CN103324950B - Human body reappearance detecting method and system based on online study - Google Patents
Human body reappearance detecting method and system based on online study Download PDFInfo
- Publication number
- CN103324950B CN103324950B CN201210077833.9A CN201210077833A CN103324950B CN 103324950 B CN103324950 B CN 103324950B CN 201210077833 A CN201210077833 A CN 201210077833A CN 103324950 B CN103324950 B CN 103324950B
- Authority
- CN
- China
- Prior art keywords
- human body
- detection
- human
- model
- module
- 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.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a human body reappearance detecting method and system based on online study. The method comprises the following steps of detecting the human body in a video frame to obtain a human body detecting result set in the video frame, tracking the human body in the human body detecting result set, studying and updating a human body characteristic model according to the human body detecting set and human body tracking results, storing and managing the human body characteristic model to obtain a matched human body characteristic model, and carrying out human body reappearance detection in a new input video frame.
Description
Technical field
The present invention relates to pattern recognition and computer vision field, particularly to a kind of, inspection is reappeared based on on-line study human body
Survey method and its system.
Background technology
So-called human body reappears detection it is simply that the picture frame being engraved under a video camera when a given people is a certain is (with this
The human detection window of people), it is engraved in, when finding every other, the image that under current camera or other video cameras, this people occurred
Frame/sequence.From these Search Results, we can obtain this people in appearance situation on diverse location for the following period of time and
The information such as the behavior occurring.Human body reappears detection and has a very wide range of applications in field of video monitoring.Detected using reappearing
Result, we can carry out long period behavior analysiss using the position that people on different periods occurs and its behavior of being occurred;
By analyzing the appearance situation of people to be checked on diverse geographic location, the operation such as missing missing and suspect's search can be carried out.
In addition, human body reappear detection can also improve existing track algorithm robustness so that track algorithm shift or with
Again recover tracking target in the case of losing, keep the seriality followed the tracks of.Existing method is mostly under concern and different visual angles
The individual matching algorithm of same pedestrian, and assume that pedestrian is detected from image well and (detect and typically pass through handss
Work marks), and as a practicality human body playback system, need to consider automatic sexual function and automatically managed according to run time
The ability of pedestrian's property data base.
Content of the invention
For solving the above problems, the present invention attempts to design intelligent video processing function, can automatically realize from video
The detection of human body, tracking and feature extraction;And can examine with other according to for the tracking situation of human body and this detection body
The relation surveying object to be upgraded human body apparent model automatically.Provide a kind of human body based on on-line study reappear detecting system and
Its method.
The present invention discloses a kind of human body based on on-line study and reappears detection method, comprises the steps:
Step 1, detects human body in the video frame, obtains the human detection set in described frame of video;
Step 2, is tracked to the human body in described human detection set;
Step 3, according to the result of described human detection set and human body tracking, is learnt and is updated characteristics of human body's mould
Type;
Step 4, storage and management characteristics of human body's model, obtain the characteristics of human body's model mating, and in new input video
Carry out human body in frame and reappear detection.
The described human body based on on-line study reappears detection method, and described step 1 also includes:
Step 21, by the way of image window scanning, detects human body using human body detector in image pyramid;When
Human body occurs in the video frame, the rectangle frame that output human body is located.
The described human body based on on-line study reappears detection method, and described step 21 also includes:
Step 31, collects the human body image window having marked and non-human image window as training data, and to training
Data extracts feature;
Step 32, based on the image feature representation extracting, by way of statistical learning, study human body image window divides
Class device is distinguishing in a certain image window whether comprise human body;
Step 33, for input picture in each yardstick resampling, constitutes image pyramid, each to image pyramid
The image zooming-out characteristics of image of individual image resampling;
Step 34, classifies to each of image pyramid image window, finds out the detection of all characteristics of human body
Window;
Classification results are carried out the operation of deduplication detection by step 35 using clustering method.
The described human body based on on-line study reappears detection method, and described step 2 also includes:
Step 41, using Optical-flow Feature, predicts position and the yardstick of target object;
Step 42, for the result initialized target track algorithm of human detection;
Step 43, extracts light stream in the present frame and next frame of video, gives light stream prediction target object in next frame
Position and size;
Step 44, the credibility of output tracking result.
The described human body based on on-line study reappears detection method, and described step 3 also includes:
Step 51, the result that human detection and human body tracking are used, as input data, is detected for each and connects
The continuous human body tracing into sets up characteristics of human body's model;
Step 52, in frame of video afterwards is processed, detects reproduction in frame of video using described characteristics of human body's model
Human body, and update its corresponding characteristics of human body's model using the reproduction human body detecting.
