
- CN104008371A CN104008371A CN201410220523.7A CN201410220523A CN104008371A CN 104008371 A CN104008371 A CN 104008371A CN 201410220523 A CN201410220523 A CN 201410220523A CN 104008371 A CN104008371 A CN 104008371A
- Authority
- CN
- China
- Prior art keywords
- target
- frame image
- detecting device
- current frame
- tracker
- 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
Landscapes
- Image Analysis (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The invention provides a regional suspicious target tracking and recognizing method based on multiple cameras. The method is used for realizing accurate recognition and tracking of a single suspicious target within a small region. The method comprises the steps of confirming the target to be tracked from one of the multiple cameras, conducting feature extraction on the target and judging the position and size of the target at the same time, and then updating the feature information of the target. Other cameras operate synchronously to detect the target in real time, a corresponding triggered target recognition unit recognizes and updates the target to be tracked once a triggering message sent by the first camera is received, then robust tracking is conducted on the suspicious target by means of a detector and a tracker, and then regional suspicious target tracking and recognition are realized under the condition of multiple cameras. The method is high in target tracking robustness under the condition of a single camera and high in target recognizing and tracking accuracy under the condition of multiple cameras, and therefore the method has good application prospects.
Description
Technical field
The invention belongs to target tracking domain, be specifically related to a kind of region suspicious object recognition and tracking method based on multiple-camera.
Background technology
Video tracking is a fundamental research problem of computer vision field, is also a very challenging research direction.In realities of the day life, video tracking technology is widely applied in various fields, comprising video monitoring, military engineering, traffic administration, intelligent robot and man-machine interaction etc., has very high academic research and using value.Most completes based on single camera about the research of video frequency object tracking.But there are a lot of insurmountable problems in the video frequency following system based on single camera, limited comprising target occlusion, camera coverage, can not carry out the problems such as omnibearing tracking, and tracker based on multiple-camera can be good at overcoming these problems.Therefore, the target following based on multiple-camera has become study hotspot.Obviously, no matter the target following technology based on multiple-camera is in academic research field, or all has very important significance in engineering application.
Current existing method for tracking target great majority, only for solving the target following of single camera, seldom can solve the problem that a plurality of video cameras are followed the tracks of same target simultaneously, therefore cannot accomplish good robustness.The Chinese patent that is CN102881024A as publication number discloses a kind of " video target tracking method based on TLD ", and the method only has good robustness based on TLD framework when single camera is followed the tracks of.This patent has proposed a kind of region suspicious object recognition and tracking method and framework thinking based on multiple-camera, the method can solve the accurate recognition and tracking of single suspicious object in small area, and designed framework clear thinking, be convenient to realize and expansion, method is simply accurate, real-time and practical.Target following under the method not only has good robustness in single camera situation, when multiple-camera target following, still can guarantee higher accuracy, therefore has good application prospect.
Summary of the invention
The technical problem to be solved in the present invention is: overcome prior art deficiency, propose a kind of region suspicious object recognition and tracking method based on multiple-camera, solve the problem of the accurate recognition and tracking of single suspicious object in small area.
The technical solution adopted for the present invention to solve the technical problems is:
A region suspicious object recognition and tracking method based on multiple-camera, comprises the steps:
In a video camera in multiple-camera, the suspicious object foreground picture of trigger alarm is extracted, carries out following steps:
Steps A, in the suspicious object foreground picture extracting, to this target, use boundary rectangle frame at the enterprising line display of video image, then use man-machine interaction mode to confirm rapidly the target that will follow the tracks of, and utilize the position, size of selected target and the characteristic information extracting, detecting device and tracker are carried out respectively to initialization, trigger target following and target detection, the target recognition unit to other video camera sends the message that triggers target identification simultaneously;
Step B, tracker, according to position and the size information of target in previous frame image, is estimated position and the size of target in current frame image; Meanwhile, use detecting device to detect current frame image, find out all possible target in current frame image;
Step C, merges the result of tracker and detecting device, judges position and size that whether current frame image exists target and target:
C-1, if there is no target, returns to step B and starts next frame image to detect;
C-2, if the target of existence, the position of target, size and corresponding characteristic information are preserved, and bring learning object into, by learning object, complete the online updating to detecting device, the movement destination image being now detected also wants interval to preserve, and then the detecting device of training and target image is passed to the target recognition unit of other video camera, then jump to step B, start the processing to next frame image;
For other video camera, with the aforementioned video camera synchronous operation that photographs suspicious object, specifically carry out following steps:
Step D, detects moving target: if target do not detected, continue to carry out to detect, if the target of detecting enters next step;
Step e, judges whether to receive the message of the triggering target identification described in steps A, if do not had, jumps to step D, otherwise triggers target recognition unit, enters next step;
Step F, by being delivered to the target image of this video camera target recognition unit, all isolated movement destination images in assessment current frame image, the result of the detecting device that fusion had been trained simultaneously to moving target in current frame image, judge current frame image and whether have the target that will follow the tracks of: if N continuous two field picture all detects target, judge that suspicious object occurs at this video camera, otherwise, judge that suspicious object does not occur, wherein N is natural number;
Step G, if suspicious object do not occur, return to the target that step F continues next frame separation of images to go out and identify; If suspicious object occurs, according to step C-2, starts next frame image to process.
Further, a kind of region suspicious object recognition and tracking method based on multiple-camera of the present invention, is to adopt background subtraction method to carry out target prospect figure extraction in described steps A.
