CN109034171A - Unlicensed vehicle checking method and device in video flowing - Google Patents
Unlicensed vehicle checking method and device in video flowing Download PDFInfo
- Publication number
- CN109034171A CN109034171A CN201810821013.3A CN201810821013A CN109034171A CN 109034171 A CN109034171 A CN 109034171A CN 201810821013 A CN201810821013 A CN 201810821013A CN 109034171 A CN109034171 A CN 109034171A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- license plate
- training
- sample
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07B—TICKET-ISSUING APPARATUS; FARE-REGISTERING APPARATUS; FRANKING APPARATUS
- G07B15/00—Arrangements or apparatus for collecting fares, tolls or entrance fees at one or more control points
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Abstract
The present invention provides the unlicensed vehicle checking methods and device in a kind of video flowing.The unlicensed vehicle checking method from multiple video clips of the video flowing the following steps are included: obtain the training sample set for training linear svm classifier model;The HOG feature of all positive negative samples is extracted from training sample set obtained;Off-line training is carried out to the svm classifier model using extracted HOG feature, to obtain the disaggregated model of initial optimization;Using the disaggregated model of the video information in the video flowing again on-line training initial optimization, to obtain optimal disaggregated model;Vehicle detection is carried out to the video flowing using optimal disaggregated model.Unlicensed vehicle checking method of the invention and device, which can not only charge to automate for parking lot entrance, provides help, moreover it is possible to effectively adapt to the complicated variation of various traffic scenes.
Description
Technical field
Unlicensed vehicle checking method and device the present invention relates to intelligent transport technology, in particular in a kind of video flowing.
Background technique
Both of which is mainly used to the management of parking lot entrance vehicle at present.One mode is that ground sensation captures mould
Formula.Ground sensation needs to be laid with ground induction coil on ground, and construction is inconvenient, at high cost.Moreover, some scene ground sensations are not up to
To requirement.Especially, any one ironwork can all be arrived by coil region by candid photograph, will lead to the comparison of false grasp shooting in this way
It is more, such as bicycle, tricycle, motorcycle etc., it will lead to parking lot entrance confusion in this way, the information of interference is more, gives the later period
The processing of charge system increases difficulty.Another mode is manual control candid photograph and mode of arguing for the sake of arguing.In managing system of car parking,
Ultimate target is unattended duty to be carried out, complete barrier free accessibility payment, and this scheme runs in the opposite direction with target.
Summary of the invention
In order to solve to carry out the detection of vehicle for unlicensed vehicle in parking lot entrance, for putting automatically for vehicle of marching into the arena
Row and appearance provide information without license plate number information vehicle match and reduce artificial workload, and the present invention provides a kind of video flowings
In unlicensed vehicle checking method and device.
According to an aspect of the invention, there is provided the unlicensed vehicle checking method in a kind of video flowing.The unlicensed vehicle
Detection method includes the following steps: obtaining from multiple video clips of the video flowing for training linear supporting vector
The training sample set of machine (SVM:support vector machines) disaggregated model, the training sample set include being used as vehicle
The positive sample of sample and negative sample as non-vehicle sample;All positive negative samples are extracted from training sample set obtained
Histograms of oriented gradients (HOG:Histogram of Oriented Gradient) feature;Utilize extracted HOG feature
Off-line training is carried out to the svm classifier model, to obtain the disaggregated model of initial optimization;Utilize the video in the video flowing
The disaggregated model of information on-line training initial optimization again, to obtain optimal disaggregated model;Utilize optimal disaggregated model pair
The video flowing carries out vehicle detection.
In a modification of the present invention embodiment, it is described extracted from training sample set obtained it is all positive and negative
The step of HOG feature of sample includes: to carry out gray processing to sample image;Color is carried out to described image using Gamma correction method
The normalization in space;Calculate the gradient of each pixel in described image;Divide the image into each unit;Count each list
The histogram of gradients of member, that is, obtain description of each unit;The unit of every predetermined quantity is formed into a block, in a block
All units description son be together in series to obtain the HOG feature of the block;By all pieces of the HOG feature series connection in described image
Get up to obtain the feature vector used for training.