The described human body based on on-line study reappears detection method, and described step 51 also includes:
Step 61, judges whether to comprise emerging human body in the human detection result of input video frame;
Step 62, if emerging human body, trains characteristics of human body's model, and adds characteristics of human body's model set to it
In;
Step 63, conversely, human detection is associated with human body tracking result, based on described human detection and human body with
Track result updates its corresponding characteristics of human body's model.
The described human body based on on-line study also includes after reappearing detection method, described step 4:
Step 5, records the accessed time of the characteristics of human body's model having built up, is not accessed for characteristics of human body's mould for a long time
Type is all deleted.
The described human body based on on-line study reappears detection method, and described step 4 also includes:
Step 81, when managing characteristics of human body's model, after the N frame of interval, re-starts human detection, and by human detection
Characteristics of human body's model that result comparison has built up, described N is the natural number more than or equal to 1;
Step 82, for the human detection process not having the characteristics of human body's model mating, repeated execution of steps 1 is to step 4;
Step 83, for the characteristics of human body's model being matched, the characteristics of human body's model using described new detection is updated.
The present invention discloses a kind of human body based on on-line study and reappears detecting system, including:
Human detection module, for detecting human body in the video frame, obtains the human detection set in described frame of video;
Target tracking module, for being tracked to the human body in described human detection set;
Study and update module, for the result according to described human detection set and human body tracking, carry out study and more
New characteristics of human body's model;
Storage and management module, for storing and managing characteristics of human body's model, obtain mate characteristics of human body's model and
Carry out human body in new input video frame and reappear detection.
The described human body based on on-line study reappears detecting system, and described human detection module also includes:
Extract video frame module, for extracting characteristics of image in frame of video, and using image window scanning, in image gold word
Human body is detected in tower;When human body occurs in the video frame, the rectangle frame that output human body is located.
Described reappears detecting system based on on-line study human body, and described extraction video frame module also includes:
Training data module, for collecting the human body image window having marked and non-human image window as training number
According to, and feature is extracted to training data;
Statistical learning module, for based on the image feature representation extracting, by way of statistical learning, learns human body
Image window grader is distinguishing in a certain image window whether comprise human body;
Image pyramid module, constitutes image pyramid, to image for for input picture in each yardstick resampling
The image zooming-out characteristics of image of each image resampling pyramidal;
Window sort module, for classifying to each of image pyramid image window, finds out all human bodies
The detection window of feature;
Cluster module, for carrying out the operation of deduplication detection using clustering method to classification results.
Described reappears detecting system based on on-line study human body, and described target tracking module also includes:
Optical-flow Feature module, for adopting Optical-flow Feature, the position of prediction target object and yardstick;
Initialization module, for the result initialized target track algorithm for human detection;
Light stream prediction module, for extracting light stream in the present frame and next frame of frame of video, gives light stream prediction target
Object is in the position of next frame and size;
Output module, for the credibility of output tracking result.
Described reappears detecting system based on on-line study human body, and described study and update module also include:
Set up characteristics of human body's model module, for the result for human detection and human body tracking as input data, right
Set up characteristics of human body's model in the human body that each newly detects;
Update characteristics of human body's model module, for afterwards frame of video process in, using described characteristics of human body's model Lai
Human body in detection frame of video reappears, and updates its corresponding characteristics of human body's model using the human body reappearing.
Described reappears detecting system based on on-line study human body, and described characteristics of human body's model module of setting up also includes:
Judge testing result module, for judging whether to comprise emerging human body in the human detection result newly obtaining;
Training module, for if emerging human body, the model that training of human body surface is seen, and it is added to human detection collection
In conjunction;
Result relating module, for human detection is associated with human body tracking result, based on described human detection with
Human body tracking result updates current characteristics of human body's model.
Described reappears detecting system based on on-line study human body, also includes:
Record update module, for recording the accessed time of the characteristics of human body's model having built up, is not accessed for people
Body characteristicses model is all deleted.
Described reappears detecting system based on on-line study human body, and storage and management module also includes:
Again detection module, for when managing characteristics of human body's model, after the N frame of interval, re-starting human detection, and
Human detection result is compared the characteristics of human body's model having built up, described N is the natural number more than or equal to 1;
Repeat module, for the human detection process for the characteristics of human body's model not having coupling, repeat behaviour
Make;
Coupling characteristics of human body's model module, for for the characteristics of human body's model being matched, using the people of described new detection
Body characteristicses model is updated.
Invention, due to taking above technical scheme, has following technique effect:
(1) technology of automatic for human body detection technique, target following technology and target recognition is combined by this method,
Therefore full automatic can manage, detect and update characteristics of human body's model, carry out online full automatic human body and reappear detection;
(2) using human body appearance features so that reappearing detection there is certain visual angle robustness, and can apply to point
The relatively low environment of resolution;
(3) due to characteristics of human body's model of human body detector and on-line study, in human body due to seriously blocking or moving
Reproduction after going out the visual field has the ability again detecting.
