CN106204640A - A kind of moving object detection system and method - Google Patents
A kind of moving object detection system and method Download PDFInfo
- Publication number
- CN106204640A CN106204640A CN201610498648.5A CN201610498648A CN106204640A CN 106204640 A CN106204640 A CN 106204640A CN 201610498648 A CN201610498648 A CN 201610498648A CN 106204640 A CN106204640 A CN 106204640A
- Authority
- CN
- China
- Prior art keywords
- moving target
- rectangle frame
- point
- moving
- pedestrian
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Abstract
The present invention proposes a kind of moving object detection system, including movement destination image pretreatment module, movement destination image area determination module and moving target automatic tracking module, utilize optical flow method and grey level histogram coupling to carry out the moving target in moving target rectangle frame region from motion tracking.So, multiple moving targets can be tracked by described moving object detection system simultaneously, and realizes pedestrian detection and tracking, Statistics Bar people's quantity are carried out the functions such as motion tracking, and environmental suitability is strong, monitoring tracking accuracy is high.The invention allows for a kind of moving target detecting method.
Description
Technical field
The present invention relates to field of video monitoring, particularly to a kind of moving object detection system and method.
Background technology
Pedestrian in intelligent video monitoring detects and follows the tracks of the daily life in the affairs such as public administration, public safety in real time
In there is vital effect.Pedestrian's intelligent monitor system is made up of four main parts: pedestrian detection, pedestrian tracking and
Pedestrian's classification and pedestrian's Deviant Behavior identification.Owing to follow-up Classification and Identification process is highly dependent on target following accurately, just
True detect and track target is critically important.At present, in video, rest image target recognition has been used for reference in pedestrian target identification
Method, also have some videos pedestrian's target identification system apply method based on learning model to obtain preferably detection with
Track effect.And quickly and accurately detecting pedestrian and following the tracks of is the key technology in intelligent video monitoring system, it can not only
Transmit real-time monitoring information for management of public affairs person, additionally it is possible to moving target automatic tracking round-the-clockly, improve burst thing
The ability of part early warning.
The most in daily life, intelligent video monitoring, in municipal intelligent traffic manages, can detect for vehicle
And tracking, and the data such as the speed of vehicle, position can be calculated as required, thus improve break in traffic rules and regulations, vehicle accident and traffic
The process blocked up and emergency capability；In safety and protection monitoring, can be for bank, hospital, supermarket, residential block, warehouse and there is Gao An
Pedestrian is detected and follows the tracks of by full occasion place in real time, effectively takes precautions against large-scale groups event or the generation of accident, hits
The illegal activities such as robbery, theft, wired home management aspect can also be to child, old man, anemia of pregnant woman and there is major disease to cause
Unexpected notice hospital sues and labours etc. significant in time.But, in the Rapid Expansion of city, produced problem is managed with city
Contradiction between reason level seriously governs the demand for development in city, the cities and towns such as Mass disturbance, riot, attack of terrorism burst society
Can security incident, drastically influence cities and towns public safety, and the generation often of the social security events that happens suddenly, to a great extent with
Human body behavioral activity tight association.The most effectively determine human body behavioral activity, and its exception, questionable conduct are identified automatically, will
Contribute to Security Officer and process crisis in time, rapidly, safe precaution ability is substantially improved, thus builds the society of harmony, safety
Environment, oneself becomes an important topic of current international community.For effectively determining human body behavioral activity, it is it is crucial that the most effective
Detection and the pedestrian position followed the tracks of in video scene.Therefore it provides a kind of environmental suitability is higher, tracking accuracy is transported more accurately
Moving target detection technique is that those skilled in the art need badly and solve the technical problem that.
Summary of the invention
The technical problem to be solved in the present invention is: provide a kind of moving object detection system and method, its environmental suitability
By force, tracking accuracy is high.
The solution of the present invention is achieved in that a kind of moving object detection system, including
Movement destination image pretreatment module, during to target area image information gathering, by only using a two field picture complete
Become background modeling, for each pixel gathered in image information in the first frame, the picture of the random neighbours' point selecting it
Element value, as the sample value of its model, obtains the binary image of moving target；
Movement destination image area determination module, does connected region and carries out minute the foreground pixel point in binary image
Analysis, calculates the profile point set of each connected region, then clusters more than the profile of certain threshold value for height, finally obtain
Obtain moving target rectangle frame region completely；
Moving target automatic tracking module, utilizes optical flow method and grey level histogram to mate moving target rectangle frame region
Moving target is carried out from motion tracking.
So, described moving object detection system, by movement destination image pretreatment module, first carry out foreground detection,
I.e. background modeling, is obtained the binary image of moving target, then is clustered by minimum rectangle frame, obtains complete motion mesh
Mark rectangle frame region, then utilizes optical flow method and grey level histogram coupling to carry out the moving target in moving target rectangle frame region
From motion tracking, so, multiple moving targets can be tracked simultaneously, then judge moving target by multiframe data fusion
Type, finally realizes with tracking, Statistics Bar people's quantity, multiple target is carried out the functions such as motion tracking from motion tracking, pedestrian detection,
Its environmental suitability is strong, monitoring tracking accuracy is high.
Another technical scheme of the present invention is on above-mentioned basis, in movement destination image area determination module,
Described threshold value is that the minimum rectangle frame of 20 pixels clusters.
Another technical scheme of the present invention is on above-mentioned basis, also includes determining mould with movement destination image region
The pedestrian detection module that block connects, uses Adaboost algorithm to carry out pedestrian's inspection in complete moving target rectangle frame region
Survey.Specifically, after obtaining the motion target area of candidate, in rectangular area, carry out pedestrian detection.First extract in candidate frame
Multi-channel feature, multi-channel feature is to act on certain of original image to possess the mapping of translation invariance, described multichannel
Feature includes color characteristic (LUV), gradient magnitude feature and edge direction characteristic.After feature calculation completes, use Adaboost
Algorithm calculates this feature and belongs to the probability of pedestrian, finally according to whether there is pedestrian, and pedestrian in probability judgment rectangular area
Quantity.
Another technical scheme of the present invention is on above-mentioned basis, in described moving target automatic tracking module, and profit
Mating with optical flow method and grey level histogram carries out from the step of motion tracking as follows to moving target；
First the moving target rectangle frame region of former frame in video is carried out feature point extraction, secondly its characteristic point is adopted
Carry out Feature Points Matching with optical flow method and current frame motion target rectangle frame region, obtain the characteristic point of moving target, further according to
Whether the position of characteristic point changes and whether characteristic point belongs to foreground point, filters out the characteristic point of mistake, finally uses all
Characteristic point is filtered by value deviation method again, thus obtains the tracking knot of this moving target according to the match condition of characteristic point
Really.