It is described again online using the video information in the video flowing in a modification of the present invention embodiment
Scene when the step of disaggregated model of training initial optimization includes: analysis, never vehicle of the basis to the video information
Middle extraction negative sample, also, using the position of license plate as reference, the region around license plate is extracted as positive sample;Using being mentioned
The positive negative sample taken carries out re -training to the disaggregated model of the initial optimization.
It is described that the video flowing is carried out using optimal disaggregated model in a modification of the present invention embodiment
The step of vehicle detection includes: carry out Car license recognition, and determines whether license plate result;In the case where there is license plate result, with vehicle
Memorial tablet is set to standard and automatically intercepts headstock image, and exports license plate result;In the case where no license plate result, judge whether
It is set up background model;In the case where not establishing background model, with local binary patterns (LBP:Local Binary
Pattern) feature carries out background modeling;In the case where having built up background model, sentenced using disaggregated model and background model
The disconnected vehicle with the presence or absence of no license plate result;In the presence of the vehicle of no license plate result, the vehicle conduct that will test
Unlicensed vehicle output.
In a modification of the present invention embodiment, the svm classifier model is two classification about target and background
Model.
According to another aspect of the present invention, the unlicensed vehicle detection apparatus in a kind of video flowing is provided.It is described unlicensed
Vehicle detection apparatus includes: that training sample set obtains module, for being used for from multiple video clips of the video flowing
The training sample set of the linear svm classifier model of training, the training sample set includes the positive sample and work as vehicle sample
For the negative sample of non-vehicle sample;HOG characteristic extracting module, for extracting all positive and negative samples from training sample set obtained
This HOG feature;Off-line training module, for being instructed offline using extracted HOG feature to the svm classifier model
Practice, to obtain the disaggregated model of initial optimization;On-line training module, for being existed again using the video information in the video flowing
The disaggregated model of line training initial optimization, to obtain optimal disaggregated model;Detection module, for utilizing optimal disaggregated model
Vehicle detection is carried out to the video flowing.
In a modification of the present invention embodiment, the HOG characteristic extracting module is also used to: being carried out to sample image
Gray processing;The normalization of color space is carried out to described image using Gamma correction method;Calculate each pixel in described image
Gradient;Divide the image into each unit;Count the histogram of gradients of each unit, that is, obtain the description of each unit
Son;The unit of every predetermined quantity is formed into a block, description of all units in a block is together in series to obtain the block
HOG feature;All pieces of HOG feature in described image is together in series to obtain the feature vector used for training.
In a modification of the present invention embodiment, the on-line training module is also used to: being believed according to the video
Negative sample is extracted in scene when the analysis, never vehicle of breath, also, using the position of license plate as reference, extracts license plate week
The region enclosed is as positive sample;Re -training is carried out using disaggregated model of the extracted positive negative sample to the initial optimization.
In a modification of the present invention embodiment, the detection module is also used to: being carried out Car license recognition, and is judged have
Without license plate result;In the case where there is license plate result, headstock image is automatically intercepted as standard using license plate position, and export license plate
As a result;In the case where no license plate result, judge whether to have built up background model;The case where not establishing background model
Under, background modeling is carried out with LBP feature;In the case where having built up background model, disaggregated model and background model are utilized
Judge whether there is the vehicle of no license plate result;In the presence of the vehicle of no license plate result, the vehicle that will test is made
For the output of unlicensed vehicle.
In a modification of the present invention embodiment, the svm classifier model is two classification about target and background
Model.
According to another aspect of the present invention, a kind of computer storage medium is additionally provided, is deposited in the computer storage medium
Computer program code is contained, which is performed realization as described in any one of claim 1-5
Unlicensed vehicle checking method in video flowing.
In unlicensed vehicle checking method and device in video flowing of the invention, with HOG+SVM object detection method
The detection of unlicensed vehicle in video is carried out, and is merged with license plate result, such correct output without license plate number information vehicle is reached,
Help is provided for parking lot entrance charge automation.Meanwhile the method for on-line training is additionally used, further enhance mould
Type can effectively adapt to the complicated variation of various traffic scenes, for example, night illumination, moves shade, the environment item such as bad weather
Part.
It will be apparent to a skilled person that can be not limited to the objects and advantages that the present invention realizes above specific
It is described, and the above and other purpose that the present invention can be realized will be more clearly understood according to following detailed description.