Further, a kind of region suspicious object recognition and tracking method based on multiple-camera of the present invention, in described step B, tracker is according to position and the size information of target in previous frame image, estimates that the position of target in current frame image and big or small concrete steps are as follows:
B1, obtains object boundary frame according to target in the position of previous frame image and size information, and in object boundary frame extract minutiae;
B2, completes between adjacent two two field pictures the coupling of unique point in object boundary frame by optical flow method and follows the tracks of; When the accumulative displacement deviation of single unique point is greater than the threshold value of setting, think and deleted characteristic of correspondence point trail-and-error in target, otherwise retain;
B3, utilizes the unique point remaining to assess the object boundary frame of current frame image, thereby estimates position and the size of target in current frame image.
Further, a kind of region suspicious object recognition and tracking method based on multiple-camera of the present invention, is used detecting device to detect current frame image in described step B, finds out in current frame image the concrete steps of all possible target as follows:
A, in calculating current frame image, the variance yields of the object boundary frame of the scanning that is useful on, compares by the variance of all scan box and the variance of target frame, filters out the scan box that all variances than target frame are less than ratio threshold values;
B, define M basic classification device, for all scan box that are not filtered, all to carry out n kind change of scale, each basic classification device can carry out the comparison of pixel average in the position, m place of choosing at random in scan box after conversion, result relatively can produce a posterior probability, the mean value that calculates M all posterior probability of basic classification device, if mean value is greater than ratio threshold values, so now classifies as target by scan box; M, n, m are natural number;
C, is used support vector machines (Support Vector Machine) to classify to remaining scan box, if scan box is supported vector machine SVM, is divided into positive class, and this scan box is chosen as candidate target so.
Further, a kind of region suspicious object recognition and tracking method based on multiple-camera of the present invention, described step C merges the result of tracker and detecting device, judge current frame image whether exist the position of target and target and big or small concrete steps as follows:
C1, if tracker and detecting device do not trace into or detect target information, thinks in current frame image and does not exist target or target to lose;
C2, if tracker has tracking results, and detecting device does not detect target, and the degree of confidence of the target tracing into is while being greater than a certain threshold value, the target using the result of tracker as current frame image;
C3, if tracker does not trace into target, and detecting device detects target, and the degree of confidence of the target detecting is while being greater than a certain threshold value, the target using the result of detecting device as current frame image;
C4, if when tracker and detecting device all detect target, is divided into two kinds of situations: if the target that a tracker and detecting device trace into is complete when overlapping, think that this target is the target that current frame image will be followed the tracks of; If the target that two trackers and detecting device obtain is not exclusively overlapping, calculate the degree of confidence of detecting device and tracker, if the degree of confidence of detecting device is greater than the degree of confidence of tracker and overlapping area while being less than certain threshold value, the target using the result of detecting device as present frame, otherwise, the target using the testing result of tracker as present frame.
Further, a kind of region suspicious object recognition and tracking method based on multiple-camera of the present invention, in described step C-2, the concrete steps of detecting device online updating are as follows:
101, in learning object, if detecting device is non-target by last definite object judgement in step C, is changed into target, and added in training set and upgrade;
102, in learning object, if detecting device has detected a plurality of targets, comprise the target that will follow the tracks of, according to target, in every two field picture, only have the constraint condition of one, change other target outside target definite in step C into non-target, and add in training set and upgrade;
103, in learning object, if target recognition unit has identified target, added in training set and upgraded;
104, utilize the training set having upgraded to be updated in line target detection model, i.e. the online detecting device of Training Support Vector Machines SVM to obtain training again.
Further, a kind of region suspicious object recognition and tracking method based on multiple-camera of the present invention, the method in described step D, moving target being detected is background subtraction method.
Further, a kind of region suspicious object recognition and tracking method based on multiple-camera of the present invention, the concrete steps of described step F are as follows:
F1, extracts SURF (Speeded Up Robust Features accelerates robust features) feature and calculates color histogram foreground target image being detected in step D;
F2, also extracts SURF feature and calculates color histogram being delivered to all real goal images of this video camera;
F3, by the SURF feature of extracting on foreground target image and real goal image, compare between two, find out the some to angle point of mutual coupling, set up the corresponding relation between target, assess matching degree between the two simultaneously, and the target in current frame image and the matching degree that is delivered to all real goals of this video camera are carried out to statistical study;
F4, the target in the color histogram graph evaluation current frame image calculating by step F 2 be delivered to the matching degree of all real goals of this video camera, and matching degree is carried out to statistical study;
F5, by step b and c calculate target in current frame image and with the matching degree that is delivered to all real goals of this video camera, and matching degree is also carried out to statistical study;
F6, if the matching degree in step F 3 to F5 is all greater than certain proportion, thinks and the target that will follow the tracks of detected, otherwise, this target is lost;
F7, if when being consecutively detected the frame number of target and surpassing N, illustrate that suspicious object occurs at this video camera, otherwise suspicious object does not occur.
The technical solution used in the present invention compared with prior art, has following technique effect:
The present invention proposes a kind of region suspicious object recognition and tracking method based on multiple-camera, adopt the target following of described method to be no longer confined within sweep of the eye target be followed the tracks of at single camera, but can between a plurality of video cameras, combine target is carried out to recognition and tracking, thereby greatly extended the field range of video camera, this security protection for a lot of important places and monitoring have important using value.The target following of described method not only has good robustness in single camera situation, when multiple-camera target recognition and tracking, still can guarantee higher accuracy, therefore has good application prospect.
Accompanying drawing explanation
Fig. 1 is the region suspicious object recognition and tracking method flow diagram based on multiple-camera.
Embodiment
Clearer for technical matters, technical scheme and technique effect that the region suspicious object recognition and tracking method based on multiple-camera of the present invention will be solved, below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.