If video exists multiple moving target, wherein there is the distance between multiple moving target position less than certain
Pixel number and moving target rectangle frame between occur overlap situation, be expressed as multiple moving target overlap.By meter
Calculate multiple moving target overlapping regions pixel point density, i.e. density=moving target pixel/moving target rectangle frame pixel,
Its density span is between 0 to 1, when its value is less than a certain particular value, it is judged that for multiple moving targets of overlapping region
Separate；Re-use grey level histogram the moving target of the moving target in foreground detection and tracking is mated, moved
Target separate after accurately follow the tracks of result.
Another technical scheme of the present invention is on above-mentioned basis, also includes statistical analysis module, described statistical
Analysis module is connected with described moving target automatic tracking module and pedestrian detection module, to the judgement of the direction of motion of moving target,
At least one information in movement velocity calculating, movement tendency prediction and pedestrian's quantity is added up.
A kind of moving target detecting method, comprises the following steps:
Movement destination image pretreatment, during to target area image information gathering, by only using a two field picture to complete the back of the body
Scape models, for each pixel gathered in image information in the first frame, the pixel value of the random neighbours' point selecting it
As the sample value of its model, obtain the binary image of moving target；
Movement destination image region determines, the foreground pixel point in binary image is done connected region and is analyzed, meter
Calculate the profile point set of each connected region, then cluster more than the profile of certain threshold value for height, finally obtained
Whole moving target rectangle frame region；
Moving target automatic tracking, utilizes optical flow method and the motion to moving target rectangle frame region of the grey level histogram coupling
Target is carried out from motion tracking.
So, described moving target detecting method, first pass through movement destination image pretreatment, carry out foreground detection, i.e. carry on the back
Scape models, and obtains the binary image of moving target, then is clustered by minimum rectangle frame, obtains complete moving target square
Shape frame region, then utilizes optical flow method and grey level histogram coupling to carry out the moving target in moving target rectangle frame region automatically
Follow the tracks of, so, can multiple moving targets be tracked simultaneously, then judged the class of moving target by multiframe data fusion
Type, finally realizes with tracking, Statistics Bar people's quantity, multiple target is carried out the functions such as motion tracking from motion tracking, pedestrian detection, its
Environmental suitability is strong, monitoring tracking accuracy is high.
Another technical scheme of the present invention is on above-mentioned basis, in movement destination image area determination step,
Described threshold value is that the minimum rectangle frame of 20 pixels clusters.
Another technical scheme of the present invention is on above-mentioned basis, and described movement destination image pre-treatment step is concrete
For, during to target area image information gathering, by only using a two field picture to complete background modeling, for gathering in image information
Each pixel in first frame, the pixel value of the random neighbours' point selecting it is as the sample value of its model, at random
The process selected is as follows: first translates the coordinate (x, y) of each pixel, then obtains coordinate translation amount, finally according to
Corresponding coordinate translation amount obtains the pixel value of random neighbours' point, i.e. model initialization；Obtaining such a sample
After collection, when updating every time, when the most often running into a new pixel, just by the collection in this pixel value and this sample set
Point contrasts, it is judged that whether this new pixel value is background dot, the process of the i.e. described classification to background dot, thus
Binary image to moving target.
Another technical scheme of the present invention is on above-mentioned basis, in described movement destination image area determination step
Clustering method particularly as follows:
1) barycenter of minimum rectangle frame is determined；
2) distance between each two minimum rectangle frame is calculated；
3), when the distance between two minimum rectangle frames is less than a certain particular value, just two minimum rectangle frames are merged,
Realize the cluster to moving target.
Another technical scheme of the present invention is on above-mentioned basis, at shown movement destination image area determination step
The most also include pedestrian detection step, use Adaboost algorithm to carry out pedestrian's inspection in complete moving target rectangle frame region
Survey；
Specifically, after obtaining the motion target area of candidate, carry out pedestrian detection in rectangular area.First candidate is extracted
Multi-channel feature in frame, then carry out feature calculation, then use Adaboost algorithm to calculate this feature and belong to the probability of pedestrian,
Finally according to whether probability judgment rectangular area exists pedestrian and pedestrian's quantity.
Another technical scheme of the present invention is that, on above-mentioned basis, described multi-channel feature is for acting on original image
Certain possess the mapping of translation invariance, including color characteristic, gradient magnitude feature and/or edge direction characteristic.
Another technical scheme of the present invention is on above-mentioned basis, and described moving target automatic tracking step is concrete
For:
First the moving target rectangle frame region of former frame in video is carried out feature point extraction, secondly its characteristic point is adopted
Carry out Feature Points Matching with optical flow method and current frame motion target rectangle frame region, obtain the characteristic point of moving target, further according to
Whether the position of characteristic point changes and whether characteristic point belongs to foreground point, filters out the characteristic point of mistake, finally uses all
Characteristic point is filtered by value deviation method again, thus obtains the tracking knot of this moving target according to the match condition of characteristic point
Really；Or,
If video exists multiple moving target, wherein there is the distance between multiple moving target position less than certain
Pixel number and moving target rectangle frame between occur overlap situation, then it represents that overlap for multiple moving targets, pass through
Calculating multiple moving target overlapping regions pixel point density, its density span is between 0 to 1, when its value is less than a certain spy
During definite value, it is judged that the multiple moving targets for overlapping region separate；Re-use grey level histogram to the motion mesh in foreground detection
Mark and follow the tracks of moving target mate, obtain moving target separate after accurately follow the tracks of result.
Another technical scheme of the present invention is, on above-mentioned basis, also to wrap after moving target automatic tracking step
Include statistical analysis step, in conjunction with pedestrian detection and the result of motion target tracking, moving target can be carried out direction of motion judgement,
Movement velocity calculates, movement tendency is predicted and pedestrian's quantity statistics.
1) moving target detected is carried out the pedestrian detection tracking result in conjunction with moving target, it is possible to statistics
The quantity of pedestrian；
2) characteristic point matched for optical flow method, according to the displacement of characteristic point N continuous frame in video image, can differentiate
Go out the direction of moving target；
3) characteristic point matched for optical flow method, during according to displacement and the interframe of the forward and backward frame of characteristic point in video image
Between poor, the speed of moving target can be calculated；
4) combine the speed of moving target, displacement, the trend of moving target can be doped.
Accompanying drawing explanation
The accompanying drawing of the part constituting the present invention is used for providing a further understanding of the present invention, and the present invention's is schematic real
Execute example and illustrate for explaining the present invention, being not intended that inappropriate limitation of the present invention.
Fig. 1 is the structured flowchart representing the moving object detection system involved by present embodiment.