And it is to be understood that aforementioned description substantially and subsequent detailed description are exemplary illustration and explanation, not
The limitation to the claimed content of the present invention should be used as.
Detailed description of the invention
With reference to the attached drawing of accompanying, the more purposes of the present invention, function and advantage are by the as follows of embodiment through the invention
Description is illustrated, in which:
Fig. 1 is the flow chart according to the unlicensed vehicle checking method of exemplary embodiment of the present invention;
Fig. 2 is the block diagram according to the unlicensed vehicle detection apparatus of exemplary embodiment of the present invention.
Specific embodiment
The preferred embodiment that the present invention will be described in more detail below with reference to accompanying drawings.Although showing the present invention in attached drawing
Preferred embodiment, however, it is to be appreciated that may be realized in various forms the present invention without the embodiment party that should be illustrated here
Formula is limited.On the contrary, thesing embodiments are provided so that the present invention is more thorough and complete, and can will be of the invention
Range is completely communicated to those skilled in the art.
The present invention proposes a kind of based on HOG+ for shortcoming present in existing parking lot entrance management
Unlicensed vehicle checking method and device in the video flowing of SVM.Wherein, it is carried out in video with HOG+SVM object detection method
The detection of unlicensed vehicle, and merged with license plate result, reach such correct output without license plate number information vehicle, goes out for parking lot
Entrance charge automation provides help.Meanwhile the method for on-line training is additionally used, and model is further enhanced, it can be effectively
The complicated variation of various traffic scenes is adapted to, for example, night illumination, moves shade, the environmental conditions such as bad weather.
It is below with reference to accompanying drawings and right in conjunction with specific embodiments in order to be clearer and more clear technical solution of the present invention
The present invention is described in detail.
Fig. 1 shows the flow chart of the unlicensed vehicle checking method of an exemplary embodiment of the present invention.Such as Fig. 1 institute
Show, unlicensed vehicle checking method include training sample set obtain step S1, HOG characteristic extraction step S2, off-line training step S3,
On-line training step S4 and detecting step S5.Each step is detailed below.
Step S1: it is obtained from multiple video clips of collected video flowing for training linear svm classifier model
Training sample set.According to camera in the collected vehicle sample of parking lot entrance, positive sample is trained (that is, vehicle
Sample) mark.The positive sample of mark includes the vehicle sample of various postures and illumination condition.It is special according to scene is taken on site
Property, the region of mark mainly select vehicle face part, rather than entire vehicle body, and contain the background area of part.In addition, using shining
Jpeg stream of the camera when the collected non-vehicle of parking lot entrance is passed through, carries out negative sample with the method for sliding window
(i.e. the sample of non-vehicle may include tricycle, pedestrian, motorcycle, building, ground and any shooting be appeared in addition to vehicle
The target of picture) extraction.Therefore, the training sample set of acquisition includes as the positive sample of vehicle sample and as non-vehicle sample
This negative sample.
Step S2: the training sample set obtained from step S1 extracts the HOG feature of all positive negative samples.Firstly, right
Sample image carries out gray processing and needs image being considered as 3-D image here.Then, after using Gamma correction method to gray processing
Image carry out color space normalization.Normalized purpose is the contrast in order to adjust image, reduces image local
Influence caused by shade and illumination variation, while the interference of noise can be inhibited.Then, in the image after calculating normalization
The gradient of each pixel, including gradient magnitude and direction.This is done to capture profile information, and further, weakened light shines
Interference.Next, small unit (cell) is divided an image into, for example, each cell size is 8*8 pixel.Then, it counts
The histogram of gradients of each cell, that is, the description for obtaining each unit is sub (descriptor).Then, by every predetermined quantity
Unit forms a block (block), and the descriptor of all cell in a block is together in series to obtain the HOG of the block
Feature descriptor.Finally, the HOG feature descriptor of all block in image is together in series to obtain for training
The feature vector used.Generalized time and effect consider that the size of training sample is selected as 128*64 pixel by inventor, often
One cell size is 8*8 pixel size, and 2*2 cell forms a block, and angular range selection is 0 to 360 degree,
18 equal parts finally obtain the feature vector of 7560 dimensions.