Fig. 1 is the region suspicious object recognition and tracking method flow diagram based on multiple-camera.
A region suspicious object recognition and tracking method based on multiple-camera, comprises the steps:
Steps A, a video camera in multiple-camera (is assumed to video camera one, but be not defined as a certain particular camera, between multiple-camera, can there be overlapping region visual field, also can zero lap region, but the distance between each video camera is unsuitable excessive, thereby be called regional target tracking identification) inner, adopt background subtraction method, the suspicious object foreground picture of trigger alarm is extracted and use boundary rectangle frame at the enterprising line display of video image to this target, then use the man-machine interaction modes such as keyboard or mouse click to confirm rapidly the target that will follow the tracks of, and utilize the position of selected target, size and the characteristic information extracting, detecting device and tracker are carried out respectively to initialization, trigger target following and target detection, notify other video camera can trigger target recognition unit simultaneously,
Step B, tracker, according to position and the size information of target in previous frame image, is estimated position and the size of target in current frame image; Meanwhile, use detecting device to detect current frame image, find out all possible target in current frame image;
Step C, merges the result of tracker and detecting device, judges position and size that whether current frame image exists target and target:
C-1, if there is no target, returns to step B and starts next frame image to detect;
C-2, if the target of existence, the position of target, size and corresponding characteristic information are preserved, and bring learning object into, by learning object, complete the online updating to detecting device, the movement destination image being now detected also wants interval to preserve, then the detecting device of training and target image are passed to other video camera (in order to guarantee the precision of identification, here require model and the parameter of each video camera consistent as far as possible) target recognition unit, then jump to step B, start the processing to next frame image;
Step D, other video camera and first video camera synchronous operation, detect moving target: if target do not detected, continue to carry out to detect, if the target of detecting enters next step;
Step e, judges whether to receive the message of the triggering target identification that first video camera sends, if do not had, jumps to step D, otherwise triggers target recognition unit, enters next step;
Step F, by being delivered to the target image of this video camera target recognition unit, all isolated movement destination images in assessment current frame image, the result of the detecting device that fusion had been trained simultaneously to moving target in current frame image, judge current frame image and whether have the target that will follow the tracks of: if (N represents N continuous if N continuous two field picture all detects target, illustrate that suspicious object appears at the possibility of this video camera large, thereby prevent that the mistake identification producing due to indivedual two field pictures from causing mistake identification and the tracking error of whole algorithm, N is the value of artificially setting as the case may be, can test and provide by reality) two field picture all detects target, judge that suspicious object occurs at this video camera, otherwise, judge that suspicious object does not occur, wherein N is constant,
Step G, if suspicious object do not occur, return to the target that step F continues next frame separation of images to go out and identify; If suspicious object occurs, according to step C-2, starts next frame image to process.
A region suspicious object recognition and tracking method based on multiple-camera, in described step B, tracker, according to position and the size information of target in previous frame image, estimates that the position of target in current frame image and big or small concrete steps are as follows:
B1, obtains object boundary frame according to target in the position of previous frame image and size information, and in object boundary frame extract minutiae;
B2, completes between adjacent two two field pictures the coupling of unique point in object boundary frame by optical flow method and follows the tracks of; When the accumulative displacement deviation of single unique point is greater than the threshold value of setting, think and deleted characteristic of correspondence point trail-and-error in target, otherwise retain;
Optical flow method comes from visually-perceptible principle, and the motion spatially of objective object is generally relatively continuous, and in motion process, its image that projects in sensor plane should be also continually varying, i.e. gray scale unchangeability hypothesis.According to this basic assumption, can obtain light stream Basic Constraint Equation.
If the point (x, y) on image is I (x, y, t) in the gray scale of moment t, establishing light stream w (u, v) is u (x, y) and v (x, y) at the horizontal and vertical mobile component of this point:
After super-interval dt, the gray scale of corresponding point is I (x+dx, y+dy, t+dt), when dt → 0, gray scale I remains unchanged, and obtains I (x, y, t)=I (x+dx, y+dy, t+dt), this formula is launched by Taylor, ignore second order infinitesimal, arrange the Basic Constraint Equation (optical flow constraint equation) that obtains light stream:
The Basic Constraint Equation of light stream can be noted by abridging and is:
I
x·u+I
y·v+I
t＝0 (3-3)
Light stream Basic Constraint Equation represents that pixel grey scale equals the spatial gradient of gray scale and the dot product of light stream speed to the rate of change of time.In order to solve the Basic Constraint Equation of light stream, must give formula (3-3) additional other constraint conditions, different constraint condition produces different optical flow analysis methods.
LK optical flow method is a kind of conventional and effective local parameter light stream method of estimation based on gradient.In the neighborhood that this algorithm supposition is Ω a bulk, light stream vector is constant, then uses weighted least-squares method to estimate light stream.
On a little spatial neighborhood Ω, light stream evaluated error is defined as:
W in formula
2(x) represent window weighting function, the impact that it makes centre of neighbourhood region produce constraint is larger than outer peripheral areas, often adopts Gaussian function.
Suppose IS=(I
t, I
t+1..., I
t+N) be one group of image sequence, x
tbe the corresponding position of t unique point constantly, use LK optical flow method to x
tcarry out forward direction and follow the tracks of N two field picture, resulting feature point tracking track is
here f represents forward direction tracking, and the length that N is pursuit path, in like manner, carries out backward tracking N two field picture, and resulting feature point tracking track is
b represents backward tracking,
the accumulative displacement deviation of so single unique point may be defined as
when the accumulative displacement deviation TE of single unique point (TE is the artificial value of setting as the case may be, can test and be provided by reality) is greater than the threshold value of setting, think and deleted characteristic of correspondence point trail-and-error in target, otherwise reservation;
B3, utilizes the unique point remaining to assess the object boundary frame of current frame image, thereby estimates position and the size of target in current frame image.