Fig. 2 is to represent the moving target detecting method flow chart involved by present embodiment；
Fig. 3 is the knot after representing the result of background subtraction algorithm involved by present embodiment and clustering target
Really；
Fig. 4 is the target following result after representing the moving target coincidence involved by present embodiment.
Detailed description of the invention
Describing the present invention below in conjunction with the accompanying drawings, the description of this part is only exemplary and explanatory, should not
Protection scope of the present invention is had any restriction effect.Additionally, those skilled in the art are according to the description of presents, can be right
In presents, the feature in embodiment and in different embodiment carries out respective combination.
The embodiment of the present invention is as follows, refers to Fig. 1, a kind of moving object detection system, the fortune being connected including signal successively
Moving-target image pre-processing module 1, during to target area image information gathering, builds by only using a two field picture to complete background
Mould, for each pixel gathered in image information in the first frame, the pixel value conduct of the random neighbours' point selecting it
The sample value of its model, obtains the binary image of moving target；Movement destination image area determination module 2, to binaryzation
Foreground pixel point in image does connected region and is analyzed, and calculates the profile point set of each connected region, so that it is determined that bag
Enclose the region of the minimum rectangle frame of each profile, then carry out more than or equal to the minimum rectangle frame of 20 pixels for height
Cluster, finally obtains complete moving target rectangle frame region；Moving target automatic tracking module 3, utilizes optical flow method and gray scale
The moving target in moving target rectangle frame region is carried out from motion tracking by Histogram Matching.
So, described moving object detection system, by movement destination image pretreatment module, first carry out foreground detection,
I.e. background modeling, is obtained the binary image of moving target, then is clustered by minimum rectangle frame, obtains complete motion mesh
Mark rectangle frame region, then utilizes optical flow method and grey level histogram coupling to carry out the moving target in moving target rectangle frame region
From motion tracking, so, multiple moving targets can be tracked simultaneously, then judge moving target by multiframe data fusion
Type, finally realizes with tracking, Statistics Bar people's quantity, multiple target is carried out the functions such as motion tracking from motion tracking, pedestrian detection,
Its environmental suitability is strong, monitoring tracking accuracy is high.
On the basis of above-described embodiment, in another embodiment of the present invention, also include determining with movement destination image region
The pedestrian detection module that module connects, uses Adaboost algorithm to carry out pedestrian's inspection in complete moving target rectangle frame region
Survey.Specifically, after obtaining the motion target area of candidate, in rectangular area, carry out pedestrian detection.First extract in candidate frame
Multi-channel feature, multi-channel feature refers to act on certain of original image and possesses the mapping of translation invariance, described manifold
Road feature includes color characteristic (LUV), gradient magnitude feature and edge direction characteristic.After feature calculation completes, use
Adaboost algorithm calculates this feature and belongs to the probability of pedestrian, whether there is pedestrian finally according in probability judgment rectangular area,
And pedestrian's quantity.
On the basis of above-described embodiment, in another embodiment of the present invention, described moving object detection unit, utilize described
After movement destination image information carries out foreground detection, then carry out multi-channel feature extraction and Adaboost algorithm pedestrian detection, right
Multiple moving targets are tracked.
On the basis of above-described embodiment, in another embodiment of the present invention, in described moving target automatic tracking module, profit
Mating with optical flow method and grey level histogram carries out from the step of motion tracking as follows to moving target；
First the moving target rectangle frame region of former frame in video is carried out feature point extraction, secondly its characteristic point is adopted
Carry out Feature Points Matching with optical flow method and current frame motion target rectangle frame region, obtain the characteristic point of moving target, further according to
Whether the position of characteristic point changes and whether characteristic point belongs to foreground point, filters out the characteristic point of mistake, finally uses all
Characteristic point is filtered by value deviation method again, thus obtains the tracking knot of this moving target according to the match condition of characteristic point
Really.
If video exists multiple moving target, wherein there is the distance between multiple moving target position less than certain
Pixel number and moving target rectangle frame between occur overlap situation, be expressed as multiple moving target overlap.By meter
Calculate multiple moving target overlapping regions pixel point density, i.e. density=moving target pixel/moving target rectangle frame pixel,
Its density span is between 0 to 1, when its value is less than a certain particular value, it is judged that for multiple moving targets of overlapping region
Separate；Re-use grey level histogram the moving target of the moving target in foreground detection and tracking is mated, moved
Target separate after accurately follow the tracks of result.
On the basis of above-described embodiment, in another embodiment of the present invention, also include statistical analysis module, described statistical
Analysis module is connected with described moving target automatic tracking module and pedestrian detection module, to the judgement of the direction of motion of moving target,
At least one information in movement velocity calculating, movement tendency prediction and pedestrian's quantity is added up.
On the basis of above-described embodiment, in another embodiment of the present invention, as it is shown in figure 1, a kind of moving object detection side
Method, comprises the following steps:
Movement destination image pretreatment S1, during to target area image information gathering, by only using a two field picture to complete
Background modeling, for each pixel gathered in image information in the first frame, the pixel of the random neighbours' point selecting it
It is worth the sample value of the model as it, obtains the binary image of moving target；
Movement destination image region determines S2, the foreground pixel point in binary image is done connected region and is analyzed,
Calculate the profile point set of each connected region, so that it is determined that surround the region of the minimum rectangle frame of each profile, then for
Highly cluster more than or equal to the minimum rectangle frame of 20 pixels, finally obtain complete moving target rectangle frame district
Territory；
Moving target automatic tracking S3, utilizes optical flow method and the grey level histogram coupling fortune to moving target rectangle frame region
Moving-target is carried out from motion tracking.
So, described moving target detecting method, first pass through movement destination image pretreatment, carry out foreground detection, i.e. carry on the back
Scape models, and obtains the binary image of moving target, then is clustered by minimum rectangle frame, obtains complete moving target square
Shape frame region, then utilizes optical flow method and grey level histogram coupling to carry out the moving target in moving target rectangle frame region automatically
Follow the tracks of, so, can multiple moving targets be tracked simultaneously, then judged the class of moving target by multiframe data fusion
Type, finally realizes with tracking, Statistics Bar people's quantity, multiple target is carried out the functions such as motion tracking from motion tracking, pedestrian detection, its
Environmental suitability is strong, monitoring tracking accuracy is high.
On the basis of above-described embodiment, in another embodiment of the present invention, described movement destination image pre-treatment step has
Body is, during to target area image information gathering, by only using a two field picture to complete background modeling, for gathering image information
In each pixel in the first frame, the pixel value of the random neighbours' point selecting it as the sample value of its model, with
The process that machine selects is as follows: first translates the coordinate (x, y) of each pixel, then obtains coordinate translation amount, last root
The pixel value of random neighbours' point, i.e. model initialization is obtained according to corresponding coordinate translation amount；Obtaining such a sample
After this collection, when updating every time, when the most often running into a new pixel, just by adopting in this pixel value and this sample set
Collection point contrasts, it is judged that whether this new pixel value is background dot, the process of the i.e. described classification to background dot, thus
Obtain the binary image of moving target.