Step S3: carrying out off-line training to svm classifier model using the HOG feature extracted in step S2, initial to obtain
The disaggregated model of optimization.Here, svm classifier model is two linear disaggregated models.In this step, firstly, setting includes SVM
The parameters such as parameter, the number of iterations, the error of disaggregated model.Then, it based on set parameter, will be concentrated from training sample at random
The HOG feature input svm classifier model of the training sample of the certain amount (for example, 1000 to 10000) of extraction is trained, and is obtained
To the disaggregated model of initial optimization.
Step S4: using the disaggregated model of the video information in video flowing again on-line training initial optimization, to obtain most
Excellent disaggregated model.Specifically, firstly, extracting negative sample in scene when according to analysis, never vehicle to video information,
Also, using the position of license plate as reference, the region around license plate is extracted as positive sample.Then, using extracted positive and negative
Sample carries out re -training to the disaggregated model of initial optimization, so as to improve verification and measurement ratio and reduce false detection rate.
Step S5: vehicle detection is carried out to video flowing using disaggregated model optimal obtained in step S4.In the step
In, firstly, carrying out Car license recognition, and determine whether license plate result.In the case where there is license plate result, using license plate position as standard
Headstock image is automatically intercepted, and exports license plate result.However, judging whether to have built up in the case where no license plate result
Background model.
On the other hand, if having built up background model, nothing is judged whether there is using disaggregated model and background model
The vehicle of license plate result.Moreover, the vehicle that will test is defeated as unlicensed vehicle in the presence of the vehicle of no license plate result
Out.
It follows that vehicle detection result is merged with license plate result in detecting step S5.If passing through disaggregated model
And background modeling, had been acknowledged that vehicle passes through, but for there is board vehicle, need preferentially to export license plate as a result, if this
Process does not have license plate as a result, the vehicle that can be will test is exported as unlicensed vehicle.Guarantee that a vehicle comes, and has license plate in this way
Number, output license plate number without license plate number as a result, capture to vehicle pictures.
Using the unlicensed vehicle checking method of the present embodiment, can provide license plate recognition result at parking lot entrance and
Continuous video flowing, and complete the output without license plate number vehicle.Meanwhile to unlicensed license number and there can be license plate in Entrance
The vehicle of number result carries out clearance of arguing for the sake of arguing automatically.Furthermore it is also possible in parking exit further to the vehicle for having license plate number
It charges automatically by the matching of license plate number, for the vehicle of no license plate number, then searches for vehicle of the entrance without license plate number information, carry out
It is artificial to search, or Auto-matching is carried out, it finds information of marching into the arena and charges.
The present invention also provides a kind of devices for realizing above-mentioned unlicensed vehicle checking method.Fig. 2 shows according to this hair
The block diagram of the unlicensed vehicle detection apparatus 100 of bright exemplary embodiment.As shown in Fig. 2, unlicensed vehicle detection apparatus 100 includes
Training sample set obtains module 101, HOG characteristic extracting module 102, off-line training module 103, on-line training module 104 and inspection
Survey module 105.
Training sample set obtains module 101 and is used to obtained from multiple video clips of video flowing for training linearly
The training sample set of svm classifier model.According to camera in the collected vehicle sample of parking lot entrance, it is trained positive sample
The mark of this (that is, vehicle sample).The positive sample of mark includes the vehicle sample of various postures and illumination condition.It is clapped according to scene
Take the photograph scene characteristics, the region of mark mainly selects vehicle face part, rather than entire vehicle body, and contains the background area of part.This
Outside, using camera when the collected non-vehicle of parking lot entrance is passed through jpeg stream, with sliding window method into
(i.e. the sample of non-vehicle may include tricycle, pedestrian, motorcycle, building, ground and any except vehicle is outgoing to row negative sample
The target of present shooting picture) extraction.Therefore, the training sample set of acquisition includes the positive sample and conduct as vehicle sample
The negative sample of non-vehicle sample.
HOG characteristic extracting module 102, which is used to obtain the training sample set that module 101 obtains from training sample set and extracts, to be owned
Positive negative sample HOG feature.Firstly, needing image being considered as 3-D image here to sample image progress gray processing.So
Afterwards, the normalization of color space is carried out to the image after gray processing using Gamma correction method.Normalized purpose is to adjust
The contrast of image, reduce image local shade and illumination variation caused by influence, while the interference of noise can be inhibited.