A region suspicious object recognition and tracking method based on multiple-camera, is used detecting device to detect current frame image in described step B, finds out in current frame image the concrete steps of all possible target as follows:
A, calculate the variance yields of the object boundary frame (abbreviation scan box) of the scanning that is useful in current frame image, by the variance of all scan box and the variance of target frame, compare, filter out the scan box that all variances than target frame are less than ratio threshold values, this ratio threshold values is depending on experience, for example be set to 30%, 40%, 50%;
B, define M basic classification device, for all scan box that are not filtered, all to carry out n kind change of scale, each basic classification device can carry out the comparison of pixel average in the position, m place of choosing at random in scan box after conversion, result relatively can produce a m dimension dual code X, then by index, compare and normalization with the dual code of wanting tracking target simultaneously, just can produce a posterior probability P
i(Y|X), Y ∈ { 0,1}, 1≤i≤M wherein, the mean value that calculates M basic classification device posterior probability, if mean value is greater than ratio threshold values, so now classifies as target by scan box, and this ratio threshold values is depending on experience, for example can be set to 50%, 60%, 65%;
C, is used support vector machines to classify to remaining scan box, if scan box is supported vector machine SVM, is divided into positive class, and this scan box is chosen as candidate target so.
A region suspicious object recognition and tracking method based on multiple-camera, described step C merges the result of tracker and detecting device, judge current frame image whether exist the position of target and target and big or small concrete steps as follows:
C1, if tracker and detecting device do not trace into or detect target information, thinks in current frame image and does not exist target or target to lose;
C2, if tracker has tracking results, and detecting device does not detect target, and the degree of confidence of the target tracing into is while being greater than a certain threshold value, the target using the result of tracker as current frame image;
C3, if tracker does not trace into target, and detecting device detects target, and the degree of confidence of the target detecting is while being greater than a certain threshold value, the target using the result of detecting device as current frame image;
C4, if when tracker and detecting device all detect target, is divided into two kinds of situations: if the target that a tracker and detecting device trace into is complete when overlapping, think that this target is the target that current frame image will be followed the tracks of; If the target that two trackers and detecting device obtain is not exclusively overlapping, calculate the degree of confidence of detecting device and tracker, if the degree of confidence of detecting device is greater than the degree of confidence of tracker and overlapping area while being less than certain threshold value, the target using the result of detecting device as present frame, otherwise, the target using the testing result of tracker as present frame.
A region suspicious object recognition and tracking method based on multiple-camera, in described step C-2, the concrete steps of detecting device online updating are as follows:
101, in learning object, if detecting device is non-target by last definite object judgement in step C, is changed into target, and added in training set and upgrade;
102, in learning object, if detecting device has detected a plurality of targets, comprise the target that will follow the tracks of, according to target, in every two field picture, only have the constraint condition of one, change other target outside target definite in step C into non-target, and add in training set and upgrade;
103, in learning object, if target recognition unit has identified target, added in training set and upgraded;
104, utilize the training set having upgraded to be updated in line target detection model, i.e. the online detecting device of Training Support Vector Machines SVM to obtain training again.
A region suspicious object recognition and tracking method based on multiple-camera, the concrete steps of described step F are as follows:
F1, extracts SURF feature and calculates color histogram foreground target image being detected in step D;
F2, also extracts SURF feature and calculates color histogram being delivered to all real goal images of this video camera;
F3, by the SURF feature of extracting on foreground target image and real goal image, compare between two, find out the some to angle point of mutual coupling, set up the corresponding relation between target, assess matching degree between the two simultaneously, and the target in current frame image and the matching degree that is delivered to all real goals of this video camera are carried out to statistical study;
F4, the target in the color histogram graph evaluation current frame image calculating by step F 2 be delivered to the matching degree of all real goals of this video camera, and matching degree is carried out to statistical study;
F5, by step b and c calculate target in current frame image and with the matching degree that is delivered to all real goals of this video camera, and matching degree is also carried out to statistical study;
F6, if the matching degree in step F 3 to F5 is all greater than certain proportion, thinks and the target that will follow the tracks of detected, otherwise, this target is lost;
F7, if when being consecutively detected the frame number of target and surpassing N, illustrate that suspicious object occurs at this video camera, otherwise suspicious object does not occur.
Obviously, it will be appreciated by those skilled in the art that a kind of disclosed region suspicious object recognition and tracking method based on multiple-camera of the invention described above, can also on the basis that does not depart from content of the present invention, make various improvement.Therefore, protection scope of the present invention should be determined by the content of appending claims.