On the basis of above-described embodiment, in another embodiment of the present invention, described movement destination image area determination step
In clustering method particularly as follows:
1) barycenter of minimum rectangle frame is determined；
2) distance between each two minimum rectangle frame is calculated；
3), when the distance between two minimum rectangle frames is less than a certain particular value, just two minimum rectangle frames are merged,
Realize the cluster to moving target.
On the basis of above-described embodiment, in another embodiment of the present invention, determine step in shown movement destination image region
Also include pedestrian detection step after rapid, use Adaboost algorithm to carry out pedestrian in complete moving target rectangle frame region
Detection；
Specifically, after obtaining the motion target area of candidate, carry out pedestrian detection in rectangular area.First candidate is extracted
Multi-channel feature in frame, then carry out feature calculation, then use Adaboost algorithm to calculate this feature and belong to the probability of pedestrian,
Finally according to whether probability judgment rectangular area exists pedestrian and pedestrian's quantity.
On the basis of above-described embodiment, in another embodiment of the present invention, described multi-channel feature is for acting on original graph
Certain of picture possesses the mapping of translation invariance, including color characteristic, gradient magnitude feature and/or edge direction characteristic.
On the basis of above-described embodiment, in another embodiment of the present invention, described moving target automatic tracking step is concrete
For:
First the moving target rectangle frame region of former frame in video is carried out feature point extraction, secondly its characteristic point is adopted
Carry out Feature Points Matching with optical flow method and current frame motion target rectangle frame region, obtain the characteristic point of moving target, further according to
Whether the position of characteristic point changes and whether characteristic point belongs to foreground point, filters out the characteristic point of mistake, finally uses all
Characteristic point is filtered by value deviation method again, thus obtains the tracking knot of this moving target according to the match condition of characteristic point
Really；Or,
If video exists multiple moving target, wherein there is the distance between multiple moving target position less than certain
Pixel number and moving target rectangle frame between occur overlap situation, then it represents that overlap for multiple moving targets, pass through
Calculating multiple moving target overlapping regions pixel point density, its density span is between 0 to 1, when its value is less than a certain spy
During definite value, it is judged that the multiple moving targets for overlapping region separate；Re-use grey level histogram to the motion mesh in foreground detection
Mark and follow the tracks of moving target mate, obtain moving target separate after accurately follow the tracks of result.
On the basis of above-described embodiment, in another embodiment of the present invention, after moving target automatic tracking step also
Including statistical analysis step, in conjunction with pedestrian detection and the result of motion target tracking, moving target can be carried out the direction of motion and sentence
Disconnected, movement velocity calculates, movement tendency is predicted and pedestrian's quantity statistics.
1) moving target detected is carried out the pedestrian detection tracking result in conjunction with moving target, it is possible to statistics
The quantity of pedestrian；
2) characteristic point matched for optical flow method, according to the displacement of characteristic point N continuous frame in video image, can differentiate
Go out the direction of moving target；
3) characteristic point matched for optical flow method, during according to displacement and the interframe of the forward and backward frame of characteristic point in video image
Between poor, the speed of moving target can be calculated；
4) combine the speed of moving target, displacement, the trend of moving target can be doped.
Specifically, below as a example by the application in video monitoring system, it is 720*576 for acquisition monitoring video sequence
Pixel, frame per second is 25fps process, and this video static scene has fixing automobile, plant etc., and moving object mainly has row
People, is also blocked by between pedestrian, certainly blocks and the factors such as interference such as leaves disturbance, and the present embodiment comprises the following steps:
1, collection and the foreground detection of video image are gathered.After completing target area video image information collection, by only
The first two field picture in video is used to complete model initialization, each pixel (x, neighbours y) in random selection image
The pixel value of point, as the sample value of its model, builds sample set；Wherein the pixel value at x point is Vi, background sample at x
Collection M (N)={ V1,V2,...,VN, for, sample set size is N=20；One ball of the region representation as radius of the R centered by x
Body is SR (v (x)), and whether distance and radial difference by calculating pixel x to these sampled points are more than given threshold value, if
Be more than or equal to, then judge that this pixel x belongs to background dot, i.e. complete pixel be background dot or foreground point point
Class process, thus obtain the binary image of moving target.When a pixel is classified as background dot, then have that 1/ φ's is general
Rate updates the background model of oneself, and wherein φ determines according to the accuracy detecting foreground point, in its sample set
One sample point of random selection is filled with updating with this pixel.It also has the probability of 1/ φ to update its neighbour simultaneously
Occupying the background model of point, finally it has the probability of 1/ φ to update the background model of oneself, constantly updates to external diffusion successively,
Thus update the sample set of background pixel point, complete detection real-time to moving target.
2, as shown in Figure 3 and Figure 4, on the basis of the binary image that above-mentioned steps obtains, first to two detected
Foreground pixel point in value image does connected domain analysis, uses Cany edge detection algorithm to calculate the profile of each connected domain
Point set, progressively scans each profile order from top to bottom, from left to right, finds the most left extreme point of each profile
(XL, YL), the rightest extreme point (XR,YR), top extreme point (XT,YT), least significant end extreme point (XB,YB), so that it is determined that surround every
The region of the minimum rectangle frame of individual profile, then poly-more than or equal to doing of the minimum rectangle frame of 20 pixels for height
Class, the most in default situations, in video, the height of moving target is more than or equal to 20 pixels, finally obtains complete
Moving target rectangle frame area image.
After the profile finding detection moving target, clustering method is specific as follows: calculate the matter of all minimum rectangle frames
The heart (xi, yi), according to the distance between each two minimum rectangle frame barycenter, when distance is less than a certain particular value, then it is judged as closing
And, otherwise, then be judged as separately, draw according to experiment, particular value be 10 pixels be optimal；Between two of which rectangle frame
The computing formula of distance be:
The coordinate assuming the barycenter of two rectangle frames is (xa, ya) and (xb, yb), width is respectively wa、wb, height is respectively
ha、hb, according to the rectangle frame characteristic that human body is exclusive, depth-width ratio is 7:3.