Brief description
Figure 1A -1B is that the human body based on on-line study for the present invention reappears detecting system specific embodiment schematic flow sheet;
Fig. 2 is that the human body based on on-line study for the present invention reappears detection method flow chart;
Fig. 3 is that the human body based on on-line study for the present invention reappears detecting system flow chart.
Specific embodiment
The specific embodiment of the present invention is given below, in conjunction with accompanying drawing, detailed description is made that to the present invention.
As shown in Fig. 2 the present invention discloses a kind of human body based on on-line study reappears detection method, comprise the steps:
Step 1, detects human body in the video frame, obtains the human detection set in described frame of video;
Step 2, is tracked to the human body in described human detection set;
Step 3, according to the result of described human detection set and human body tracking, is learnt and is updated characteristics of human body's mould
Type;
Step 4, storage and management characteristics of human body's model, obtain the characteristics of human body's model mating, and in new input video
Carry out human body in frame and reappear detection.
Technique effect:By strategies such as comprehensive detection, tracking, the study of online anthropometric dummy and dynamic management so that this
The bright human body that carries out that can be on-line Full reappears detection.And can automatically manage storage so that system can be long-term height
Effect is run.
The described human body based on on-line study reappears detection method, and described step 1 also includes:
Step 21, by the way of image window scanning, detects human body using human body detector in image pyramid;When
Human body occurs in the video frame, the rectangle frame that output human body is located.
The described human body based on on-line study reappears detection method, and described step 21 also includes:
Step 31, collects the human body image window having marked and non-human image window as training data, and to training
Data extracts feature;
Step 32, based on the image feature representation extracting, by way of statistical learning, study human body image window divides
Class device is distinguishing in a certain image window whether comprise human body;
Step 33, for input picture in each yardstick resampling, constitutes image pyramid, each to image pyramid
The image zooming-out characteristics of image of individual image resampling;
Step 34, classifies to each of image pyramid image window, finds out the detection of all characteristics of human body
Window;
Classification results are carried out the operation of deduplication detection by step 35 using clustering method.
Technique effect:By way of pyramid scans with using simple histogram of gradients, this detection algorithm can
Efficiently, the human body occurring in quick positioning video.
The described human body based on on-line study reappears detection method, and described step 2 also includes:
Step 41, using Optical-flow Feature, predicts position and the yardstick of target object;
Step 42, for the result initialized target track algorithm of human detection;
Step 43, extracts light stream in the present frame and next frame of video, gives light stream prediction target object in next frame
Position and size;
Step 44, the credibility of output tracking result.
Technique effect:Method using light stream predicts position of human body, improves tracking velocity, and the real-time for system provides
Ensure.
The described human body based on on-line study reappears detection method, and described step 3 also includes:
Step 51, the result that human detection and human body tracking are used, as input data, is detected for each and connects
The continuous human body tracing into sets up characteristics of human body's model;
Step 52, in frame of video afterwards is processed, detects reproduction in frame of video using described characteristics of human body's model
Human body, and update its corresponding characteristics of human body's model using the reproduction human body detecting.
The described human body based on on-line study reappears detection method, and described step 51 also includes:
Step 61, judges whether to comprise emerging human body in the human detection result of input video frame;
Step 62, if emerging human body, trains characteristics of human body's model, and adds characteristics of human body's model set to it
In;
Step 63, conversely, human detection is associated with human body tracking result, based on described human detection and human body with
Track result updates its corresponding characteristics of human body's model.
Technique effect:Using online characteristics of human body's model learning method so that system full automatic can carry out human body
Reappear detection;The system of ensure that automatically updates of model has the change at light according to current environment, the visual angle of people etc. and adjusts
Model parameter, self adaptation scene changes.
The described human body based on on-line study also includes after reappearing detection method, described step 4:
Step 5, records the accessed time of the characteristics of human body's model having built up, is not accessed for characteristics of human body's mould for a long time
Type is all deleted..Described long-term be not accessed existence time threshold value, can be designed according to concrete scene, in an experiment we
Generally use 100 frames (or 4 seconds).
Technique effect:Anthropometric dummy memory management ensure that system operationally will not internal memory be overflowed, and ensure that mould
The speed of type coupling, which ensure that our reproduction detecting system can be with long-play.
The described human body based on on-line study reappears detection method, and described step 4 also includes:
Step 81, when managing characteristics of human body's model, after the N frame of interval, re-starts human detection, and by human detection
Characteristics of human body's model that result comparison has built up, described N is the natural number more than or equal to 1;
Step 82, for the human detection process not having the characteristics of human body's model mating, repeated execution of steps 1 is to step 4;
Step 83, for the characteristics of human body's model being matched, the characteristics of human body's model using described new detection is updated.