Claims (8)
1. the region suspicious object recognition and tracking method based on multiple-camera, is characterized in that,
In a video camera in multiple-camera, the suspicious object foreground picture of trigger alarm is extracted, carries out following steps:
Steps A, in the suspicious object foreground picture extracting, to this target, use boundary rectangle frame at the enterprising line display of video image, then use man-machine interaction mode to confirm rapidly the target that will follow the tracks of, and utilize the position, size of selected target and the characteristic information extracting, detecting device and tracker are carried out respectively to initialization, trigger target following and target detection, the target recognition unit to other video camera sends the message that triggers target identification simultaneously;
Step B, tracker, according to position and the size information of target in previous frame image, is estimated position and the size of target in current frame image; Meanwhile, use detecting device to detect current frame image, find out all possible target in current frame image;
Step C, merges the result of tracker and detecting device, judges position and size that whether current frame image exists target and target:
C-1, if there is no target, returns to step B and starts next frame image to detect;
C-2, if the target of existence, the position of target, size and corresponding characteristic information are preserved, and bring learning object into, by learning object, complete the online updating to detecting device, the movement destination image being now detected also wants interval to preserve, and then the detecting device of training and target image is passed to the target recognition unit of other video camera, then jump to step B, start the processing to next frame image;
For other video camera, with the aforementioned video camera synchronous operation that photographs suspicious object, specifically carry out following steps:
Step D, detects moving target: if target do not detected, continue to carry out to detect, if the target of detecting enters next step;
Step e, judges whether to receive the message of the triggering target identification described in steps A, if do not had, jumps to step D, otherwise triggers target recognition unit, enters next step;
Step F, by being delivered to the target image of this video camera target recognition unit, all isolated movement destination images in assessment current frame image, the result of the detecting device that fusion had been trained simultaneously to moving target in current frame image, judge current frame image and whether have the target that will follow the tracks of: if N continuous two field picture all detects target, judge that suspicious object occurs at this video camera, otherwise, judge that suspicious object does not occur, wherein N is natural number;
Step G, if suspicious object do not occur, return to the target that step F continues next frame separation of images to go out and identify; If suspicious object occurs, according to step C-2, starts next frame image to process.
2. a kind of region suspicious object recognition and tracking method based on multiple-camera as claimed in claim 1, is characterized in that, is to adopt background subtraction method to carry out target prospect figure extraction in described steps A.
3. a kind of region suspicious object recognition and tracking method based on multiple-camera as claimed in claim 1, it is characterized in that, in described step B, tracker is according to position and the size information of target in previous frame image, estimates that the position of target in current frame image and big or small concrete steps are as follows:
B1, obtains object boundary frame according to target in the position of previous frame image and size information, and in object boundary frame extract minutiae;
B2, completes between adjacent two two field pictures the coupling of unique point in object boundary frame by optical flow method and follows the tracks of; When the accumulative displacement deviation of single unique point is greater than the threshold value of setting, think and deleted characteristic of correspondence point trail-and-error in target, otherwise retain;
B3, utilizes the unique point remaining to assess the object boundary frame of current frame image, thereby estimates position and the size of target in current frame image.
4. a kind of region suspicious object recognition and tracking method based on multiple-camera as claimed in claim 1, it is characterized in that, in described step B, use detecting device to detect current frame image, find out in current frame image the concrete steps of all possible target as follows:
A, in calculating current frame image, the variance yields of the object boundary frame of the scanning that is useful on, compares by the variance of all scan box and the variance of target frame, filters out the scan box that all variances than target frame are less than ratio threshold values;
B, define M basic classification device, for all scan box that are not filtered, all to carry out n kind change of scale, each basic classification device can carry out the comparison of pixel average in the position, m place of choosing at random in scan box after conversion, result relatively can produce a posterior probability, the mean value that calculates M all posterior probability of basic classification device, if mean value is greater than ratio threshold values, so now classifies as target by scan box; M, n, m are natural number;
C, is used support vector machines to classify to remaining scan box, if scan box is supported vector machine SVM, is divided into positive class, and this scan box is chosen as candidate target so.
5. a kind of region suspicious object recognition and tracking method based on multiple-camera as claimed in claim 1, it is characterized in that, described step C merges the result of tracker and detecting device, judge current frame image whether exist the position of target and target and big or small concrete steps as follows:
C1, if tracker and detecting device do not trace into or detect target information, thinks in current frame image and does not exist target or target to lose;
C2, if tracker has tracking results, and detecting device does not detect target, and the degree of confidence of the target tracing into is while being greater than a certain threshold value, the target using the result of tracker as current frame image;
C3, if tracker does not trace into target, and detecting device detects target, and the degree of confidence of the target detecting is while being greater than a certain threshold value, the target using the result of detecting device as current frame image;
C4, if when tracker and detecting device all detect target, is divided into two kinds of situations: if the target that a tracker and detecting device trace into is complete when overlapping, think that this target is the target that current frame image will be followed the tracks of; If the target that two trackers and detecting device obtain is not exclusively overlapping, calculate the degree of confidence of detecting device and tracker, if the degree of confidence of detecting device is greater than the degree of confidence of tracker and overlapping area while being less than certain threshold value, the target using the result of detecting device as present frame, otherwise, the target using the testing result of tracker as present frame.
6. a kind of region suspicious object recognition and tracking method based on multiple-camera as claimed in claim 1, is characterized in that, in described step C-2, the concrete steps of detecting device online updating are as follows:
101, in learning object, if detecting device is non-target by last definite object judgement in step C, is changed into target, and added in training set and upgrade;
102, in learning object, if detecting device has detected a plurality of targets, comprise the target that will follow the tracks of, according to target, in every two field picture, only have the constraint condition of one, change other target outside target definite in step C into non-target, and add in training set and upgrade;
103, in learning object, if target recognition unit has identified target, added in training set and upgraded;
104, utilize the training set having upgraded to be updated in line target detection model, i.e. the online detecting device of Training Support Vector Machines SVM to obtain training again.
7. a kind of region suspicious object recognition and tracking method based on multiple-camera as claimed in claim 1, is characterized in that, the method in described step D, moving target being detected is background subtraction method.