If xa-xb≤ (wa+wb) * 0.3 and ya-yb≤ (ha+hb) * 0.7, then judge that rectangle frame is overlapping；Between it away from
From for 0；
If xa-xb≤ (wa+wb) * 0.3, then judge that rectangle frame x direction is overlapping；Distance between it is | ya-yb|-(ha
+hb)*0.7；
If ya-yb≤ (ha+hb) * 0.7, then judge that rectangle frame y direction is overlapping；Distance between it is | xa-xb|-(wa
+wb)*0.3；
If xa-xb>=(wa+wb) * 0.3 and ya-yb>=(ha+hb) * 0.7, then judge that rectangle frame is the most overlapping；Between it
Distance is
Constantly repeat this process until canonical measure function starts convergence, the most all use mean square deviation as standard
Measure function, is tracked its moving target with this.
3, use Adaboost algorithm that complete moving target rectangle frame region is carried out pedestrian detection；
After obtaining the motion target area of candidate, use slip window sampling detection pedestrian.First, scale according to a certain percentage
Image in candidate rectangle window, until image window is less than 64*32, thus builds the image pyramid of candidate rectangle window.
At pyramidal each layer, use the window of fixing 64*32 size to slide in the picture, calculate the multichannel in 64*32 window
Feature, and utilize Adaboost algorithm to calculate this feature to belong to the probability of pedestrian, finally give the pedestrian in motion target area
Position.
In each position of each level of image pyramid, it is required for calculating multi-channel feature.Multi-channel feature bag
Including color characteristic (LUV), gradient direction feature and edge amplitude Characteristics, wherein color characteristic 3-dimensional, gradient direction feature 6 is tieed up,
Edge amplitude Characteristics 3-dimensional, during Practical Calculation, is divided into the fritter of 16*8 4*4 by the image block of 64*32, and each fritter has 10 dimensions
Feature, so the image block of 64*32 a total of 16*8*10 dimensional feature.First each pixel position in the image block of 64*32 is calculated
The transverse gradients value put and longitudinal Grad, and obtain the gradient magnitude of each pixel.Then, the image block of 64*32 is divided into
The fritter of 4*4, in each fritter, calculates gradient magnitude feature and gradient direction feature respectively.Gradient magnitude is characterized as 4*4
The gradient magnitude average of pixel, and when calculating gradient direction feature, the gradient direction of 0-360 degree is divided into 6 deciles, each etc.
Points 60 degree, the transverse gradients of 4*4 pixel of fritter with in longitudinal direction gradient projection to these 6 deciles, thus obtain 6 directions
On projection accumulation, just obtained gradient direction feature after normalization.
After multi-channel feature has calculated, many Weak Classifiers can be formed according to each component of multi-channel feature, I
Use Adaboost algorithm by integrated for these a Weak Classifiers strong classifier, its training process as follows:
Assuming that X represents that sample space, Y represent sample class logo collection, owing to current problem only exists pedestrian and non-row
People hinders two classes, so Y={1 ,-1}.Make S={ (x1,y1),(x2,y2)…(xN,yN) it is sample training collection, wherein xi∈ X, yi
∈ Y, i=1,2...N.
The first step, initialization sample weight, align sample,To negative sampleWherein m, n are respectively
The number of positive negative sample, and meet m+n=N.
Second step, initializes maximum iteration time T=50；
3rd step, For t=1,2 ..., T
1) Weak Classifier learning algorithm is called, it is thus achieved that optimal Weak Classifier ht: X → Y so that the false drop rate of its correspondence is:
2) all sample weights are updated,Wherein βt=εt/(1-εt), ei=1 represents xiBy ht
X () is correctly classified, ei=0 represents xiBy htX () is correctly classified；
3) weight of all samples is normalized so that
4th step, T the optimum Weak Classifier generated according to step 3, finally giving pedestrian's identification and classification device is:
Herein willThe probability of pedestrian is belonged to as sample,Non-pedestrian is belonged to as sample
Probability.
4, optical flow method and grey level histogram coupling is utilized to carry out moving target from motion tracking；
First a continuous print sequence of frames of video is processed, extract the moving target rectangle frame of former frame in video
Provincial characteristics point, secondly uses optical flow method and current frame motion target rectangle frame region to carry out Feature Points Matching its characteristic point,
Obtaining the characteristic point of moving target, whether the position further according to characteristic point changes and whether characteristic point belongs to foreground point, i.e.
For any two adjacent video frames afterwards, find the key feature points occurred in previous frame optimum bit in the current frame
Put, thus obtain foreground target position coordinates in the current frame, obtain effective tracking characteristics point, filter out in estimation range
The characteristic point that matching degree is high, first filters out the characteristic point of mistake, finally uses mean bias method again to filter characteristic point,
So iteration is carried out, thus obtains the tracking result of this moving target according to the match condition of characteristic point.
If there is multiple moving target in video, the distance between plurality of moving target position is less than certain picture
The situation of overlap occurs between vegetarian refreshments number and moving target rectangle frame, is expressed as multiple moving target and overlaps.Many by calculating
Individual moving target overlapping region pixel point density represents the degree of overlapping between motion target area, i.e. density=moving target picture
Vegetarian refreshments/moving target rectangle frame pixel, its density span is between 0 to 1, when its value is less than a certain particular value, sentences
The multiple moving targets for overlapping region that break separate, otherwise are judged as overlap, and such as this density particular value can be drawn by experiment
0.1 is optimal, and namely when less than 0.1, moving target separates；When more than 0.1, it is judged that overlapping for multiple moving targets；
Re-use grey level histogram the moving target of the moving target in foreground detection and tracking is mated, obtain moving target and divide
Result is accurately followed the tracks of after opening.
5, combine the result of pedestrian detection and motion target tracking, moving target can be carried out direction of motion judgement, motion
The prediction of speed calculation, movement tendency and pedestrian's quantity statistics.
1) moving target detected is carried out the pedestrian detection tracking result in conjunction with moving target, it is possible to statistics
The quantity of pedestrian；
2) characteristic point matched for optical flow method, according to the displacement of characteristic point N continuous frame in video image, can differentiate
Go out the direction of moving target；
3) characteristic point matched for optical flow method, during according to displacement and the interframe of the forward and backward frame of characteristic point in video image
Between poor, the speed of moving target can be calculated；
4) combine the speed of moving target, displacement, the trend of moving target can be doped.
The above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For Yuan, under the premise without departing from the principles of the invention, it is also possible to make some improvements and modifications, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (10)
1. a moving object detection system, it is characterised in that include
Movement destination image pretreatment module, during to target area image information gathering, by only using a two field picture to complete the back of the body
Scape models, for each pixel gathered in image information in the first frame, the pixel value of the random neighbours' point selecting it
As the sample value of its model, obtain the binary image of moving target；
Movement destination image area determination module, does connected region to the foreground pixel point in binary image and is analyzed, meter
Calculate the profile point set of each connected region, then cluster more than the profile of certain threshold value for height, finally obtained
Whole moving target rectangle frame region；
Moving target automatic tracking module, utilizes optical flow method and the motion to moving target rectangle frame region of the grey level histogram coupling
Target is carried out from motion tracking.