Due to the difference of the conditions such as visual angle and illumination, even same person, also can show huge under different cameras
Big difference, these all make human body reappear detection becomes extremely difficult.In the existing research work of international and national, research
Persons attempt to carry out human body reproduction detection by information such as face and gaits, and make some progress.But, people is just
Face is many times sightless in monitor video, and carrying of facial characteristics of low resolution monitor video strong influence
Take;For gait, because the change at visual angle and the impact (such as sick) of physiological situation have typically exhibited very big difference
The opposite sex.According to these observations, the present invention proposes to represent human body using apparent (wearing clothes) feature of human body, and utilizes on-line study
Mode setting up the apparent discrimination model of human body, and then realize comparing robust and reappear detection.
The present invention passes through to extract the current characteristic model observing human body, is engraved in phase by way of model compares when other
Find the people that can be good at being described by this feature model to carry out human body reproduction in the picture frame that same or different cameras capture
Detection;Especially, a kind of method based on Human Detection and on-line study of present invention proposition is automatic from monitor video
Obtain human body target, and reappear detection for the human body target training detecting and the automatic body updating characteristics of human body's model and calculate
Method.
Human body based on on-line study reappears detecting system, specifically includes following step:(1) give input video
First frame, carries out human detection by human body detector to this frame, obtains the human detection frame set in frame of video;(2) pass through
Track algorithm is tracked to the human body that these detect;(3) learnt and more with the result followed the tracks of according to above-mentioned human detection
New characteristics of human body's model;(4) after being spaced several frames, re-start human detection, and these testing results are compared have built up
Characteristics of human body's model;To (3), new spy is set up for the human detection repeat step (1) not finding characteristics of human body's Model Matching
Levy model;For the characteristic model on being matched, carry out model modification using these new observed data;(5) it is being spaced one section
After time, delete the characteristic model that these are not updated.
Because the purpose of the present invention is to set up the high human body playback system of a set of computational efficiency, design detection module, with
In track module, we employ some fast algorithms.
Using a kind of full automatic human detection algorithm, including following operating procedure:(1) collect the human body image having marked
Window and non-human image window are as training data, and extract feature, such as histogram of gradients to training data;(2) it is based on
The image feature representation extracting, is learnt human body image window grader by way of statistical learning and distinguishes a certain image window
Whether human body is comprised in mouthful.In an experiment, we adopt bootstrapping (boosting) learning algorithm as Statistical learning model;(3)
For input picture resampling on each yardstick, constitute image pyramid, each image resampling to image pyramid
Image zooming-out histogram of gradients feature;Wherein said image pyramid is exactly that image is carried out down-sampling for multiple yardsticks, this
The image construction image pyramid of a little different scales;(4) each of image pyramid window is classified, find out all
It is probably the detection window of human body target;(5) using clustering method using clustering method be the present invention a part, but cluster
Algorithm can carry out the operation of deduplication detection using existing algorithm to classification results.The window that described deduplication refers to it
Between overlapping very big, corresponding is same human body.This human detection algorithm, because histogram of gradients can be fast based on integrogram
The calculating of speed and bootstrapping (boosting) algorithm can be used to design detector, therefore can quickly determine the position of human body target
And yardstick.
According to testing result, it is proposed that a kind of quick target tracking algorism the object detecting is tracked,
Track algorithm includes three below step:(1) the result initialized target track algorithm based on human detection;(2) in video
Extract light stream in present frame and next frame, give light stream prediction target object in the position of next frame and size;(3) output tracking
The credibility of result.The operating rate of system due to adopting optical flow tracking, can be further speeded up.
On the basis of online characteristics of human body's model-learning algorithm is the result building on above-mentioned human detection and the result followed the tracks of,
There is automatic management, training and more New function, its workflow is as follows:(1) judge in the new human detection result obtaining
Whether comprise emerging human body;(2) it is based on the judgement draw in step (1) and make following operation:(2.1) if new occur
Human body, the model that training of human body surface is seen, and be added to somatic data storehouse;(2.2) conversely, close testing result with tracking result
Connection gets up, and updates current characteristics of human body's model based on tracking result.
Specifically, for characteristics of human body's model module, the present invention is learnt using online mode and updates human body spy
Levy model.As shown in Figure 1, observe in frame whether include the corresponding human body of characteristic model according to current, characteristic model is drawn
It is divided into two classes:After the corresponding people of characteristic model is through the human detection and Model Matching of system, identification is to observe current
When occurring in frame, we are called " active grader ", otherwise, are called " inactive grader ".Wherein Figure 1A be as
The human body that newly detects of fruit is all not up to threshold value through all inactive grader T_I after judging, i.e. itself and existing human body
Characteristic model all mismatches, then initialize a new grader that enlivens and it is tracked;When Figure 1B is to reappear human body to occur,
Its corresponding grader is activated and continues human body and update sorter model.