Then, the gradient of each pixel in image after calculating normalization, including gradient magnitude and direction.This is done to capture
Profile information, and the interference that further weakened light shines.Next, small unit (cell) is divided an image into, for example, each
Cell size is 8*8 pixel.Then, the histogram of gradients of each cell is counted, that is, obtain description of each unit
(descriptor).Then, the unit of every predetermined quantity is formed into a block (block), all cell's in a block
Descriptor is together in series to obtain the HOG feature descriptor of the block.Finally, the HOG of all block in image is special
Sign descriptor is together in series to obtain the feature vector used for training.Generalized time and effect consider that inventor will train
Size is selected as 128*64 pixel, each cell size is 8*8 pixel size, and 2*2 cell forms one
Block, angular range selection is 0~360 degree, and 18 equal parts finally obtain the feature vector of 7560 dimensions.
Off-line training module 103 be used for using HOG characteristic extracting module 102 extract HOG feature to svm classifier model into
Row off-line training, to obtain the disaggregated model of initial optimization.Here, svm classifier model is two linear disaggregated models.In the step
In rapid, firstly, setting includes the parameters such as parameter, the number of iterations, the error of svm classifier model.Then, based on set ginseng
Number, will concentrate the HOG feature of the training sample for the certain amount (for example, 1000 to 10000) randomly selected defeated from training sample
Enter svm classifier model to be trained, obtains the disaggregated model of initial optimization.
On-line training module 104 is used for the classification mould using the video information in video flowing again on-line training initial optimization
Type, to obtain optimal disaggregated model.Specifically, firstly, in scene when according to analysis, never vehicle to video information
Negative sample is extracted, also, using the position of license plate as reference, extracts the region around license plate as positive sample.Then, institute is utilized
The positive negative sample extracted carries out re -training to the disaggregated model of initial optimization, so as to improve verification and measurement ratio and reduce erroneous detection
Rate.
The optimal disaggregated model that detection module 105 is used to obtain using on-line training module 104 carries out vehicle to video flowing
Detection.Firstly, detection module 105 carries out Car license recognition, and determine whether license plate result.In the case where there is license plate result,
Headstock image is automatically intercepted as standard using license plate position, and exports license plate result.However, in the case where no license plate result,
Judge whether to have built up background model.
If not establishing background model, detection module 105 carries out background modeling with LBP feature.Specifically, resonable
Near the output position for the vehicle thought, one piece of small region is selected, carries out background modeling with LBP feature.It is equivalent to and establishes
The trigger area of one simulation, when there is target by this part region, LBP characteristic value can deviate background value, when target is left
When, the value under background state can be restored.The process for extracting LBP is that original image is converted to LBP figure first, is then counted
The LBP histogram of LBP figure, and original image is indicated with the histogram of this vector form.Light can be overcome with LBP feature
According to interference, trigger it is sensitive, accurate.There are two the purposes done so: i) when a vehicle passes through, multiframe picture can be handled,
With the available repeated detection window of disaggregated model, but for actual application, same vehicle passes through, it is only necessary to have a knot
Fruit, that is to say, that vehicle must be occurred being divided into a process to disappearance, in this process, guaranteed as far as possible in a fixation
Position export primary result;Ii) when simulation trigger area changes, just output as a result, guarantee solid at one as far as possible
Fixed position result output.Occur using the carry out vehicle of LBP characteristic background modeling as follows to disappearance cutting procedure: if 1) existed
Simulation trigger area has continuous multiple frames not move, then extracts the LBP histogram of current region, and expression establishes background.If
It does not move, then constantly updates background characteristics value;If 2) detect movement, present day analog trigger area prospect and back are calculated
The distance of the LBP histogram of scape, if distance has been greater than fixed threshold, then it is assumed that current region is capped, it may be possible to vehicle
It blocks or what other were interfered blocks, need according to single frames picture as a result, judging whether it is that vehicle passes through;3) continue to calculate and work as
The distance of the LBP histogram of front simulation trigger area foreground and background is recognized if being less than fixed threshold value apart from lasting multiframe
It is had been moved off for current vehicle.For example, when a vehicle comes, the LBP histogram for the simulation trigger area that present frame calculates and
The distance of background histogram can be greater than specified threshold value, this is to represent the region to be capped.If vehicle is walked, triggering is simulated
The LBP histogram in region can be less than specified threshold value at a distance from background histogram, and expression has restored background, only restore back
Scape state just allows the triggering of next vehicle.