Moving object detection system the most according to claim 1, it is characterised in that also include and movement destination image region
Determine the pedestrian detection module that module connects, use Adaboost algorithm to go in complete moving target rectangle frame region
People detects.
Moving object detection system the most according to claim 2, it is characterised in that also include statistical analysis module, described
Statistical analysis module is connected with described moving target automatic tracking module and pedestrian detection module, the direction of motion to moving target
At least one information in judgement, movement velocity calculating, movement tendency prediction and pedestrian's quantity is added up.
4. a moving target detecting method, it is characterised in that comprise the following steps:
Movement destination image pretreatment, during to target area image information gathering, builds by only using a two field picture to complete background
Mould, for each pixel gathered in image information in the first frame, the pixel value conduct of the random neighbours' point selecting it
The sample value of its model, obtains the binary image of moving target；
Movement destination image region determines, the foreground pixel point in binary image is done connected region and is analyzed, calculate
The profile point set of each connected region, then clusters more than the profile of certain threshold value for height, finally obtains complete
Moving target rectangle frame region；
Moving target automatic tracking, utilizes optical flow method and the grey level histogram coupling moving target to moving target rectangle frame region
Carry out from motion tracking.
Moving target detecting method the most according to claim 4, it is characterised in that described movement destination image region determines
Clustering method in step particularly as follows:
1) barycenter of minimum rectangle frame is determined；
2) distance between each two minimum rectangle frame is calculated；
3), when the distance between two minimum rectangle frames is less than a certain particular value, just two minimum rectangle frames are merged, it is achieved
Cluster to moving target.
Moving target detecting method the most according to claim 4, it is characterised in that true in shown movement destination image region
Also include pedestrian detection step after determining step, use Adaboost algorithm to carry out in complete moving target rectangle frame region
Pedestrian detection.
Moving target detecting method the most according to claim 6, it is characterised in that described pedestrian detection step specifically,
In described moving target rectangle frame region, extract the multi-channel feature in candidate frame, after feature calculation completes, use
Adaboost algorithm calculates this feature and belongs to the probability of pedestrian, finally according to whether probability judgment rectangular area exists pedestrian and
Pedestrian's quantity.
Moving target detecting method the most according to claim 7, it is characterised in that described multi-channel feature is former for acting on
Certain of beginning image possesses the mapping of translation invariance, including color characteristic, gradient magnitude feature and/or edge direction characteristic.
Moving target detecting method the most according to claim 4, it is characterised in that described moving target automatic tracking step
Particularly as follows:
First the moving target rectangle frame region of former frame in video is carried out feature point extraction, secondly its characteristic point is used light
Stream method and current frame motion target rectangle frame region carry out Feature Points Matching, obtain the characteristic point of moving target, further according to feature
Whether the position of point changes and whether characteristic point belongs to foreground point, filters out the characteristic point of mistake, finally uses average inclined
Characteristic point is filtered by difference method again, thus obtains the tracking result of this moving target according to the match condition of characteristic point；Or
Person,
If video exists multiple moving target, wherein there is the distance between multiple moving target position less than certain picture
The situation of overlap occurs, then it represents that overlap for multiple moving targets between vegetarian refreshments number and moving target rectangle frame, by calculating
Multiple moving target overlapping regions pixel point density, its density span is between 0 to 1, when its value is less than a certain particular value
Time, it is judged that the multiple moving targets for overlapping region separate；Re-use grey level histogram to the moving target in foreground detection and
Follow the tracks of moving target mate, obtain moving target separate after accurately follow the tracks of result.
10. according to the moving target detecting method according to any one of claim 4 to 9, it is characterised in that at moving target certainly
Also include statistical analysis step after motion tracking step, specifically include
1) moving target detected is carried out pedestrian detection, in conjunction with the tracking result of moving target, the number of statistics pedestrian
Amount；
2) characteristic point matched for optical flow method, according to the displacement of characteristic point N continuous frame in video image, it determines moving target
Direction；
3) characteristic point matched for optical flow method, displacement and inter frame temporal according to the forward and backward frame of characteristic point in video image are poor,
Calculate the speed of moving target；
4) speed of moving target, displacement are combined, it was predicted that the trend of moving target.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610498648.5A CN106204640A (en) | 2016-06-29 | 2016-06-29 | A kind of moving object detection system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610498648.5A CN106204640A (en) | 2016-06-29 | 2016-06-29 | A kind of moving object detection system and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106204640A true CN106204640A (en) | 2016-12-07 |
Family
ID=57463547
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610498648.5A Pending CN106204640A (en) | 2016-06-29 | 2016-06-29 | A kind of moving object detection system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106204640A (en) |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897742A (en) * | 2017-02-21 | 2017-06-27 | 北京市商汤科技开发有限公司 | Method, device and electronic equipment for detecting object in video |
| CN107044855A (en) * | 2017-05-05 | 2017-08-15 | 南京信息工程大学 | A kind of inertial measuring unit and method based on camera array |
| CN107248173A (en) * | 2017-06-08 | 2017-10-13 | 深圳市智美达科技股份有限公司 | Method for tracking target, device, computer equipment and storage medium |
| CN108154090A (en) * | 2017-12-11 | 2018-06-12 | 北京小米移动软件有限公司 | Face identification method and device |
| CN108665478A (en) * | 2018-05-11 | 2018-10-16 | 西安天和防务技术股份有限公司 | Multi-target Data Association under a kind of complex scene |
| CN108710828A (en) * | 2018-04-18 | 2018-10-26 | 北京汽车集团有限公司 | The method, apparatus and storage medium and vehicle of identification object |
| CN108932496A (en) * | 2018-07-03 | 2018-12-04 | 北京佳格天地科技有限公司 | The quantity statistics method and device of object in region |
| CN109685833A (en) * | 2018-12-28 | 2019-04-26 | 镇江市高等专科学校 | Method for tracking moving target |
| CN109697387A (en) * | 2017-10-23 | 2019-04-30 | 北京京东尚科信息技术有限公司 | Direction of motion prediction technique, device, electronic equipment and storage medium |
| CN109754411A (en) * | 2018-11-22 | 2019-05-14 | 济南艾特网络传媒有限公司 | Building pivot frame larceny detection method and system are climbed based on optical flow method target following |
| CN109766756A (en) * | 2018-12-10 | 2019-05-17 | 中国科学院自动化研究所南京人工智能芯片创新研究院 | Make a dash across the red light data processing method, device, computer equipment and storage medium |
| CN109785363A (en) * | 2018-12-29 | 2019-05-21 | 中国电子科技集团公司第五十二研究所 | A kind of unmanned plane video motion Small object real-time detection and tracking |
| CN109932281A (en) * | 2017-12-19 | 2019-06-25 | 中国科学院沈阳自动化研究所 | The liquid viscosity On-line Measuring Method of view-based access control model |
| CN110008951A (en) * | 2019-03-14 | 2019-07-12 | 深兰科技（上海）有限公司 | A kind of object detection method and device |
| CN110097578A (en) * | 2019-05-09 | 2019-08-06 | 电子科技大学 | Plastic grains tracking |
| CN110407052A (en) * | 2019-08-02 | 2019-11-05 | 浙江新再灵科技股份有限公司 | Strenuous exercise's behavioral value method and system in a kind of elevator |
| CN110472614A (en) * | 2019-08-22 | 2019-11-19 | 四川自由健信息科技有限公司 | A kind of recognition methods for behavior of falling in a swoon |
| CN110555863A (en) * | 2019-09-11 | 2019-12-10 | 湖南德雅坤创科技有限公司 | moving object detection method and device and computer readable storage medium |
| CN110909607A (en) * | 2019-10-24 | 2020-03-24 | 北京中润惠通科技发展有限公司 | Device system for sensing passenger flow in intelligent subway operation |
| CN111145151A (en) * | 2019-12-23 | 2020-05-12 | 维沃移动通信有限公司 | Motion area determination method and electronic equipment |
| CN111383244A (en) * | 2020-02-28 | 2020-07-07 | 浙江大华技术股份有限公司 | Target detection tracking method |
| CN112184751A (en) * | 2019-07-04 | 2021-01-05 | 虹软科技股份有限公司 | Object identification method and system and electronic equipment |
| TWI732374B (en) * | 2019-12-10 | 2021-07-01 | 財團法人工業技術研究院 | Method and apparatus for object recognition |
| CN113313007A (en) * | 2021-05-26 | 2021-08-27 | 每日互动股份有限公司 | Pedestrian static state identification method based on video, electronic equipment and storage medium |
| US11250269B2 (en) | 2019-04-18 | 2022-02-15 | Fujitsu Limited | Recognition method and apparatus for false detection of an abandoned object and image processing device |
| CN116434126A (en) * | 2023-06-13 | 2023-07-14 | 清华大学 | Method and device for detecting micro-vibration speed of crops |
| CN116453062A (en) * | 2023-06-12 | 2023-07-18 | 青岛义龙包装机械有限公司 | Packaging machine assembly risk monitoring method based on robot high-precision compliant assembly |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120243733A1 (en) * | 2011-03-22 | 2012-09-27 | Morpho, Inc. | Moving object detecting device, moving object detecting method, moving object detection program, moving object tracking device, moving object tracking method, and moving object tracking program |
| CN103955947A (en) * | 2014-03-21 | 2014-07-30 | 南京邮电大学 | Multi-target association tracking method based on continuous maximum energy and apparent model |
| CN104392469A (en) * | 2014-12-15 | 2015-03-04 | 辽宁工程技术大学 | Target tracking method based on soft characteristic theory |
| CN104866823A (en) * | 2015-05-11 | 2015-08-26 | 重庆邮电大学 | Vehicle detection and tracking method based on monocular vision |
-
2016
- 2016-06-29 CN CN201610498648.5A patent/CN106204640A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120243733A1 (en) * | 2011-03-22 | 2012-09-27 | Morpho, Inc. | Moving object detecting device, moving object detecting method, moving object detection program, moving object tracking device, moving object tracking method, and moving object tracking program |
| CN103955947A (en) * | 2014-03-21 | 2014-07-30 | 南京邮电大学 | Multi-target association tracking method based on continuous maximum energy and apparent model |
| CN104392469A (en) * | 2014-12-15 | 2015-03-04 | 辽宁工程技术大学 | Target tracking method based on soft characteristic theory |
| CN104866823A (en) * | 2015-05-11 | 2015-08-26 | 重庆邮电大学 | Vehicle detection and tracking method based on monocular vision |
Non-Patent Citations (4)
| Title |
|---|
| 刘赏等: """帧平均加速度"人群异常运动检测算法"", 《计算机工程与应用》 * |
| 吴鸿: ""视频监控中多目标检测与跟踪算法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 李文辉等: ""一种改进的Adaboost 训练算法"", 《吉林大学学报( 理学版)》 * |
| 邱祯艳: ""基于实时视频的运动目标检测算法"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897742A (en) * | 2017-02-21 | 2017-06-27 | 北京市商汤科技开发有限公司 | Method, device and electronic equipment for detecting object in video |
| CN106897742B (en) * | 2017-02-21 | 2020-10-27 | 北京市商汤科技开发有限公司 | Method and device for detecting object in video and electronic equipment |
| WO2018153323A1 (en) * | 2017-02-21 | 2018-08-30 | 北京市商汤科技开发有限公司 | Method and device for detecting object in video, and electronic device |
| CN107044855A (en) * | 2017-05-05 | 2017-08-15 | 南京信息工程大学 | A kind of inertial measuring unit and method based on camera array |
| CN107248173A (en) * | 2017-06-08 | 2017-10-13 | 深圳市智美达科技股份有限公司 | Method for tracking target, device, computer equipment and storage medium |
| CN109697387B (en) * | 2017-10-23 | 2021-07-30 | 北京京东尚科信息技术有限公司 | Motion direction prediction method and device, electronic equipment and storage medium |
| CN109697387A (en) * | 2017-10-23 | 2019-04-30 | 北京京东尚科信息技术有限公司 | Direction of motion prediction technique, device, electronic equipment and storage medium |
| CN108154090B (en) * | 2017-12-11 | 2021-10-01 | 北京小米移动软件有限公司 | Face recognition method and device |
| CN108154090A (en) * | 2017-12-11 | 2018-06-12 | 北京小米移动软件有限公司 | Face identification method and device |
| CN109932281A (en) * | 2017-12-19 | 2019-06-25 | 中国科学院沈阳自动化研究所 | The liquid viscosity On-line Measuring Method of view-based access control model |
| CN109932281B (en) * | 2017-12-19 | 2021-08-17 | 中国科学院沈阳自动化研究所 | Vision-based liquid viscosity on-line measuring method |