For input monitoring video, when initial before enliven grader set TAAnd inactive grader set TIIt is sky.
We carry out a human detection every n (in an experiment, n=50) frame, obtain detection block set BD.If BDIt is not empty, each inspection
Survey frame and enliven grader (characteristic model) for one as initial alignment position initialization, and add it to enliven grader set
TAIn.
Ensuing each frame, TAIn grader all target is tracked and carries out online characteristic model more simultaneously
Newly, TACorresponding tracking box collection is combined into BT.If someone disappears from video, its corresponding grader state is set to inactive shape
State, is added to inactive grader set TIIn.
If BDIn detection block BiWith BTThe common factor area ratio of all tracking box be typically designed less than certain threshold
For detection block size 50% is then it is assumed that BiBelong to the new people entering video camera, need it to be carried out reappear detection, that is, judge TI
In with the presence or absence of this people grader.
With inactive grader set TIIn model to BiCarry out classification to judge, obtain the probability belonging to this people, take all
The maximum of class probability, if maximum is more than certain threshold value, its threshold value is such as 0.7, then start this grader and continue to Bi
It is tracked, otherwise it is assumed that being emerging people, initialization one enlivens grader with regard to the new of this people, is added to TAIn.
For characteristics of human body's model management aspect, for the consideration of matching speed, internal memory and disk space, in a period of time
Inside all do not become the characteristics of human body's model enlivening grader, we are deleted.The present invention is using " nearest privilege of access "
Principle is managing characteristics of human body's model.Set up a characteristic model queue, every be accessed for recently characteristic model (i.e. with new
The characteristic model that testing result matches) it is placed in head of the queue.Every one section of set time, the characteristic model of tail of the queue is deleted to release
Put memory space.
As shown in figure 3, the present invention discloses a kind of human body based on on-line study reappears detecting system, including:
Human detection module 10, for detecting human body in the video frame, obtains the human detection set in described frame of video;
Target tracking module 20, for being tracked to the human body in described human detection set;
Study with update module 30, for the result according to described human detection set and human body tracking, carry out study and
Update characteristics of human body's model;
Storage and management module 40, for storing and managing characteristics of human body's model, obtains the characteristics of human body's model mating simultaneously
Carry out human body and reappear detection in new input video frame.
The described human body based on on-line study reappears detecting system, and described human detection module also includes:
Extract video frame module, for extracting characteristics of image in frame of video, and using image window scanning, in image gold word
Human body is detected in tower;When human body occurs in the video frame, the rectangle frame that output human body is located.
Described reappears detecting system based on on-line study human body, and described extraction video frame module also includes:
Training data module, for collecting the human body image window having marked and non-human image window as training number
According to, and feature is extracted to training data;
Statistical learning module, for based on the image feature representation extracting, by way of statistical learning, learns human body
Image window grader is distinguishing in a certain image window whether comprise human body;
Image pyramid module, constitutes image pyramid, to image for for input picture in each yardstick resampling
The image zooming-out characteristics of image of each image resampling pyramidal;
Window sort module, for classifying to each of image pyramid image window, finds out all human bodies
The detection window of feature;
Cluster module, for carrying out the operation of deduplication detection using clustering method to classification results.
Described reappears detecting system based on on-line study human body, and described target tracking module also includes:
Optical-flow Feature module, for adopting Optical-flow Feature, the position of prediction target object and yardstick;
Initialization module, for the result initialized target track algorithm for human detection;
Light stream prediction module, for extracting light stream in the present frame and next frame of frame of video, gives light stream prediction target
Object is in the position of next frame and size;
Output module, for the credibility of output tracking result.
Described reappears detecting system based on on-line study human body, and described study and update module also include:
Set up characteristics of human body's model module, for the result for human detection and human body tracking as input data, right
Set up characteristics of human body's model in the human body that each newly detects;
Update characteristics of human body's model module, for afterwards frame of video process in, using described characteristics of human body's model Lai
Human body in detection frame of video reappears, and updates its corresponding characteristics of human body's model using the human body reappearing.
Described reappears detecting system based on on-line study human body, and described characteristics of human body's model module of setting up also includes:
Judge testing result module, for judging whether to comprise emerging human body in the human detection result newly obtaining;
Training module, for if emerging human body, the model that training of human body surface is seen, and it is added to human detection collection
In conjunction;
Result relating module, for human detection is associated with human body tracking result, based on described human detection with
Human body tracking result updates current characteristics of human body's model.