On the other hand, if having built up background model, nothing is judged whether there is using disaggregated model and background model
The vehicle of license plate result.Moreover, the vehicle that will test is defeated as unlicensed vehicle in the presence of the vehicle of no license plate result
Out.
As an example, unlicensed vehicle detection apparatus of the invention may include memory and processor, deposit on memory
Computer program code is contained, when processor is configured as executing the computer program code stored on memory, is realized such as
The step S1-S5 of the preceding unlicensed vehicle checking method.
In conclusion the unlicensed vehicle checking method and device of the present embodiment, carry out with HOG+SVM object detection method
The detection of unlicensed vehicle in video, and merged with license plate result, reach such correct output without license plate number information vehicle, to stop
Entrance charge automation in parking lot provides help.Meanwhile the method for on-line training is additionally used, further enhance model, energy
The complicated variation of various traffic scenes is effectively adapted to, for example, night illumination, moves shade, the environmental conditions such as bad weather.
Each section of the invention can be realized with hardware, software, firmware or their combination.In above embodiment
In, software or firmware that multiple steps or method can be executed in memory and by suitable instruction execution system with storage come
It realizes.For example, if realized with hardware, in another embodiment, the known following technology in this field can be used
Any one of or their combination realize: have for data-signal is realized the logic gates of logic function from
Logic circuit is dissipated, the specific integrated circuit with suitable combinational logic gate circuit, programmable gate array (PGA), scene can compile
Journey gate array (FPGA) etc..
The logic and/or step for indicating or describing in other ways herein in flow charts, for example, being considered
For realizing the order list of the executable instruction of logic function, may be embodied in any computer-readable medium, with
For instruction execution system, device or equipment (such as computer based system, including the system of processor or other can be from instruction
Execute system, device or equipment instruction fetch and the system that executes instruction) use, or combine these instruction execution systems, device or
Equipment and use.
As above it describes for one embodiment and/or the feature that shows can be in a manner of same or similar at one or more
It is used in a number of other embodiments, and/or combines or substitute the feature in other embodiments with the feature in other embodiments
It uses.
In conjunction with the explanation and practice of the invention disclosed here, the other embodiment of the present invention is for those skilled in the art
It all will be readily apparent and understand.Illustrate and embodiment is regarded only as being exemplary, true scope of the invention and purport are equal
It is defined in the claims.
Claims (11)
1. the unlicensed vehicle checking method in a kind of video flowing, which is characterized in that the unlicensed vehicle checking method includes following
Step:
The training sample set for training linear svm classifier model, the instruction are obtained from multiple video clips of video flowing
Practicing sample set includes the positive sample as vehicle sample and the negative sample as non-vehicle sample;
The HOG feature of all positive negative samples is extracted from training sample set obtained;
Off-line training is carried out to the svm classifier model using extracted HOG feature, to obtain the classification mould of initial optimization
Type;
Using the disaggregated model of the video information in the video flowing again on-line training initial optimization, to obtain optimal classification
Model;
Vehicle detection is carried out to the video flowing using optimal disaggregated model.
2. unlicensed vehicle checking method according to claim 1, which is characterized in that described from training sample obtained
Collection extracts the step of HOG feature of all positive negative sample and includes:
Gray processing is carried out to sample image;
The normalization of color space is carried out to described image using Gamma correction method;
Calculate the gradient of each pixel in described image;
Divide the image into each unit;
Count the histogram of gradients of each unit, that is, obtain description of each unit;
The unit of every predetermined quantity is formed into a block, description of all units in a block is together in series to obtain the block
HOG feature;
All pieces of HOG feature in described image is together in series to obtain the feature vector used for training.