| CN108710828A (en) * | 2018-04-18 | 2018-10-26 | 北京汽车集团有限公司 | The method, apparatus and storage medium and vehicle of identification object |
| CN108710828B (en) * | 2018-04-18 | 2021-01-01 | 北京汽车集团有限公司 | Method, device and storage medium for identifying target object and vehicle |
| CN108665478A (en) * | 2018-05-11 | 2018-10-16 | 西安天和防务技术股份有限公司 | Multi-target Data Association under a kind of complex scene |
| CN108932496A (en) * | 2018-07-03 | 2018-12-04 | 北京佳格天地科技有限公司 | The quantity statistics method and device of object in region |
| CN108932496B (en) * | 2018-07-03 | 2022-03-25 | 北京佳格天地科技有限公司 | Method and device for counting number of target objects in area |
| CN109754411A (en) * | 2018-11-22 | 2019-05-14 | 济南艾特网络传媒有限公司 | Building pivot frame larceny detection method and system are climbed based on optical flow method target following |
| CN109766756A (en) * | 2018-12-10 | 2019-05-17 | 中国科学院自动化研究所南京人工智能芯片创新研究院 | Make a dash across the red light data processing method, device, computer equipment and storage medium |
| CN109685833A (en) * | 2018-12-28 | 2019-04-26 | 镇江市高等专科学校 | Method for tracking moving target |
| CN109685833B (en) * | 2018-12-28 | 2023-04-07 | 镇江市高等专科学校 | Moving target tracking method |
| CN109785363A (en) * | 2018-12-29 | 2019-05-21 | 中国电子科技集团公司第五十二研究所 | A kind of unmanned plane video motion Small object real-time detection and tracking |
| CN110008951A (en) * | 2019-03-14 | 2019-07-12 | 深兰科技（上海）有限公司 | A kind of object detection method and device |
| US11250269B2 (en) | 2019-04-18 | 2022-02-15 | Fujitsu Limited | Recognition method and apparatus for false detection of an abandoned object and image processing device |
| CN110097578A (en) * | 2019-05-09 | 2019-08-06 | 电子科技大学 | Plastic grains tracking |
| CN110097578B (en) * | 2019-05-09 | 2021-08-17 | 电子科技大学 | Plastic particle tracking method |
| CN112184751A (en) * | 2019-07-04 | 2021-01-05 | 虹软科技股份有限公司 | Object identification method and system and electronic equipment |
| CN110407052A (en) * | 2019-08-02 | 2019-11-05 | 浙江新再灵科技股份有限公司 | Strenuous exercise's behavioral value method and system in a kind of elevator |
| CN110407052B (en) * | 2019-08-02 | 2020-11-10 | 浙江新再灵科技股份有限公司 | Method and system for detecting violent movement behaviors in elevator |
| CN110472614A (en) * | 2019-08-22 | 2019-11-19 | 四川自由健信息科技有限公司 | A kind of recognition methods for behavior of falling in a swoon |
| CN110555863A (en) * | 2019-09-11 | 2019-12-10 | 湖南德雅坤创科技有限公司 | moving object detection method and device and computer readable storage medium |
| CN110909607A (en) * | 2019-10-24 | 2020-03-24 | 北京中润惠通科技发展有限公司 | Device system for sensing passenger flow in intelligent subway operation |
| TWI732374B (en) * | 2019-12-10 | 2021-07-01 | 財團法人工業技術研究院 | Method and apparatus for object recognition |
| CN111145151A (en) * | 2019-12-23 | 2020-05-12 | 维沃移动通信有限公司 | Motion area determination method and electronic equipment |
| CN111145151B (en) * | 2019-12-23 | 2023-05-26 | 维沃移动通信有限公司 | Motion area determining method and electronic equipment |
| CN111383244A (en) * | 2020-02-28 | 2020-07-07 | 浙江大华技术股份有限公司 | Target detection tracking method |
| CN111383244B (en) * | 2020-02-28 | 2023-09-01 | 浙江大华技术股份有限公司 | Target detection tracking method |
| CN113313007A (en) * | 2021-05-26 | 2021-08-27 | 每日互动股份有限公司 | Pedestrian static state identification method based on video, electronic equipment and storage medium |
| CN113313007B (en) * | 2021-05-26 | 2022-10-14 | 每日互动股份有限公司 | Pedestrian static state identification method based on video, electronic equipment and storage medium |
| CN116453062A (en) * | 2023-06-12 | 2023-07-18 | 青岛义龙包装机械有限公司 | Packaging machine assembly risk monitoring method based on robot high-precision compliant assembly |
| CN116453062B (en) * | 2023-06-12 | 2023-08-22 | 青岛义龙包装机械有限公司 | Packaging machine assembly risk monitoring method based on robot high-precision compliant assembly |
| CN116434126A (en) * | 2023-06-13 | 2023-07-14 | 清华大学 | Method and device for detecting micro-vibration speed of crops |
| CN116434126B (en) * | 2023-06-13 | 2023-09-19 | 清华大学 | Method and device for detecting micro-vibration speed of crops |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| 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 | |
| CN106910203B (en) | The quick determination method of moving target in a kind of video surveillance | |
| CN103077423B (en) | To run condition detection method based on crowd's quantity survey of video flowing, local crowd massing situation and crowd | |
| CN104599502B (en) | Method for traffic flow statistics based on video monitoring | |
| CN101141633B (en) | Moving object detecting and tracing method in complex scene | |
| CN101719216B (en) | Movement human abnormal behavior identification method based on template matching | |
| CN102214359B (en) | Target tracking device and method based on hierarchic type feature matching | |
| CN105787472B (en) | A kind of anomaly detection method based on the study of space-time laplacian eigenmaps | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| CN107230267B (en) | Intelligence In Baogang Kindergarten based on face recognition algorithms is registered method | |
| CN100525395C (en) | Pedestrian tracting method based on principal axis marriage under multiple vedio cameras | |
| CN101833771B (en) | Tracking device and method for solving multiple-target meeting dodging | |
| CN107622258A (en) | A kind of rapid pedestrian detection method of combination static state low-level image feature and movable information | |
| CN111860274B (en) | Traffic police command gesture recognition method based on head orientation and upper half skeleton characteristics | |
| CN105678803A (en) | Video monitoring target detection method based on W4 algorithm and frame difference | |
| CN102903124A (en) | Moving object detection method | |
| CN103942575A (en) | System and method for analyzing intelligent behaviors based on scenes and Markov logic network | |
| CN103854027A (en) | Crowd behavior identification method | |
| CN106127812A (en) | A kind of passenger flow statistical method of non-gate area, passenger station based on video monitoring | |
| CN106373146A (en) | Target tracking method based on fuzzy learning | |
| CN107491749A (en) | Global and local anomaly detection method in a kind of crowd's scene | |
| CN105956568A (en) | Abnormal behavior detecting and early warning method based on monitored object identification | |
| CN109086682A (en) | A kind of intelligent video black smoke vehicle detection method based on multi-feature fusion | |
| CN104680554A (en) | SURF-based compression tracing method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161207 |
|
| RJ01 | Rejection of invention patent application after publication |