8. a kind of region suspicious object recognition and tracking method based on multiple-camera as claimed in claim 4, is characterized in that, the concrete steps of described step F are as follows:
F1, extracts SURF feature and calculates color histogram foreground target image being detected in step D;
F2, also extracts SURF feature and calculates color histogram being delivered to all real goal images of this video camera;
F3, by the SURF feature of extracting on foreground target image and real goal image, compare between two, find out the some to angle point of mutual coupling, set up the corresponding relation between target, assess matching degree between the two simultaneously, and the target in current frame image and the matching degree that is delivered to all real goals of this video camera are carried out to statistical study;
F4, the target in the color histogram graph evaluation current frame image calculating by step F 2 be delivered to the matching degree of all real goals of this video camera, and matching degree is carried out to statistical study;
F5, by step b and c calculate target in current frame image and with the matching degree that is delivered to all real goals of this video camera, and matching degree is also carried out to statistical study;
F6, if the matching degree in step F 3 to F5 is all greater than certain proportion, thinks and the target that will follow the tracks of detected, otherwise, this target is lost;
F7, if when being consecutively detected the frame number of target and surpassing N, illustrate that suspicious object occurs at this video camera, otherwise suspicious object does not occur.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410220523.7A CN104008371B (en) | 2014-05-22 | 2014-05-22 | Regional suspicious target tracking and recognizing method based on multiple cameras |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410220523.7A CN104008371B (en) | 2014-05-22 | 2014-05-22 | Regional suspicious target tracking and recognizing method based on multiple cameras |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104008371A true CN104008371A (en) | 2014-08-27 |
| CN104008371B CN104008371B (en) | 2017-02-15 |
Family
ID=51369020
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410220523.7A Active CN104008371B (en) | 2014-05-22 | 2014-05-22 | Regional suspicious target tracking and recognizing method based on multiple cameras |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104008371B (en) |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104537688A (en) * | 2014-12-24 | 2015-04-22 | 南京邮电大学 | Moving object detecting method based on background subtraction and HOG features |
| CN104992453A (en) * | 2015-07-14 | 2015-10-21 | 国家电网公司 | Target tracking method under complicated background based on extreme learning machine |
| CN105812746A (en) * | 2016-04-21 | 2016-07-27 | 北京格灵深瞳信息技术有限公司 | Target detection method and system |
| CN106571014A (en) * | 2016-10-24 | 2017-04-19 | 上海伟赛智能科技有限公司 | Method for identifying abnormal motion in video and system thereof |
| CN106707296A (en) * | 2017-01-09 | 2017-05-24 | 华中科技大学 | Dual-aperture photoelectric imaging system-based unmanned aerial vehicle detection and recognition method |
| CN106960446A (en) * | 2017-04-01 | 2017-07-18 | 广东华中科技大学工业技术研究院 | A kind of waterborne target detecting and tracking integral method applied towards unmanned boat |
| CN107066990A (en) * | 2017-05-04 | 2017-08-18 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
| CN107094223A (en) * | 2017-03-17 | 2017-08-25 | 奥克斯空调股份有限公司 | Obtain the method and device of the high-definition image of intrusion target |
| CN107277460A (en) * | 2017-07-29 | 2017-10-20 | 安徽博威康信息技术有限公司 | A kind of intellectual monitoring alarm method of identity-based identification |
| CN107277461A (en) * | 2017-07-29 | 2017-10-20 | 安徽博威康信息技术有限公司 | A kind of intellectual monitoring alarm method based on fixation and recognition |
| CN107705324A (en) * | 2017-10-20 | 2018-02-16 | 中山大学 | A kind of video object detection method based on machine learning |
| CN108038408A (en) * | 2017-10-13 | 2018-05-15 | 西安天和防务技术股份有限公司 | Intelligent-tracking method and device, storage medium and electric terminal |
| CN108038837A (en) * | 2017-12-08 | 2018-05-15 | 苏州科达科技股份有限公司 | Object detection method and system in video |
| CN108154098A (en) * | 2017-12-20 | 2018-06-12 | 歌尔股份有限公司 | A kind of target identification method of robot, device and robot |
| CN108776481A (en) * | 2018-06-20 | 2018-11-09 | 北京智行者科技有限公司 | A kind of parallel driving control method |
| CN106023650B (en) * | 2016-07-01 | 2018-11-30 | 南京邮电大学 | Real-time pedestrian's method for early warning based on traffic intersection video and computer parallel processing system |
| CN109033922A (en) * | 2017-06-08 | 2018-12-18 | 北京君正集成电路股份有限公司 | The method and device of target identification is set in a kind of video |
| CN109074473A (en) * | 2016-04-11 | 2018-12-21 | 北京市商汤科技开发有限公司 | For the method and system to image tracing |
| CN109118523A (en) * | 2018-09-20 | 2019-01-01 | 电子科技大学 | A kind of tracking image target method based on YOLO |
| CN109344792A (en) * | 2018-10-18 | 2019-02-15 | 电子科技大学 | A kind of Motion parameters tracking |
| CN109598885A (en) * | 2018-12-21 | 2019-04-09 | 广东中安金狮科创有限公司 | Monitoring system and its alarm method |
| CN109791615A (en) * | 2017-05-05 | 2019-05-21 | 京东方科技集团股份有限公司 | For detecting and tracking the method, target object tracking equipment and computer program product of target object |