3. unlicensed vehicle checking method according to claim 1, which is characterized in that described using in the video flowing
Video information again the disaggregated model of on-line training initial optimization the step of include:
Negative sample is extracted in scene when according to analysis, never vehicle to the video information, also, with the position of license plate
As reference, the region around license plate is extracted as positive sample;
Re -training is carried out using disaggregated model of the extracted positive negative sample to the initial optimization.
4. unlicensed vehicle checking method according to claim 1, which is characterized in that described to utilize optimal disaggregated model
Include: to the step of video flowing progress vehicle detection
Car license recognition is carried out, and determines whether license plate result;
In the case where there is license plate result, headstock image is automatically intercepted as standard using license plate position, and export license plate result;
In the case where no license plate result, judge whether to have built up background model;
In the case where not establishing background model, background modeling is carried out with LBP feature;
In the case where having built up background model, no license plate result is judged whether there is using disaggregated model and background model
Vehicle;
In the presence of the vehicle of no license plate result, the vehicle that will test is exported as unlicensed vehicle.
5. unlicensed vehicle checking method according to claim 1, which is characterized in that the svm classifier model is about mesh
Two disaggregated models of mark and background.
6. the unlicensed vehicle detection apparatus in a kind of video flowing, which is characterized in that the unlicensed vehicle detection apparatus includes:
Training sample set obtains module, for obtaining from multiple video clips of the video flowing for training linear SVM
The training sample set of disaggregated model, the training sample set include as the positive sample of vehicle sample and as non-vehicle sample
Negative sample;
HOG characteristic extracting module, for extracting the HOG feature of all positive negative samples from training sample set obtained;
Off-line training module, for carrying out off-line training to the svm classifier model using extracted HOG feature, to obtain
The disaggregated model of initial optimization;
On-line training module, for the classification mould using the video information in the video flowing again on-line training initial optimization
Type, to obtain optimal disaggregated model;
Detection module, for carrying out vehicle detection to the video flowing using optimal disaggregated model.
7. unlicensed vehicle detection apparatus according to claim 6, which is characterized in that the HOG characteristic extracting module is also used
In:
Gray processing is carried out to sample image;
The normalization of color space is carried out to described image using Gamma correction method;
Calculate the gradient of each pixel in described image;
Divide the image into each unit;
Count the histogram of gradients of each unit, that is, obtain description of each unit;
The unit of every predetermined quantity is formed into a block, description of all units in a block is together in series to obtain the block
HOG feature;
All pieces of HOG feature in described image is together in series to obtain the feature vector used for training.
8. unlicensed vehicle detection apparatus according to claim 6, which is characterized in that the on-line training module is also used to:
Negative sample is extracted in scene when according to analysis, never vehicle to the video information, also, with the position of license plate
As reference, the region around license plate is extracted as positive sample;
Re -training is carried out using disaggregated model of the extracted positive negative sample to the initial optimization.
9. unlicensed vehicle detection apparatus according to claim 6, which is characterized in that the detection module is also used to:
Car license recognition is carried out, and determines whether license plate result;
In the case where there is license plate result, headstock image is automatically intercepted as standard using license plate position, and export license plate result;
In the case where no license plate result, judge whether to have built up background model;
In the case where not establishing background model, background modeling is carried out with LBP feature;
In the case where having built up background model, no license plate result is judged whether there is using disaggregated model and background model
Vehicle;
In the presence of the vehicle of no license plate result, the vehicle that will test is exported as unlicensed vehicle.
10. unlicensed vehicle detection apparatus according to claim 6, which is characterized in that the svm classifier model is about mesh
Two disaggregated models of mark and background.