Described reappears detecting system based on on-line study human body, also includes:
Record update module, for recording the accessed time of the characteristics of human body's model having built up, is not accessed for people
Body characteristicses model is all deleted.
Described reappears detecting system based on on-line study human body, and storage and management module also includes:
Again detection module, for when managing characteristics of human body's model, after the N frame of interval, re-starting human detection, and
Human detection result is compared the characteristics of human body's model having built up, described N is the natural number more than or equal to 1;
Repeat module, for the human detection process for the characteristics of human body's model not having coupling, repeat behaviour
Make;
Coupling characteristics of human body's model module, for for the characteristics of human body's model being matched, using the people of described new detection
Body characteristicses model is updated.
Those skilled in the art, under conditions of the spirit and scope of the present invention determining without departing from claims, goes back
Various modifications can be carried out to above content.
Claims (16)
1. a kind of human body based on on-line study reappears detection method it is characterised in that comprising the steps:
Step 1, detects human body in the video frame, obtains the human detection set in described frame of video;
Step 2, is tracked to the human body in described human detection set;
Step 3, according to the result of described human detection set and human body tracking, is learnt and is updated characteristics of human body's model;
Step 4, storage and management characteristics of human body's model, obtain the characteristics of human body's model mating, and in new input video frame
Carry out human body and reappear detection;
Wherein said characteristics of human body's model is guaranteed replacement active grader and inactive grader, detects human body every n frame, obtains
Human detection set BDIf, human detection set BDIt is not empty, then by human detection set BDIn each detection block mark as initial
Determine one active grader of position initialization, and initialized active grader is added to active grader set TA
In;
Ensuing each frame, active grader set TAIn grader all human body is tracked and carries out online simultaneously
The renewal of characteristics of human body's model, active grader set TACorresponding human detection collection is combined into BTIf human body disappears from video
Lose, corresponding for the human body of disappearance grader state is set to an inactive state, is added to inactive grader set TIIn;
If human detection set BDIn detection block BiWith BTAll tracking box common factor area ratio be less than a threshold value, then BiBelong to
In the new people entering video camera, judge inactive grader set TIIn with the presence or absence of the new people entering video camera grader.
2. the human body based on on-line study as claimed in claim 1 reappears detection method it is characterised in that described step 1 is gone back
Including:
Step 21, by the way of image window scanning, detects human body using human body detector in image pyramid;Work as human body
Occur in the video frame, the rectangle frame that output human body is located.
3. the human body based on on-line study as claimed in claim 2 reappears detection method it is characterised in that described step 21 is gone back
Including:
Step 31, collects the human body image window having marked and non-human image window as training data, and to training data
Extract feature;
Step 32, based on the image feature representation extracting, by way of statistical learning, learns human body image window grader
To distinguish in a certain image window whether comprise human body;
Step 33, for input picture in each yardstick resampling, constitutes image pyramid, each figure to image pyramid
Image zooming-out characteristics of image as resampling;
Step 34, classifies to each of image pyramid image window, finds out the detection window of all characteristics of human body;
Classification results are carried out the operation of deduplication detection by step 35 using clustering method.
4. the human body based on on-line study as claimed in claim 1 reappears detection method it is characterised in that described step 2 is gone back
Including:
Step 41, using Optical-flow Feature, predicts position and the yardstick of target object;
Step 42, for the result initialized target track algorithm of human detection;
Step 43, extracts light stream in the present frame and next frame of video, gives light stream prediction target object in the position of next frame
Put and size;
Step 44, the credibility of output tracking result.
5. the human body based on on-line study as claimed in claim 1 reappears detection method it is characterised in that described step 3 is gone back
Including:
Step 51, is used the result of human detection and human body tracking as input data, for each detected and continuous with
Track to human body set up characteristics of human body's model;
Step 52, in frame of video afterwards is processed, detects, using described characteristics of human body's model, the human body reappearing in frame of video,
And update its corresponding characteristics of human body's model using the reproduction human body detecting.
6. the human body based on on-line study as claimed in claim 5 reappears detection method it is characterised in that described step 51 is gone back
Including:
Step 61, judges whether to comprise emerging human body in the human detection result of input video frame;
Step 62, if emerging human body, trains characteristics of human body's model, and adds in characteristics of human body's model set to it;
Step 63, conversely, human detection is associated with human body tracking result, is tied with human body tracking based on described human detection
Fruit updates its corresponding characteristics of human body's model.
7. the human body based on on-line study as claimed in claim 1 reappear detection method it is characterised in that described step 4 it
Also include afterwards:
Step 5, records the accessed time of the characteristics of human body's model having built up, is not accessed for characteristics of human body's model all for a long time
It is deleted.
8. the human body based on on-line study as claimed in claim 1 reappears detection method it is characterised in that described step 4 is gone back
Including:
Step 81, when managing characteristics of human body's model, after the N frame of interval, re-starts human detection, and by human detection result
Compare the characteristics of human body's model having built up, described N is the natural number more than or equal to 1;
Step 82, for the human detection process not having the characteristics of human body's model mating, repeated execution of steps 1 is to step 4;
Step 83, for the characteristics of human body's model being matched, is entered to characteristics of human body's model using human detection result in step 81
Row updates.
9. a kind of human body based on on-line study reappears detecting system it is characterised in that including:
Human detection module, for detecting human body in the video frame, obtains the human detection set in described frame of video;
Target tracking module, for being tracked to the human body in described human detection set;
Study and update module, for the result according to described human detection set and human body tracking, carry out study and more new person
Body characteristicses model;
Storage and management module, for storing and managing characteristics of human body's model, obtains characteristics of human body's model of coupling and new
Carry out human body in input video frame and reappear detection;
Wherein said characteristics of human body's model is guaranteed replacement active grader and inactive grader, detects human body every n frame, obtains
Human detection set BDIf, human detection set BDIt is not empty, then by human detection set BDIn each detection block mark as initial
Determine one active grader of position initialization, and initialized active grader is added to active grader set TA
In;
Ensuing each frame, active grader set TAIn grader all human body is tracked and carries out online simultaneously
The renewal of characteristics of human body's model, active grader set TACorresponding human detection collection is combined into BTIf human body disappears from video
Lose, corresponding for the human body of disappearance grader state is set to an inactive state, is added to inactive grader set TIIn;
If human detection set BDIn detection block BiWith BTAll tracking box common factor area ratio be less than a threshold value, then BiBelong to
In the new people entering video camera, judge inactive grader set TIIn with the presence or absence of the new people entering video camera grader.
10. the human body based on on-line study as claimed in claim 9 reappears detecting system it is characterised in that described people's health check-up
Survey module also to include:
Extract video frame module, for extracting characteristics of image in frame of video, and using image window scanning, in image pyramid
Detection human body;When human body occurs in the video frame, the rectangle frame that output human body is located.
11. are based on as claimed in claim 10 on-line study human body reappears detecting system it is characterised in that described extraction video
Frame module also includes:
Training data module, for collecting the human body image window having marked and non-human image window as training data, and
Feature is extracted to training data;
Statistical learning module, for based on the image feature representation extracting, by way of statistical learning, learns human body image
Window grader is distinguishing in a certain image window whether comprise human body;
Image pyramid module, constitutes image pyramid for for input picture in each yardstick resampling, to image gold word
The image zooming-out characteristics of image of each image resampling of tower;
Window sort module, for classifying to each of image pyramid image window, finds out all characteristics of human body
Detection window;
Cluster module, for carrying out the operation of deduplication detection using clustering method to classification results.
12. are based on as claimed in claim 9 on-line study human body reappears detecting system it is characterised in that described target following
Module also includes:
Optical-flow Feature module, for adopting Optical-flow Feature, the position of prediction target object and yardstick;
Initialization module, for the result initialized target track algorithm for human detection;
Light stream prediction module, for extracting light stream in the present frame and next frame of frame of video, gives light stream prediction target object
Position and size in next frame;
Output module, for the credibility of output tracking result.
13. be based on as claimed in claim 9 on-line study human body reappear detecting system it is characterised in that described study with more
New module also includes:
Set up characteristics of human body's model module, for the result for human detection and human body tracking as input data, for every
One human body newly detecting sets up characteristics of human body's model;
Update characteristics of human body's model module, in processing in frame of video afterwards, detected using described characteristics of human body's model
Human body in frame of video reappears, and updates its corresponding characteristics of human body's model using the human body reappearing.
14. are based on on-line study human body as claimed in claim 13 reappears detecting system it is characterised in that described set up human body
Characteristic model module also includes:
Judge testing result module, for judging whether to comprise emerging human body in the human detection result newly obtaining;
Training module, for if emerging human body, the model that training of human body surface is seen, and it is added to human detection set
In;
Result relating module, for associating human detection with human body tracking result, based on described human detection and human body
Tracking result updates current characteristics of human body's model.
15. are based on on-line study human body as claimed in claim 9 reappears detecting system it is characterised in that also including:
Record update module, for recording the accessed time of the characteristics of human body's model having built up, is not accessed for human body special
Levy model to be all deleted.
16. are based on as claimed in claim 9 on-line study human body reappears detecting system it is characterised in that storage and management mould
Block also includes:
Again detection module, for when managing characteristics of human body's model, after the N frame of interval, re-starting human detection, and by people
Characteristics of human body's model that the comparison of body testing result has built up, described N is the natural number more than or equal to 1;
Repeat module, for the human detection process for the characteristics of human body's model not having coupling, repeat operation;
Coupling characteristics of human body's model module, for for the characteristics of human body's model being matched, using human body in detection module again
Testing result is updated to characteristics of human body's model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210077833.9A CN103324950B (en) | 2012-03-22 | 2012-03-22 | Human body reappearance detecting method and system based on online study |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210077833.9A CN103324950B (en) | 2012-03-22 | 2012-03-22 | Human body reappearance detecting method and system based on online study |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103324950A CN103324950A (en) | 2013-09-25 |
CN103324950B true CN103324950B (en) | 2017-02-08 |
Family
ID=49193678
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210077833.9A Active CN103324950B (en) | 2012-03-22 | 2012-03-22 | Human body reappearance detecting method and system based on online study |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103324950B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104820998B (en) * | 2015-05-27 | 2019-11-26 | 成都通甲优博科技有限责任公司 | A kind of human testing based on unmanned motor platform and tracking and device |
CN106355188B (en) * | 2015-07-13 | 2020-01-21 | 阿里巴巴集团控股有限公司 | Image detection method and device |
CN109215052A (en) * | 2017-06-29 | 2019-01-15 | 沈阳新松机器人自动化股份有限公司 | Quick human body searching method and system based on motion segmentation |
CN112734805B (en) * | 2021-01-11 | 2022-04-15 | 北京深睿博联科技有限责任公司 | Pedestrian motion trajectory prediction method and device based on deep learning |
CN112766230A (en) * | 2021-02-09 | 2021-05-07 | 浙江工商大学 | Video streaming personnel online time length estimation method and corresponding system |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101231703A (en) * | 2008-02-28 | 2008-07-30 | 上海交通大学 | Method for tracing a plurality of human faces base on correlate vector machine to improve learning |
CN101561928B (en) * | 2009-05-27 | 2011-09-14 | 湖南大学 | Multi-human body tracking method based on attribute relational graph appearance model |
CN102270348B (en) * | 2011-08-23 | 2013-08-07 | 中国科学院自动化研究所 | Method for tracking deformable hand gesture based on video streaming |
-
2012
- 2012-03-22 CN CN201210077833.9A patent/CN103324950B/en active Active
Non-Patent Citations (1)
Title |
---|
基于视频分析的智能视频监控系统的研究与设计;王文勇;《中国优秀硕士学位论文全文数据库 信息科技辑》;20111231(第12期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN103324950A (en) | 2013-09-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104091176B (en) | Portrait comparison application technology in video | |
WO2021043073A1 (en) | Urban pet movement trajectory monitoring method based on image recognition and related devices | |
CN109145708B (en) | Pedestrian flow statistical method based on RGB and D information fusion | |
CN103517042B (en) | A kind of nursing house old man's hazardous act monitoring method | |
CN104978567B (en) | Vehicle checking method based on scene classification | |
Wang et al. | Automated sewer pipe defect tracking in CCTV videos based on defect detection and metric learning | |
TWI382762B (en) | Method for tracking moving object | |
Gawande et al. | Pedestrian detection and tracking in video surveillance system: issues, comprehensive review, and challenges | |
CN106204640A (en) | A kind of moving object detection system and method | |
CN106203274A (en) | Pedestrian's real-time detecting system and method in a kind of video monitoring | |
CN103324950B (en) | Human body reappearance detecting method and system based on online study | |
WO2014050518A1 (en) | Information processing device, information processing method, and information processing program | |
CN107230267B (en) | Intelligence In Baogang Kindergarten based on face recognition algorithms is registered method | |
CN109829382B (en) | Abnormal target early warning tracking system and method based on intelligent behavior characteristic analysis | |
CN107657232B (en) | Pedestrian intelligent identification method and system | |
CN102722698A (en) | Method and system for detecting and tracking multi-pose face | |
Dubuisson et al. | A survey of datasets for visual tracking | |
CN106355604A (en) | Target image tracking method and system | |
CN111652035B (en) | Pedestrian re-identification method and system based on ST-SSCA-Net | |
CN113362374A (en) | High-altitude parabolic detection method and system based on target tracking network | |
CN112541403B (en) | Indoor personnel falling detection method by utilizing infrared camera | |
CN113963399A (en) | Personnel trajectory retrieval method and device based on multi-algorithm fusion application | |
CN103049748A (en) | Behavior-monitoring method and behavior-monitoring system | |
Huang et al. | A method for fast fall detection | |
CN113963371A (en) | Human body abnormal posture detection method |
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 |