| CN110059578A (en) * | 2019-03-27 | 2019-07-26 | 东软睿驰汽车技术（沈阳）有限公司 | A kind of method and device of vehicle tracking |
| CN110798653A (en) * | 2018-08-01 | 2020-02-14 | 华为技术有限公司 | Image processing method and video monitoring system based on multi-machine cooperation |
| CN111242974A (en) * | 2020-01-07 | 2020-06-05 | 重庆邮电大学 | Vehicle real-time tracking method based on twin network and back propagation |
| CN111627046A (en) * | 2020-05-15 | 2020-09-04 | 北京百度网讯科技有限公司 | Target part tracking method and device, electronic equipment and readable storage medium |
| CN111798643A (en) * | 2020-05-25 | 2020-10-20 | 中国原子能科学研究院 | Novel global critical accident alarm method and system |
| CN113542671A (en) * | 2020-04-21 | 2021-10-22 | 株式会社日立制作所 | Event analysis system and event analysis method |
| CN114253253A (en) * | 2020-09-24 | 2022-03-29 | 科沃斯商用机器人有限公司 | Target identification method and device based on artificial intelligence and robot |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040100563A1 (en) * | 2002-11-27 | 2004-05-27 | Sezai Sablak | Video tracking system and method |
| CN101751677A (en) * | 2008-12-17 | 2010-06-23 | 中国科学院自动化研究所 | Target continuous tracking method based on multi-camera |
| CN102289822A (en) * | 2011-09-09 | 2011-12-21 | 南京大学 | Method for tracking moving target collaboratively by multiple cameras |
| CN102629385A (en) * | 2012-02-28 | 2012-08-08 | 中山大学 | Object matching and tracking system based on multiple camera information fusion and method thereof |
-
2014
- 2014-05-22 CN CN201410220523.7A patent/CN104008371B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040100563A1 (en) * | 2002-11-27 | 2004-05-27 | Sezai Sablak | Video tracking system and method |
| CN101751677A (en) * | 2008-12-17 | 2010-06-23 | 中国科学院自动化研究所 | Target continuous tracking method based on multi-camera |
| CN102289822A (en) * | 2011-09-09 | 2011-12-21 | 南京大学 | Method for tracking moving target collaboratively by multiple cameras |
| CN102629385A (en) * | 2012-02-28 | 2012-08-08 | 中山大学 | Object matching and tracking system based on multiple camera information fusion and method thereof |
Non-Patent Citations (2)
| Title |
|---|
| 明安龙 等: ""多摄像机之间基于区域SIFT描述子的目标匹配"", 《计算机学报》 * |
| 衡林: ""多摄像机视域中的运动目标检测与跟踪研究"", 《中国优秀硕士学位论文全文数据库·信息科技辑》 * |
Cited By (39)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104537688A (en) * | 2014-12-24 | 2015-04-22 | 南京邮电大学 | Moving object detecting method based on background subtraction and HOG features |
| CN104992453A (en) * | 2015-07-14 | 2015-10-21 | 国家电网公司 | Target tracking method under complicated background based on extreme learning machine |
| CN104992453B (en) * | 2015-07-14 | 2018-10-23 | 国家电网公司 | Target in complex environment tracking based on extreme learning machine |
| CN109074473A (en) * | 2016-04-11 | 2018-12-21 | 北京市商汤科技开发有限公司 | For the method and system to image tracing |
| CN105812746A (en) * | 2016-04-21 | 2016-07-27 | 北京格灵深瞳信息技术有限公司 | Target detection method and system |
| CN105812746B (en) * | 2016-04-21 | 2019-05-10 | 北京格灵深瞳信息技术有限公司 | A kind of object detection method and system |
| CN106023650B (en) * | 2016-07-01 | 2018-11-30 | 南京邮电大学 | Real-time pedestrian's method for early warning based on traffic intersection video and computer parallel processing system |
| CN106571014A (en) * | 2016-10-24 | 2017-04-19 | 上海伟赛智能科技有限公司 | Method for identifying abnormal motion in video and system thereof |
| CN106707296A (en) * | 2017-01-09 | 2017-05-24 | 华中科技大学 | Dual-aperture photoelectric imaging system-based unmanned aerial vehicle detection and recognition method |
| CN106707296B (en) * | 2017-01-09 | 2019-03-05 | 华中科技大学 | It is a kind of based on the unmanned machine testing of Based on Dual-Aperture photo electric imaging system and recognition methods |
| CN107094223A (en) * | 2017-03-17 | 2017-08-25 | 奥克斯空调股份有限公司 | Obtain the method and device of the high-definition image of intrusion target |
| CN106960446A (en) * | 2017-04-01 | 2017-07-18 | 广东华中科技大学工业技术研究院 | A kind of waterborne target detecting and tracking integral method applied towards unmanned boat |
| CN107066990B (en) * | 2017-05-04 | 2019-10-11 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
| CN107066990A (en) * | 2017-05-04 | 2017-08-18 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
| CN109791615B (en) * | 2017-05-05 | 2023-07-28 | 京东方科技集团股份有限公司 | Method for detecting and tracking a target object, target object tracking device and computer program product |
| CN109791615A (en) * | 2017-05-05 | 2019-05-21 | 京东方科技集团股份有限公司 | For detecting and tracking the method, target object tracking equipment and computer program product of target object |
| CN109033922A (en) * | 2017-06-08 | 2018-12-18 | 北京君正集成电路股份有限公司 | The method and device of target identification is set in a kind of video |
| CN107277461A (en) * | 2017-07-29 | 2017-10-20 | 安徽博威康信息技术有限公司 | A kind of intellectual monitoring alarm method based on fixation and recognition |
| CN107277460A (en) * | 2017-07-29 | 2017-10-20 | 安徽博威康信息技术有限公司 | A kind of intellectual monitoring alarm method of identity-based identification |
| CN108038408A (en) * | 2017-10-13 | 2018-05-15 | 西安天和防务技术股份有限公司 | Intelligent-tracking method and device, storage medium and electric terminal |
| CN107705324A (en) * | 2017-10-20 | 2018-02-16 | 中山大学 | A kind of video object detection method based on machine learning |
| CN108038837A (en) * | 2017-12-08 | 2018-05-15 | 苏州科达科技股份有限公司 | Object detection method and system in video |
| CN108038837B (en) * | 2017-12-08 | 2020-09-29 | 苏州科达科技股份有限公司 | Method and system for detecting target in video |
| CN108154098A (en) * | 2017-12-20 | 2018-06-12 | 歌尔股份有限公司 | A kind of target identification method of robot, device and robot |
| CN108776481A (en) * | 2018-06-20 | 2018-11-09 | 北京智行者科技有限公司 | A kind of parallel driving control method |
| CN110798653A (en) * | 2018-08-01 | 2020-02-14 | 华为技术有限公司 | Image processing method and video monitoring system based on multi-machine cooperation |
| US11445150B2 (en) | 2018-08-01 | 2022-09-13 | Huawei Technologies Co., Ltd. | Multi-camera collaboration-based image processing method and video surveillance system |
| CN110798653B (en) * | 2018-08-01 | 2021-09-14 | 华为技术有限公司 | Image processing method and video monitoring system based on multi-machine cooperation |
| CN109118523B (en) * | 2018-09-20 | 2022-04-22 | 电子科技大学 | Image target tracking method based on YOLO |
| CN109118523A (en) * | 2018-09-20 | 2019-01-01 | 电子科技大学 | A kind of tracking image target method based on YOLO |
| CN109344792A (en) * | 2018-10-18 | 2019-02-15 | 电子科技大学 | A kind of Motion parameters tracking |
| CN109598885A (en) * | 2018-12-21 | 2019-04-09 | 广东中安金狮科创有限公司 | Monitoring system and its alarm method |
| CN110059578A (en) * | 2019-03-27 | 2019-07-26 | 东软睿驰汽车技术（沈阳）有限公司 | A kind of method and device of vehicle tracking |
| CN111242974A (en) * | 2020-01-07 | 2020-06-05 | 重庆邮电大学 | Vehicle real-time tracking method based on twin network and back propagation |
| CN111242974B (en) * | 2020-01-07 | 2023-04-11 | 重庆邮电大学 | Vehicle real-time tracking method based on twin network and back propagation |
| CN113542671A (en) * | 2020-04-21 | 2021-10-22 | 株式会社日立制作所 | Event analysis system and event analysis method |
| CN111627046A (en) * | 2020-05-15 | 2020-09-04 | 北京百度网讯科技有限公司 | Target part tracking method and device, electronic equipment and readable storage medium |
| CN111798643A (en) * | 2020-05-25 | 2020-10-20 | 中国原子能科学研究院 | Novel global critical accident alarm method and system |
| CN114253253A (en) * | 2020-09-24 | 2022-03-29 | 科沃斯商用机器人有限公司 | Target identification method and device based on artificial intelligence and robot |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104008371B (en) | 2017-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104008371A (en) | Regional suspicious target tracking and recognizing method based on multiple cameras | |
| Mukojima et al. | Moving camera background-subtraction for obstacle detection on railway tracks | |
| Keller et al. | The benefits of dense stereo for pedestrian detection | |
| Bondi et al. | Real-time people counting from depth imagery of crowded environments | |
| CN106447680B (en) | The object detecting and tracking method that radar is merged with vision under dynamic background environment | |
| CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
| CN100589561C (en) | Dubious static object detecting method based on video content analysis | |
| US20180365843A1 (en) | Method and system for tracking moving objects based on optical flow method | |
| CN103279791B (en) | Based on pedestrian's computing method of multiple features | |
| Bloisi et al. | Argos—A video surveillance system for boat traffic monitoring in Venice | |
| CN106127807A (en) | A kind of real-time video multiclass multi-object tracking method | |
| CN107301375B (en) | Video image smoke detection method based on dense optical flow | |
| CN109697420A (en) | A kind of Moving target detection and tracking towards urban transportation | |
| CN105654516B (en) | Satellite image based on target conspicuousness is to ground weak moving target detection method | |
| CN108776974B (en) | A kind of real-time modeling method method suitable for public transport scene | |
| CN103942536B (en) | Multi-target tracking method of iteration updating track model | |
| CN103971386A (en) | Method for foreground detection in dynamic background scenario | |
| CN104424638A (en) | Target tracking method based on shielding situation | |
| KR20150049529A (en) | Apparatus and method for estimating the location of the vehicle | |
| CN103336947A (en) | Method for identifying infrared movement small target based on significance and structure | |
| Wang et al. | A roadside camera-radar sensing fusion system for intelligent transportation | |
| Ryoo et al. | Observe-and-explain: A new approach for multiple hypotheses tracking of humans and objects | |
| Qing et al. | A novel particle filter implementation for a multiple-vehicle detection and tracking system using tail light segmentation | |
| CN101719280A (en) | Method for detecting petty infrared target | |
| CN112329671B (en) | Pedestrian running behavior detection method based on deep learning and related components |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20140827 Assignee: Nanjing Nanyou Information Industry Technology Research Institute Co. Ltd. Assignor: Nanjing Post & Telecommunication Univ. Contract record no.: 2018320000285 Denomination of invention: Regional suspicious target tracking and recognizing method based on multiple cameras Granted publication date: 20170215 License type: Common License Record date: 20181101 |
|
| EE01 | Entry into force of recordation of patent licensing contract | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20190611 Address after: Room 026-1, Building C, Science Park, Guanghua Road, Qinhuai District, Nanjing, Jiangsu Province, 210014 Patentee after: Nanjing Soft Xun Technology Co., Ltd. Address before: 210003 new model road 66, Gulou District, Nanjing, Jiangsu Patentee before: Nanjing Post & Telecommunication Univ. |
|
| TR01 | Transfer of patent right |