11. a kind of computer storage medium, which is characterized in that it is stored with computer program code in the computer storage medium,
The computer program code is performed the unlicensed vehicle inspection realized in the video flowing as described in any one of claim 1-5
Survey method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810821013.3A CN109034171B (en) | 2018-07-24 | 2018-07-24 | Method and device for detecting unlicensed vehicles in video stream |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810821013.3A CN109034171B (en) | 2018-07-24 | 2018-07-24 | Method and device for detecting unlicensed vehicles in video stream |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109034171A true CN109034171A (en) | 2018-12-18 |
CN109034171B CN109034171B (en) | 2022-10-04 |
Family
ID=64645606
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810821013.3A Active CN109034171B (en) | 2018-07-24 | 2018-07-24 | Method and device for detecting unlicensed vehicles in video stream |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109034171B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113256991A (en) * | 2021-07-14 | 2021-08-13 | 智广海联(天津)大数据技术有限公司 | Traffic flow statistics-based automobile plant peripheral road congestion management system and method |
CN113793508A (en) * | 2021-09-27 | 2021-12-14 | 深圳市芊熠智能硬件有限公司 | Entrance and exit unlicensed vehicle anti-interference rapid detection method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104820998A (en) * | 2015-05-27 | 2015-08-05 | 成都通甲优博科技有限责任公司 | Human body detection and tracking method and device based on unmanned aerial vehicle mobile platform |
KR20160062880A (en) * | 2014-11-26 | 2016-06-03 | 휴앤에스(주) | road traffic information management system for g using camera and radar |
CN105702048A (en) * | 2016-03-23 | 2016-06-22 | 武汉理工大学 | Automobile-data-recorder-based illegal lane occupation identification system and method for automobile on highway |
-
2018
- 2018-07-24 CN CN201810821013.3A patent/CN109034171B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20160062880A (en) * | 2014-11-26 | 2016-06-03 | 휴앤에스(주) | road traffic information management system for g using camera and radar |
CN104820998A (en) * | 2015-05-27 | 2015-08-05 | 成都通甲优博科技有限责任公司 | Human body detection and tracking method and device based on unmanned aerial vehicle mobile platform |
CN105702048A (en) * | 2016-03-23 | 2016-06-22 | 武汉理工大学 | Automobile-data-recorder-based illegal lane occupation identification system and method for automobile on highway |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113256991A (en) * | 2021-07-14 | 2021-08-13 | 智广海联(天津)大数据技术有限公司 | Traffic flow statistics-based automobile plant peripheral road congestion management system and method |
CN113256991B (en) * | 2021-07-14 | 2021-09-24 | 智广海联(天津)大数据技术有限公司 | Traffic flow statistics-based automobile plant peripheral road congestion management system and method |
CN113793508A (en) * | 2021-09-27 | 2021-12-14 | 深圳市芊熠智能硬件有限公司 | Entrance and exit unlicensed vehicle anti-interference rapid detection method |
Also Published As
Publication number | Publication date |
---|---|
CN109034171B (en) | 2022-10-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104392468B (en) | Based on the moving target detecting method for improving visual background extraction | |
CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
CN103310444B (en) | A kind of method of the monitoring people counting based on overhead camera head | |
CN106372666B (en) | A kind of target identification method and device | |
CN109559302A (en) | Pipe video defect inspection method based on convolutional neural networks | |
CN109376667A (en) | Object detection method, device and electronic equipment | |
CN103268470B (en) | Object video real-time statistical method based on any scene | |
CN109298785A (en) | A kind of man-machine joint control system and method for monitoring device | |
CN110084165A (en) | The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations | |
CN102043953A (en) | Real-time-robust pedestrian detection method aiming at specific scene | |
CN108764167A (en) | A kind of target of space time correlation recognition methods and system again | |
CN109740590A (en) | The accurate extracting method of ROI and system based on target following auxiliary | |
CN108805016A (en) | A kind of head and shoulder method for detecting area and device | |
CN104050481A (en) | Multi-template infrared image real-time pedestrian detection method combining contour feature and gray level | |
CN103593679A (en) | Visual human-hand tracking method based on online machine learning | |
CN109558790B (en) | Pedestrian target detection method, device and system | |
CN110084987A (en) | A kind of foreign matter inspecting system and method towards rail traffic | |
Tung et al. | Large-scale object detection of images from network cameras in variable ambient lighting conditions | |
Malhi et al. | Vision based intelligent traffic management system | |
CN110633678A (en) | Rapid and efficient traffic flow calculation method based on video images | |
CN109034171A (en) | Unlicensed vehicle checking method and device in video flowing | |
CN112434566A (en) | Passenger flow statistical method and device, electronic equipment and storage medium | |
CN113935395A (en) | Training of object recognition neural networks | |
CN109165872A (en) | Unmanned intelligent sphere course management system, method and court | |
CN104778676A (en) | Depth ranging-based moving target detection method and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |