CN109145715A - The space base pedestrian of rail traffic invades boundary's detection method, device and system - Google Patents
The space base pedestrian of rail traffic invades boundary's detection method, device and system Download PDFInfo
- Publication number
- CN109145715A CN109145715A CN201810710554.9A CN201810710554A CN109145715A CN 109145715 A CN109145715 A CN 109145715A CN 201810710554 A CN201810710554 A CN 201810710554A CN 109145715 A CN109145715 A CN 109145715A
- Authority
- CN
- China
- Prior art keywords
- image
- space base
- ratio
- battery limit
- invading
- 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
- 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
- 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
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Abstract
The space base pedestrian that the present invention provides a kind of rail traffic invades boundary's detection method, device and system, pass through the ratio according to pixel distance and actual distance in space base orbital image, it is determined in the space base orbital image and prevents invading battery limit (BL) area image, the space base orbital image is the image that space base image capture device takes vertically downward along track；Prevent that invading battery limit (BL) area image is cut at least two subregion figure for described, and obtaining from least two subregion figure includes the portrait class administrative division map for invading boundary pedestrian；According to position of the administrative division map of the portrait class in the space base orbital image and the corresponding scene location of the space base orbital image, determine described in invade the position of boundary pedestrian, so that the space base pedestrian for improving rail traffic invades boundary's accuracy of detection.
Description
Technical field
The present invention relates to the space base pedestrians of signal processing technology more particularly to a kind of rail traffic to invade boundary's detection method, dress
It sets and system.
Background technique
Rail traffic is the main artery of national economy, is the backbone of China's multimodal transport network, sends out in modern transportation system
Wave very important effect.Since railroad track is outdoor laying mostly, there may be village, field or public affairs along the line
Road.Although the isolation facilities such as fence, wire netting usually can be arranged along the railway, it often will appear pedestrian's injection range
Pedestrian invade zone phenomenon, and in order to avoid crashing, need to invade boundary to pedestrian and be used for quickly detecting, to take in time
Brake measure notifies administrative staff to prevent boundary's behavior is invaded.
It is usually monitoring devices such as embedding proximity detector along the railway that existing pedestrian, which invades boundary's detection method, is being detected
To when having pedestrian, vehicle proximal or into orbital region, sounding an alarm or send warning message to neighbouring administrative center.
However as the development of rail system, a large amount of detectors, but detector are arranged along the track to extend in all direction
False alarm rate it is higher, it is not high that existing pedestrian invades boundary's detection method reliability.
Summary of the invention
The space base pedestrian that the present invention provides a kind of rail traffic invades boundary's detection method, device and system, improves pedestrian and invades
The detection accuracy and efficiency on boundary.
According to the first aspect of the invention, the space base pedestrian for providing a kind of rail traffic invades boundary's detection method, comprising:
According to the ratio of pixel distance and actual distance in space base orbital image, determined in the space base orbital image anti-
Battery limit (BL) area image is invaded, the space base orbital image is the image that space base image capture device takes vertically downward along track；
Prevent that invading battery limit (BL) area image is cut at least two subregion figure for described, and is obtained from least two subregion figure
Taking includes the portrait class administrative division map for invading boundary pedestrian；
According to position of the administrative division map of the portrait class in the space base orbital image and the space base orbital image pair
The scene location answered, determine described in invade the position of boundary pedestrian.
Optionally, described according to pixel distance in space base orbital image in a kind of possible implementation of first aspect
With the ratio of actual distance, is determined in the space base orbital image and prevents invading battery limit (BL) area image, comprising:
Judge whether the ratio of pixel distance and actual distance is greater than proportion threshold value in space base orbital image；
If so, the space base orbital image is determined as preventing invading battery limit (BL) area image；
If it is not, then preventing invading battery limit (BL) domain detection model with preset, is determined in the space base orbital image and prevent invading battery limit (BL) domain
Selected frame be determined as described anti-invading battery limit (BL) area image and by the image in the selected frame.
Optionally, in the alternatively possible implementation of first aspect, described according to pixel in space base orbital image
The ratio of distance and actual distance is determined in the space base orbital image before preventing invading battery limit (BL) area image, further includes:
Obtain the horizontal pixel distance and longitudinal direction pixel distance of the space base orbital image；
The corresponding photographing information of the space base orbital image is obtained, the photographing information includes shooting height information and shooting
Viewing-angle information；
According to the corresponding photographing information of the space base orbital image, it is true to obtain the corresponding transverse direction of the space base orbital image
Distance and longitudinal actual distance；
By the ratio of the horizontal pixel distance and the lateral actual distance, it is determined as the first ratio；
By the ratio of longitudinal pixel distance and longitudinal actual distance, it is determined as the second ratio；
By the minimum scale in first ratio and the second ratio, it is determined as the ratio of the pixel distance and actual distance
Example.
Optionally, in another possible implementation of first aspect, described with preset Fang Qin circle region detection
Model is determined in the space base orbital image before preventing invading the selected frame in battery limit (BL) domain, further includes:
Faster-RCNN calculation is carried out to the anti-battery limit (BL) domain detection model of invading with the true classification that history invades battery limit (BL) area image
Method training, it is and described so that the anti-prediction classification invaded battery limit (BL) domain detection model and invade the history output of battery limit (BL) area image
History invades the penalty values between the true classification of battery limit (BL) area image and is lower than default Faster-RCNN loss threshold value.
Optionally, described to be obtained from least two subregion figure in another possible implementation of first aspect
Taking includes the portrait class administrative division map for invading boundary pedestrian, comprising:
Obtain the histograms of oriented gradients HOG characteristic information of each subregion figure；
With preset portrait detection model, portrait class HOG characteristic information is determined in each HOG characteristic information, wherein
The portrait detection model is to be supported the training of vector machine SVM algorithm with history portrait class HOG characteristic information to obtain；
By the corresponding subregion figure of the portrait class HOG characteristic information, it is determined as including the portrait for invading boundary pedestrian
Class administrative division map.
Optionally, in another possible implementation of first aspect, described with preset portrait detection model,
In each HOG characteristic information before determining portrait class HOG characteristic information, further includes:
If the ratio of pixel distance and actual distance is greater than proportion threshold value in space base orbital image, with the first history portrait
The true classification of class HOG characteristic information carries out SVM algorithm training to the portrait detection model, so that the portrait detection model
To the prediction classification of the first history portrait class HOG characteristic information output, with the first history portrait class HOG characteristic information
True classification between penalty values be lower than default SVM and lose threshold value, wherein the first history portrait class HOG characteristic information
The ratio of pixel distance and actual distance is greater than proportion threshold value in corresponding history space base orbital image；
If the ratio of pixel distance and actual distance is less than proportion threshold value in space base orbital image, with the second history portrait
The true classification of class HOG characteristic information carries out SVM algorithm training to the portrait detection model, so that the portrait detection model
To the prediction classification of the second history portrait class HOG characteristic information output, with the second history portrait class HOG characteristic information
True classification between penalty values be lower than default SVM and lose threshold value, wherein the second history portrait class HOG characteristic information
The ratio of pixel distance and actual distance is less than proportion threshold value in corresponding history space base orbital image.
According to the second aspect of the invention, the space base pedestrian for providing a kind of rail traffic invades boundary's detection device, comprising:
Anti- battery limit (BL) area image of invading obtains module, for the ratio according to pixel distance in space base orbital image and actual distance
Example determines in the space base orbital image and prevents invading battery limit (BL) area image, the space base orbital image is space base image capture device
The image taken vertically downward along track；
Portrait class administrative division map obtains module, for preventing that invading battery limit (BL) area image is cut at least two subregion figure for described, and
Obtaining from least two subregion figure includes the portrait class administrative division map for invading boundary pedestrian；
Position determination module, for position and institute of the administrative division map according to the portrait class in the space base orbital image
The corresponding scene location of space base orbital image is stated, the position of boundary pedestrian is invaded described in determination.
Optionally, in a kind of possible implementation of second aspect, the anti-battery limit (BL) area image of invading obtains module, specifically
For judging whether the ratio of pixel distance and actual distance is greater than proportion threshold value in space base orbital image；If so, will be described
Space base orbital image is determined as preventing invading battery limit (BL) area image；If it is not, then preventing invading battery limit (BL) domain detection model with preset, in the sky
It determines the anti-selected frame for invading battery limit (BL) domain in base orbital image, and by the image in the selected frame, is determined as described anti-invading battery limit (BL)
Area image.
Optionally, in the alternatively possible implementation of second aspect, the anti-battery limit (BL) area image of invading obtains module, also
For being determined in the space base orbital image in the ratio according to pixel distance and actual distance in space base orbital image
It is anti-to invade before battery limit (BL) area image, obtain the horizontal pixel distance and longitudinal direction pixel distance of the space base orbital image；Described in acquisition
The corresponding photographing information of space base orbital image, the photographing information include shooting height information and shooting visual angle information；According to institute
The corresponding photographing information of space base orbital image is stated, the corresponding lateral actual distance of the space base orbital image is obtained and longitudinal direction is true
Distance；By the ratio of the horizontal pixel distance and the lateral actual distance, it is determined as the first ratio；By longitudinal pixel
The ratio of distance and longitudinal actual distance, is determined as the second ratio；By the minimum in first ratio and the second ratio
Ratio is determined as the ratio of the pixel distance and actual distance.
Optionally, in another possible implementation of second aspect, the anti-battery limit (BL) area image of invading obtains module, also
For preventing invading battery limit (BL) domain detection model with preset described, is determined in the space base orbital image and prevent invading the selected of battery limit (BL) domain
Before frame, Faster-RCNN calculation is carried out to the anti-battery limit (BL) domain detection model of invading with the true classification that history invades battery limit (BL) area image
Method training, it is and described so that the anti-prediction classification invaded battery limit (BL) domain detection model and invade the history output of battery limit (BL) area image
History invades the penalty values between the true classification of battery limit (BL) area image and is lower than default Faster-RCNN loss threshold value.
Optionally, in another possible implementation of second aspect, the portrait class administrative division map obtains module, specifically
For obtaining the histograms of oriented gradients HOG characteristic information of each subregion figure；With preset portrait detection model, in each institute
It states and determines portrait class HOG characteristic information in HOG characteristic information, wherein the portrait detection model is special with history portrait class HOG
Reference breath is supported what the training of vector machine SVM algorithm obtained；By the corresponding subregion of the portrait class HOG characteristic information
Figure, is determined as including the portrait class administrative division map for invading boundary pedestrian.
Optionally, in another possible implementation of second aspect, the portrait class administrative division map obtains module, also uses
In described with preset portrait detection model, before determining portrait class HOG characteristic information in each HOG characteristic information,
It is special with the first history portrait class HOG if the ratio of pixel distance and actual distance is greater than proportion threshold value in space base orbital image
The true classification of reference breath carries out SVM algorithm training to the portrait detection model, so that the portrait detection model is to described
The prediction classification of first history portrait class HOG characteristic information output is true with the first history portrait class HOG characteristic information
Penalty values between classification are lower than default SVM and lose threshold value, wherein the first history portrait class HOG characteristic information is corresponding
The ratio of pixel distance and actual distance is greater than proportion threshold value in history space base orbital image；If in space base orbital image pixel away from
It is less than proportion threshold value from the ratio of actual distance, then with the true classification of the second history portrait class HOG characteristic information to described
Portrait detection model carries out SVM algorithm training, so that the portrait detection model believes the second history portrait class HOG feature
The prediction classification for ceasing output, the penalty values between the true classification of the second history portrait class HOG characteristic information are lower than pre-
If SVM loses threshold value, wherein pixel in the corresponding history space base orbital image of the second history portrait class HOG characteristic information
Distance and the ratio of actual distance are less than proportion threshold value.
According to the third aspect of the invention we, the space base pedestrian for providing a kind of rail traffic invades boundary's detection system, including track
The space base pedestrian of traffic invades boundary's detection device and at least one space base image capture device；
The space base image capture device, for acquiring space base orbital image；
The space base pedestrian of the rail traffic invades boundary's detection device, is used for from space base image capture device described at least one
The space base orbital image is obtained, and perform claim requires the space base pedestrian of 1~6 any rail traffic to invade detection side of boundary
Method.
According to the fourth aspect of the invention, the space base pedestrian for providing a kind of rail traffic invades boundary's detection device, comprising: storage
Device, processor and computer program, in the memory, the processor runs the meter for the computer program storage
Calculation machine program executes the method for first aspect present invention and the various possible designs of first aspect.
According to the fifth aspect of the invention, a kind of readable storage medium storing program for executing is provided, meter is stored in the readable storage medium storing program for executing
Calculation machine program, the computer program are performed for realizing the various possible designs of first aspect present invention and first aspect
The method.
A kind of space base pedestrian of rail traffic provided by the invention invades boundary's detection method, device and system, by according to sky
The ratio of pixel distance and actual distance in base orbital image, it is determining in the space base orbital image to prevent invading battery limit (BL) area image,
The space base orbital image is the image that space base image capture device takes vertically downward along track；Prevent invading battery limit (BL) domain by described
Image is cut at least two subregion figure, and obtaining from least two subregion figure includes the portrait class for invading boundary pedestrian
Administrative division map；It is corresponding according to position of the administrative division map of the portrait class in the space base orbital image and the space base orbital image
Scene location, determine described in invade the position of boundary pedestrian, so that the space base pedestrian for improving rail traffic invades boundary's accuracy of detection.
Detailed description of the invention
Fig. 1 is that a kind of space base pedestrian of rail traffic provided in an embodiment of the present invention invades the application scenarios of boundary's detection system and shows
It is intended to；
Fig. 2 is that a kind of space base pedestrian of rail traffic provided in an embodiment of the present invention invades the process signal of boundary's detection method
Figure；
Fig. 3 is that the space base pedestrian of another rail traffic provided in an embodiment of the present invention invades the process signal of boundary's detection method
Figure；
Fig. 4 is that the space base pedestrian of another rail traffic provided in an embodiment of the present invention invades the process signal of boundary's detection method
Figure；
Fig. 5 is that a kind of space base pedestrian of rail traffic provided in an embodiment of the present invention invades the structural representation of boundary's detection device
Figure；
Fig. 6 is that a kind of space base pedestrian of rail traffic provided in an embodiment of the present invention invades the hardware configuration of boundary's detection device and shows
It is intended to.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is only
It is only a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill
Personnel's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Description and claims of this specification and term " first ", " second ", " third " " in above-mentioned attached drawing
The (if present)s such as four " are to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should manage
The data that solution uses in this way are interchangeable under appropriate circumstances, so as to the embodiment of the present invention described herein can in addition to
Here the sequence other than those of diagram or description is implemented.
It should be appreciated that the size of the serial number of each process is not meant to execute sequence in the various embodiments of the application
It is successive, the execution of each process sequence should be determined by its function and internal logic, the implementation without coping with the embodiment of the present application
Journey constitutes any restriction.
It should be appreciated that in this application, " comprising " and " having " and their any deformation, it is intended that covering is not arranged
His includes, for example, the process, method, system, product or equipment for containing a series of steps or units are not necessarily limited to clearly
Those of list step or unit, but may include be not clearly listed or for these process, methods, product or equipment
Intrinsic other step or units.
It should be appreciated that in this application, " multiple " refer to two or more."and/or" is only a kind of description pass
Join object incidence relation, indicate may exist three kinds of relationships, for example, and/or B, can indicate: individualism A is existed simultaneously
These three situations of A and B, individualism B.Character "/" typicallys represent the relationship that forward-backward correlation object is a kind of "or".It " include A, B
And C " refers to that A, B, C three include,
It should be appreciated that in this application, " B corresponding with A ", " A and B are corresponding " or " B and A are corresponding " indicate B
It is associated with A, B can be determined according to A.Determine that B is not meant to determine B only according to A according to A, can also according to A and/or
Other information determines B.The matching of A and B is that the similarity of A and B is greater than or equal to preset threshold value.
It should be appreciated that in this application, Faster-RCNN algorithm is a kind of object detection calculation based on convolutional neural networks
Method.Faster-RCNN determines candidate region (region proposals) using neural network, and Faster-RCNN is in Fast-
The mode of shared convolutional layer is used on the basis of RCNN, the characteristic pattern after convolution is used to generate candidate region.By in Fast-
Increase by two convolutional layers on the basis of RCNN to realize that network (Region Proposal Networks, letter are extracted in candidate region
Claim: RPNs), one of convolutional layer is used to the position encoded at a vector of each characteristic pattern, another convolutional layer is then
The recurrence boundary of objectifiability scoring (an objectness score) and the candidate region k are exported to each position
(regressed bounds for k region proposals)。
It should be appreciated that in this application, ZF-net algorithm is another algorithm based on convolutional neural networks, feature exists
It is then based on the fine tuning algorithm of Alex-net progress, has used ReLU activation primitive, cross entropy cost function, and use lesser
Filter (filter) is to retain more original pixels information.
It should be appreciated that in this application, it is normal that SVM (Support Vector Machine), which refers to support vector machines,
A kind of method of discrimination seen.The learning model for having supervision in machine learning field, commonly used to carry out pattern-recognition,
Classification and regression analysis.
It should be appreciated that in this application, space base orbital image is the space base image that value shooting has orbital image.Space base image
Refer to the image with on-fixed background, the usually image of dollying head shooting.In the space base image being continuously shot, background
It is all variation with reference object.
It should be appreciated that in this application, HOG characteristic information refers to histograms of oriented gradients (Histogram of
Oriented Gradient, HOG) characteristic information is that one kind is used to carry out object detection in computer vision and image procossing
Feature Descriptor.HOG characteristic information is by calculating the gradient orientation histogram with statistical picture regional area come constitutive characteristic.
In a sub-picture, the presentation and shape of localized target can be described well by the direction Density Distribution at gradient or edge.Its
Essence are as follows: the statistical information of gradient, and gradient is primarily present in the place at edge.Implementation method is mainly: first by image point
At small connected region, these connected regions are called cell factory.Then the gradient of each pixel in cell factory is acquired
Or the direction histogram at edge.Finally altogether these set of histograms, so that it may which constitutive characteristic descriptor is the application's
HOG characteristic information.These local histograms (are called section) in the bigger range of image, and degree of comparing normalizes, can
To improve the performance of the algorithm, used method is: density of each histogram in this section is first calculated, then according to this
A density normalizes each cell factory in section.After being normalized by this, illumination variation and shade can be obtained
Better effect.
Depending on context, as used in this " if " can be construed to " ... when " or " when ... " or
" in response to determination " or " in response to detection ".
Technical solution of the present invention is described in detail with specifically embodiment below.These specific implementations below
Example can be combined with each other, and the same or similar concept or process may be repeated no more in some embodiments.
It is that a kind of space base pedestrian of rail traffic provided in an embodiment of the present invention invades the application of boundary's detection system referring to Fig. 1
Schematic diagram of a scenario.In application scenarios shown in Fig. 1, it includes that track is handed over that the space base pedestrian of rail traffic, which invades boundary's detection system mainly,
Logical space base pedestrian invades boundary's detection device and at least one space base image capture device.The space base pedestrian of rail traffic invades boundary's detection
Device can be understood as server 1 as shown in Figure 1, and space base image capture device can be understood as unmanned plane as shown in Figure 1
2.Unmanned plane 2 flies along the surface of track, and the underface of its body is arranged in the camera 21 of unmanned plane, it is possible thereby to vertically
Shooting space base orbital image downwards.Therefore, the picture centre of space base orbital image should be moved along the central axis of track, but
Since unmanned plane may be variation with respect to the flying height on ground, some space base orbital images only include that the anti-of track is invaded
Battery limit (BL) domain, some space base orbital images include the anti-scene image for invading battery limit (BL) domain and track two sides of track.Through the invention
The space base pedestrian for the rail traffic that embodiment provides invades boundary's detection system, and the space base image capture device is for acquiring space base rail
The space base pedestrian of road image, the rail traffic invades boundary's detection device, then for setting from space base Image Acquisition described at least one
It is standby to obtain the space base orbital image, and the space base pedestrian for executing the rail traffic as described in the examples of following each methods invades boundary's inspection
Survey method which thereby enhances detection accuracy and efficiency that pedestrian invades boundary.
It referring to fig. 2, is that a kind of space base pedestrian of rail traffic provided in an embodiment of the present invention invades the process of boundary's detection method
Schematic diagram, the executing subject of method shown in Fig. 2 can be software and/or hardware device.Method shown in Fig. 2 mainly includes step
S101 is specific as follows to step S103:
S101, according to the ratio of pixel distance and actual distance in space base orbital image, in the space base orbital image
It determines and prevents invading battery limit (BL) area image, the space base orbital image is the figure that space base image capture device takes vertically downward along track
Picture.
Specifically, prevent that invading battery limit (BL) domain can be understood as the area that the isolation facilities such as two side panel of track, wire netting fence up
Domain, it is understood that be the predeterminable area for radiating certain distance to two sides centered on track.Pixel in space base orbital image
The size of distance and the ratio of actual distance, characterizing space base orbital image is that space base image capture device is proximate to ground bat
It takes the photograph, or far from ground shooting.A same pedestrian, in the large percentage of pixel distance and actual distance, space base rail
Pedestrian in road image is large scale pedestrian target, such as portrait accounts for a semi-area of picture；And in pixel distance and really
When the ratio of distance is smaller, the pedestrian in space base orbital image is small size pedestrian target, such as portrait accounts for 1/5 face of picture
Product.Since in the case where shooting far from ground, space base orbital image is readily incorporated more non-orbital region image, such as
The objects such as rice field, the stone of orbital region two sides may all have an adverse effect to detection, reduce the accuracy of detection.And
In the case that the present embodiment ratio of pixel distance and actual distance in space base orbital image is different, using corresponding Fang Qin circle
Area image improves the anti-accuracy for invading battery limit (BL) area image.
It is to be understood that being obtained corresponding to the ratio by the ratio of pixel distance and actual distance in space base orbital image
Preset image procossing rule, prevents invading battery limit (BL) area image to determine in the space base orbital image.For example, each pixel away from
From a screenshot size corresponding with the ratio of actual distance, according to corresponding screenshot size in space base orbital image, with picture
The image interception of two sides symmetrical region is carried out on the basis of central axis, acquisition is corresponding to be prevented invading battery limit (BL) area image.By for example, in pixel
When distance and the ratio of actual distance are greater than preset threshold, directly space base orbital image is determined as preventing invading battery limit (BL) area image；And
When the ratio of pixel distance and actual distance is less than or equal to preset threshold, carry out preventing invading battery limit (BL) domain in space base orbital image
Image recognition, the region recognized is determined as anti-to invade battery limit (BL) area image.The present embodiment with pixel in space base orbital image away from
From being foundation with the ratio of actual distance, is determined from space base orbital image and prevent invading battery limit (BL) area image, various implementations are herein
Without limitation.
The anti-battery limit (BL) area image of invading is cut at least two subregion figure by S102, and from least two subregion
Obtaining in figure includes the portrait class administrative division map for invading boundary pedestrian.
Specifically, since the pedestrian for invading boundary's behavior may be in the anti-any position for invading battery limit (BL) domain, by that will prevent
It invades battery limit (BL) area image and is cut at least two subregion figure, and subregion is detected whether one by one for the mode of portrait class administrative division map,
Available includes the portrait class administrative division map for invading boundary pedestrian.Such as detect that in 2 adjacent sub-regions figures be all portrait class
Administrative division map then shows the pedestrian taken in this 2 adjacent sub-regions figure.
S103, according to position of the administrative division map of the portrait class in the space base orbital image and the space base trajectory diagram
As corresponding scene location, the position of boundary pedestrian is invaded described in determination.
Specifically, one space base orbital image of the every acquisition of space base image capture device, available acquisition position at that time,
And using the acquisition position as the corresponding scene location of space base orbital image.According to the corresponding scene location of space base orbital image
It can primarily determine the position range for invading boundary pedestrian, e.g. using longitude XX latitude YY as symmetrical centre, radius is 200 meters
Along track in road section scope.Then the position with the administrative division map of portrait class in the space base orbital image again, obtains invading boundary
Relative position of the pedestrian along the track in road section scope, and then can accurately obtain the position for invading boundary pedestrian.
The embodiment of the invention provides a kind of space base pedestrians of rail traffic to invade boundary's detection method, by according to space base track
The ratio of pixel distance and actual distance in image determines in the space base orbital image and prevents invading battery limit (BL) area image, the sky
Base orbital image is the image that space base image capture device takes vertically downward along track；Prevent that invading battery limit (BL) area image cuts for described
It is divided at least two subregion figure, and obtaining from least two subregion figure includes the portrait class region for invading boundary pedestrian
Figure；According to position and the space base orbital image corresponding field of the administrative division map of the portrait class in the space base orbital image
Scape position, determine described in invade the position of boundary pedestrian, so that the space base pedestrian for improving rail traffic invades boundary's accuracy of detection.
It is that the space base pedestrian of another rail traffic provided in an embodiment of the present invention invades the stream of boundary's detection method referring to Fig. 3
Journey schematic diagram, on the basis of the above embodiments, step S101 is (according to pixel distance in space base orbital image and actual distance
Ratio determines in the space base orbital image and prevents invading battery limit (BL) area image) it may alternatively be step S201 shown in Fig. 3 to step
Rapid S204, specific as follows:
S201 judges whether the ratio of pixel distance and actual distance is greater than proportion threshold value in space base orbital image.
If so, S202 is transferred to, if it is not, being then transferred to S203.
Specifically, the ratio of pixel distance and actual distance is greater than proportion threshold value in space base orbital image, shows space base rail
Road image is large-size images, should be shooting near the ground, and the scene of shooting is exactly to prevent invading battery limit (BL) area image.And space base orbital image
The ratio of middle pixel distance and actual distance is less than or equal to proportion threshold value, shows that space base orbital image is small-sized image, answers
For the shooting of remote ground, the scene of shooting in addition to it is anti-invade battery limit (BL) domain other than, also taken the scene of orbital region two sides, need into
Row image cropping extracts in space base orbital image and anti-invades battery limit (BL) area image.
The space base orbital image is determined as preventing invading battery limit (BL) area image by S202.
S203 prevents invading battery limit (BL) domain detection model, determines in the space base orbital image and prevent invading battery limit (BL) domain with preset
Selected frame.
Specifically, preset to prevent that invading battery limit (BL) domain detection model can be Faster-RCNN model.Faster-RCNN model
It can be used for the detection to objects in images, such as examined object is all that the anti-of track invades battery limit (BL) in the positive sample of training set
Domain, negative sample are the non-anti-surrounding enviroment image for invading battery limit (BL) domain, then training the Faster-RCNN model come can be to space base rail
Prevent that invading battery limit (BL) domain carries out frame choosing in road image.
(prevent invading battery limit (BL) domain detection model with preset, Fang Qin circle is determined in the space base orbital image in step S203
The selected frame in region) before, it can also include preventing invading battery limit (BL) domain detection model training process.Specifically, it can be and invaded with history
The true classification of battery limit (BL) area image carries out the training of Faster-RCNN algorithm to the anti-battery limit (BL) domain detection model of invading, so that described
The anti-prediction classification invaded battery limit (BL) domain detection model and invade the history output of battery limit (BL) area image, invades battery limit (BL) area image with the history
True classification between penalty values be lower than default Faster-RCNN and lose threshold value.
In a kind of specific real mode, before being detected to space base orbital image, first in training image library
Anti- battery limit (BL) domain training sample of invading invade battery limit (BL) domain detection model to anti-and be trained.It training image library can be from acquisition in advance
Rail traffic unmanned plane inspection video.Training picture is manually marked first, marking which picture is to prevent invading battery limit (BL) domain
Picture and which picture are other pictures, obtain preventing invading battery limit (BL) domain training sample.It optionally, can be with when selecting positive sample
It chooses only comprising the anti-maximum rectangle frame image invaded battery limit (BL) domain, do not include background area (farmland, road etc.).When picture middle orbit
Picture can be carried out truncation behaviour according to trade shape there are (track is bending) when angle by direction and picture X direction
Make, intercepts several straightway tracks respectively as independent positive sample, to reduce background area area in rectangle frame to the greatest extent.And it bears
The selection of sample can be the image for randomly selecting background area (farmland, road etc.).It is anti-to invade the use of battery limit (BL) domain detection model
Faster-RCNN model, training set sample mark sequentially input every trained picture in Faster-RCNN model after finishing
Carry out multiple model learning and training.Faster-RCNN model mainly includes that candidate regions extract network (RPN) and Fast-RCNN
Detect two convolutional neural networks of network.The method that Faster-RCNN model takes two networks alternating training in training, first
Then training RPN network fixes RPN network and Fast-RCNN with obtained candidate region training Fast-RCNN detection network
The shared depth convolutional layer for detecting network carries out tuning to the full articulamentum of RPN network, finally to the full articulamentum of detection network
Carry out tuning.Wherein, RPN network and Fast-RCNN detection network require pre-training initialization network, and in initialization network
Specific layer is added after output to obtain.The pre-training initialization network of the present embodiment can be ZFnet network.ZFnet network
Comprising 5 convolutional layers, 3 pond layers and 3 full articulamentums, the output of the 5th convolutional layer of ZFnet is known as characteristic pattern.It is utilizing
When preventing invading battery limit (BL) domain sample database training Faster-RCNN model, the ZFnet initiation parameter that can be directly obtained with pre-training,
Tuning is carried out to RPN network and Fast-RCNN detection network.
Faster-RCNN model after training in the detection process, carries out one to entire space base orbital image first
Serial convolution algorithm obtains the characteristic pattern of input picture, and this feature figure input RPN network is then generated candidate regions (region
Proposals), such as about 300 candidate regions are generated.Then by the output of characteristic pattern and RPN network as the defeated of pond layer
Enter, obtains regular length output.Classify finally by full articulamentum with softmax layers, obtains being used to indicate and prevent invading battery limit (BL)
The testing result of the selected frame in domain.
S204 is determined as the image in the selected frame described anti-to invade battery limit (BL) area image.
Specifically, selecting the image in frame is that Faster-RCNN model is identified as anti-invade in space base orbital image
The image range in battery limit (BL) domain, therefore the image interception in selected frame is come out, it is exactly anti-to invade battery limit (BL) area image.
On the basis of the above embodiments, in step S101 (according to pixel distance and actual distance in space base orbital image
Ratio, determined in the space base orbital image and anti-invade battery limit (BL) area image) before, can also include obtaining the pixel distance
With the process of the ratio of actual distance, specifically: obtain first the space base orbital image horizontal pixel distance and longitudinal picture
Plain distance.The number of pixels or the distance on image that horizontal pixel distance can be understood as space base orbital image transverse direction.
Similarly, the number of pixels or the distance on image that longitudinal pixel distance can be understood as space base orbital image longitudinal direction,
But (i.e. horizontal pixel distance and longitudinal pixel distance is all number of pixels, Huo Zhedou because consistent with the unit of horizontal pixel distance
It is apart from size).While obtaining horizontal pixel distance and longitudinal pixel distance, perhaps before or later, also acquisition institute
The corresponding photographing information of space base orbital image is stated, the photographing information includes shooting height information and shooting visual angle information；And root
According to the corresponding photographing information of the space base orbital image, the corresponding lateral actual distance of the space base orbital image and longitudinal direction are obtained
Actual distance.Wherein, shooting height information can be understood as the camera of space base image capture device from underface orbital region
The distance on ground can be obtained from the device parameter of space base image capture device.Shooting visual angle information can be understood as space base
Image capture device acquires the camera lens visual angle of image, such as horizontal view angle and vertical angle of view.After obtaining above-mentioned distance, by institute
The ratio for stating horizontal pixel distance and the lateral actual distance, is determined as the first ratio；And by longitudinal pixel distance with
The ratio of the longitudinal direction actual distance, is determined as the second ratio.It can be by the minimum ratio in first ratio and the second ratio
Example, is determined as the ratio of the pixel distance and actual distance.
In one implementation, the process for obtaining the ratio of the pixel distance and actual distance may is that acquisition is empty
Basic image acquires equipment along the visible light video of patrolling railway, and space base orbital image is extracted from video.Space base Image Acquisition
Equipment is, for example, unmanned plane, in the case where vertical lower view, obtains drone flying height H, the image size of shooting is w × h
Pixel, the visual angle of video camera are α × β degree (horizontal × vertical).Lateral actual distance can be obtained as a result, are as follows: Lhor=2Htan (α/
2), longitudinal actual distance are as follows: Lver=2Htan (β/2).Horizontal and vertical direction unit pixel represents practically in image
Reason distance is respectively the inverse of the first ratio: Phor=LhorThe inverse of/w and the second ratio: Pver=Lver/h.Assuming that PthIt is default
Proportion threshold value, such as P can be setth=0.015, if min (Phor,Pver) > Pth, indicate image in pedestrian target size compared with
It is small；Otherwise indicate that pedestrian target size is larger in image.
It referring to fig. 4, is that the space base pedestrian of another rail traffic provided in an embodiment of the present invention invades the stream of boundary's detection method
Journey schematic diagram, on the basis of the above embodiments, obtaining from least two subregion figure described in step S102 includes to invade
The process of the portrait class administrative division map of boundary pedestrian, may alternatively be step S301 shown in Fig. 4 to step S303, specific as follows:
S301 obtains the histograms of oriented gradients HOG characteristic information of each subregion figure.
Specifically, in the acquisition process of HOG characteristic information, the gray value at the pixel (x, y) of subregion figure can be set
For I, gradient magnitude G, gradient direction θ, using the One-Dimensional Center gradient operator of [- 1,0,1], horizontal direction and vertical is calculated
The gradient difference in direction is as follows:
Gx(x, y)=I (x+1, y)-I (x-1, y),
Gy(x, y)=I (x, y+1)-I (x, y-1).
Then the gradient intensity of the pixel (x, y) is calculated:
And the gradient direction of the pixel (x, y):
For RGB color image, then gradient is calculated to each Color Channel, chooses the maximum Color Channel institute of gradient magnitude
Gradient of the corresponding gradient as the pixel.Then subregion figure is divided into equally distributed unit (cell), such as 8
The unit of × 8 pixel sizes, wherein being combined into a zonule block (block) per adjacent Unit 2 × 2.In the block of zonule
Pixel is obtained using gradient intensity as weight.The positive and negative of gradient direction is not considered, i.e., direction is transformed into 0 °~180 °, histogram
Figure takes 9 steering columns.The histogram of each cell domain block is normalized, thus, it is possible to illumination, shade, edge contrast
Degree etc. has better invariance.Since each zonule block has the histogram of 49 dimensions, the zonule is obtained after normalization
The feature vector that block 36 is tieed up.Assuming that v is normalized feature vector, method for normalizing step is commonly used as step 1 to step 2:
Step 1, normalization operation is executed:Wherein ε is preset minimum constant, to avoid dividing
Mother is 0.
Step 2,0.2 is reduced to as a result, if there is element to be greater than 0.2 to step 1, executes a hyposynchronization again later
Rapid 1, until all elements are respectively less than or are equal to 0.2, terminate normalization.
After above-mentioned normalization operation, the feature vector of all zonule blocks is together in series, the subregion is formed
The HOG feature of figure.
S302 determines portrait class HOG characteristic information with preset portrait detection model in each HOG characteristic information,
Wherein, the portrait detection model is to be supported the training of vector machine SVM algorithm with history portrait class HOG characteristic information to obtain
's.
Specifically, portrait detection model can be SVM model, HOG feature input SVM model be detected, SVM model
Using Linear SVM model.The concrete mode that portrait class HOG characteristic information is determined in each HOG characteristic information, can be use
The mode of sliding window slides data acquisition window mouth, successively using each subregion figure as window figure in space base orbital image
Picture extracts HOG feature in window and inputs SVM model and classify, and invades boundary pedestrian inspection to realize in space base orbital image
It surveys and positions.
(with preset portrait detection model, determine that portrait class HOG is special in each HOG characteristic information in step S302
Reference breath) before, further include the portrait detection model training in following two implementation:
In one implementation, if the ratio of pixel distance and actual distance is greater than ratio threshold in space base orbital image
Value, expression space base orbital image are large-size images, therefore use and equally carried out with the SVM model that large-size images training obtains
Detection.Specifically, SVM calculation is carried out to the portrait detection model with the true classification of the first history portrait class HOG characteristic information
Method training, so that the prediction classification that the portrait detection model exports the first history portrait class HOG characteristic information, with institute
It states the penalty values between the true classification of the first history portrait class HOG characteristic information and is lower than default SVM loss threshold value, wherein institute
It is big to state the ratio of pixel distance and actual distance in the corresponding history space base orbital image of the first history portrait class HOG characteristic information
In proportion threshold value.
In another implementation, if the ratio of pixel distance and actual distance is less than ratio threshold in space base orbital image
Value, expression space base orbital image are small-sized image, therefore use and equally carried out with the SVM model that small-sized image training obtains
Detection.Specifically, SVM calculation is carried out to the portrait detection model with the true classification of the second history portrait class HOG characteristic information
Method training, so that the prediction classification that the portrait detection model exports the second history portrait class HOG characteristic information, with institute
It states the penalty values between the true classification of the second history portrait class HOG characteristic information and is lower than default SVM loss threshold value, wherein institute
It is small to state the ratio of pixel distance and actual distance in the corresponding history space base orbital image of the second history portrait class HOG characteristic information
In proportion threshold value.
Since the characteristic quantity that large scale and small size portrait are capable of providing has bigger difference, if with large scale portrait instruction
Practice the SVM model that collection obtains, the portrait of space base orbital image small-medium size is detected, omission factor is higher, therefore above-mentioned
In embodiment, according to the size grades of portrait in the ratio-dependent image of pixel distance in space base orbital image and actual distance
(such as large scale or small size) is then used the SVM model obtained with the image training of the size grades to be detected, improved
The precision of Liao Qin circle pedestrian detection.
Optionally, cross validation method selection SVM optimized parameter can be used in the training of above-mentioned SVM model, improve
Nicety of grading.
The corresponding subregion figure of the portrait class HOG characteristic information is determined as including to invade boundary pedestrian by S303
Portrait class administrative division map.
Specifically, SVM model only identifies portrait class HOG characteristic information in the present embodiment, without to 1 people still
2 people distinguish.Therefore, if the corresponding subregion figure of portrait class HOG characteristic information is the subregion figure of more people, portrait
The possible adjacent subregion figure of the corresponding subregion figure of class HOG characteristic information, it is also possible to non-adjacent subregion figure.Phase
Ying Di is using the position of the corresponding all subregion figures of portrait class HOG characteristic information as the position for invading boundary pedestrian.
It is that a kind of space base pedestrian of rail traffic provided in an embodiment of the present invention invades the structure of boundary's detection device referring to Fig. 5
Schematic diagram specifically includes that
Anti- battery limit (BL) area image of invading obtains module 51, for the ratio according to pixel distance in space base orbital image and actual distance
Example determines in the space base orbital image and prevents invading battery limit (BL) area image, the space base orbital image is space base image capture device
The image taken vertically downward along track；
Portrait class administrative division map obtains module 52, for preventing that invading battery limit (BL) area image is cut at least two subregion figure for described,
And obtaining from least two subregion figure includes the portrait class administrative division map for invading boundary pedestrian；
Position determination module 53, for position of the administrative division map according to the portrait class in the space base orbital image and
The corresponding scene location of the space base orbital image, determine described in invade the position of boundary pedestrian.
The space base pedestrian of the rail traffic of embodiment illustrated in fig. 5, which invades boundary's detection device, accordingly can be used for executing shown in Fig. 2
Step in embodiment of the method, it is similar that the realization principle and technical effect are similar, and details are not described herein again.
On the basis of the above embodiments, the anti-battery limit (BL) area image of invading obtains module 51, is specifically used for judging space base rail
Whether the ratio of pixel distance and actual distance is greater than proportion threshold value in road image；If so, by the space base orbital image, really
It is set to and prevents invading battery limit (BL) area image；If it is not, then preventing invading battery limit (BL) domain detection model with preset, determined in the space base orbital image
The anti-selected frame for invading battery limit (BL) domain, and by the image in the selected frame is determined as described anti-invading battery limit (BL) area image.
On the basis of the above embodiments, the anti-battery limit (BL) area image of invading obtains module 51, is also used to described according to sky
The ratio of pixel distance and actual distance in base orbital image, determined in the space base orbital image it is anti-invade battery limit (BL) area image it
Before, obtain the horizontal pixel distance and longitudinal direction pixel distance of the space base orbital image；It is corresponding to obtain the space base orbital image
Photographing information, the photographing information includes shooting height information and shooting visual angle information；According to the space base orbital image pair
The photographing information answered obtains the corresponding lateral actual distance of the space base orbital image and longitudinal actual distance；By the transverse direction
The ratio of pixel distance and the lateral actual distance, is determined as the first ratio；By longitudinal pixel distance and the longitudinal direction
The ratio of actual distance is determined as the second ratio；By the minimum scale in first ratio and the second ratio, it is determined as described
The ratio of pixel distance and actual distance.
On the basis of the above embodiments, the anti-battery limit (BL) area image of invading obtains module 51, is also used to described with default
It is anti-invade battery limit (BL) domain detection model, determined in the space base orbital image it is anti-invade the selected frame in battery limit (BL) domain before, invaded with history
The true classification of battery limit (BL) area image carries out the training of Faster-RCNN algorithm to the anti-battery limit (BL) domain detection model of invading, so that described
The anti-prediction classification invaded battery limit (BL) domain detection model and invade the history output of battery limit (BL) area image, invades battery limit (BL) area image with the history
True classification between penalty values be lower than default Faster-RCNN and lose threshold value.
On the basis of the above embodiments, the portrait class administrative division map obtains module 52, is specifically used for obtaining each son
The histograms of oriented gradients HOG characteristic information of administrative division map；With preset portrait detection model, in each HOG characteristic information
Determine portrait class HOG characteristic information, wherein the portrait detection model is supported with history portrait class HOG characteristic information
The training of vector machine SVM algorithm obtains；By the corresponding subregion figure of the portrait class HOG characteristic information, be determined as include
The portrait class administrative division map of You Qin circle pedestrian.
On the basis of the above embodiments, the portrait class administrative division map obtains module 52, is also used to described with preset
Portrait detection model, before determining portrait class HOG characteristic information in each HOG characteristic information, if in space base orbital image
The ratio of pixel distance and actual distance is greater than proportion threshold value, then with the true classification of the first history portrait class HOG characteristic information
SVM algorithm training is carried out to the portrait detection model, so that the portrait detection model is to the first history portrait class HOG
The prediction classification of characteristic information output, the penalty values between the true classification of the first history portrait class HOG characteristic information
Threshold value is lost lower than default SVM, wherein the corresponding history space base orbital image of the first history portrait class HOG characteristic information
The ratio of middle pixel distance and actual distance is greater than proportion threshold value；If the ratio of pixel distance and actual distance in space base orbital image
Example is less than proportion threshold value, then is carried out with the true classification of the second history portrait class HOG characteristic information to the portrait detection model
SVM algorithm training, so that the prediction class that the portrait detection model exports the second history portrait class HOG characteristic information
Not, the penalty values between the true classification of the second history portrait class HOG characteristic information are lower than default SVM loss threshold value,
Wherein, pixel distance and actual distance in the corresponding history space base orbital image of the second history portrait class HOG characteristic information
Ratio be less than proportion threshold value.
It is that a kind of space base pedestrian of rail traffic provided in an embodiment of the present invention invades the hardware of boundary's detection device referring to Fig. 6
Structural schematic diagram, the device include: processor 61, memory 62 and computer program；Wherein
Memory 62, for storing the computer program, which can also be flash memory (flash).The calculating
Machine program is, for example, to realize application program, the functional module etc. of the above method.
Processor 61, for executing the computer program of the memory storage, to realize, device is executed in the above method
Each step.It specifically may refer to the associated description in previous methods embodiment.
Optionally, memory 62 can also be integrated with processor 61 either independent.
When the memory 62 is independently of the device except processor 61, described device can also include:
Bus 63, for connecting the memory 62 and processor 61.
The present invention also provides a kind of readable storage medium storing program for executing, computer program is stored in the readable storage medium storing program for executing, it is described
The method that computer program is performed for realizing above-mentioned various embodiments offer.
Wherein, readable storage medium storing program for executing can be computer storage medium, be also possible to communication media.Communication media includes just
In from a place to any medium of another place transmission computer program.Computer storage medium can be general or special
Any usable medium enough accessed with computer capacity.For example, readable storage medium storing program for executing is coupled to processor, to enable a processor to
Information is read from the readable storage medium storing program for executing, and information can be written to the readable storage medium storing program for executing.Certainly, readable storage medium storing program for executing can also be with
It is the component part of processor.Processor and readable storage medium storing program for executing can be located at specific integrated circuit (Application
Specific Integrated Circuits, referred to as: ASIC) in.In addition, the ASIC can be located in user equipment.Certainly,
Processor and readable storage medium storing program for executing can also be used as discrete assembly and be present in communication equipment.
The present invention also provides a kind of program product, the program product include execute instruction, this execute instruction be stored in it is readable
In storage medium.At least one processor of equipment can read this from readable storage medium storing program for executing and execute instruction, at least one processing
Device executes this and executes instruction so that equipment implements the method that above-mentioned various embodiments provide.
It is invaded in the embodiment of boundary's detection device in the space base pedestrian of above-mentioned rail traffic, it should be appreciated that during processor can be
Central Processing Unit (English: Central Processing Unit, referred to as: CPU), it can also be other general processors, number
Signal processor (English: Digital Signal Processor, referred to as: DSP), specific integrated circuit (English:
Application Specific Integrated Circuit, referred to as: ASIC) etc..General processor can be microprocessor
Or the processor is also possible to any conventional processor etc..It can direct body in conjunction with the step of method disclosed in the present application
Now executes completion for hardware processor, or in processor hardware and software module combine and execute completion.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (10)
1. a kind of space base pedestrian of rail traffic invades boundary's detection method characterized by comprising
According to the ratio of pixel distance and actual distance in space base orbital image, Fang Qin circle is determined in the space base orbital image
Area image, the space base orbital image are the image that space base image capture device takes vertically downward along track；
Prevent that invading battery limit (BL) area image is cut at least two subregion figure for described, and obtains packet from least two subregion figure
Contain the portrait class administrative division map for invading boundary pedestrian；
It is corresponding according to position of the administrative division map of the portrait class in the space base orbital image and the space base orbital image
Scene location, determine described in invade the position of boundary pedestrian.
2. the method according to claim 1, wherein described according to pixel distance in space base orbital image and true
The ratio of distance determines in the space base orbital image and prevents invading battery limit (BL) area image, comprising:
Judge whether the ratio of pixel distance and actual distance is greater than proportion threshold value in space base orbital image；
If so, the space base orbital image is determined as preventing invading battery limit (BL) area image；
If it is not, then preventing invading battery limit (BL) domain detection model with preset, the anti-choosing for invading battery limit (BL) domain is determined in the space base orbital image
Determine frame, and by the image in the selected frame, is determined as described anti-invading battery limit (BL) area image.
3. method according to claim 1 or 2, which is characterized in that described according to pixel distance in space base orbital image
With the ratio of actual distance, determined in the space base orbital image before preventing invading battery limit (BL) area image, further includes:
Obtain the horizontal pixel distance and longitudinal direction pixel distance of the space base orbital image；
The corresponding photographing information of the space base orbital image is obtained, the photographing information includes shooting height information and shooting visual angle
Information；
According to the corresponding photographing information of the space base orbital image, the corresponding lateral actual distance of the space base orbital image is obtained
With longitudinal actual distance；
By the ratio of the horizontal pixel distance and the lateral actual distance, it is determined as the first ratio；
By the ratio of longitudinal pixel distance and longitudinal actual distance, it is determined as the second ratio；
By the minimum scale in first ratio and the second ratio, it is determined as the ratio of the pixel distance and actual distance.
4. according to the method described in claim 2, it is characterized in that, it is described with it is preset it is anti-invade battery limit (BL) domain detection model,
Anti- invade before the selected frame in battery limit (BL) domain is determined in the space base orbital image, further includes:
Faster-RCNN algorithm instruction is carried out to the anti-battery limit (BL) domain detection model of invading with the true classification that history invades battery limit (BL) area image
Practice, so that the anti-prediction classification invaded battery limit (BL) domain detection model and invade the history output of battery limit (BL) area image, with the history
The penalty values invaded between the true classification of battery limit (BL) area image are lower than default Faster-RCNN and lose threshold value.
5. the method according to claim 1, wherein described obtain from least two subregion figure includes
The portrait class administrative division map of You Qin circle pedestrian, comprising:
Obtain the histograms of oriented gradients HOG characteristic information of each subregion figure；
With preset portrait detection model, portrait class HOG characteristic information is determined in each HOG characteristic information, wherein described
Portrait detection model is to be supported the training of vector machine SVM algorithm with history portrait class HOG characteristic information to obtain；
By the corresponding subregion figure of the portrait class HOG characteristic information, it is determined as including the portrait class area for invading boundary pedestrian
Domain figure.
6. according to the method described in claim 5, it is characterized in that, described with preset portrait detection model, each described
In HOG characteristic information before determining portrait class HOG characteristic information, further includes:
If the ratio of pixel distance and actual distance is greater than proportion threshold value in space base orbital image, with the first history portrait class
The true classification of HOG characteristic information carries out SVM algorithm training to the portrait detection model, so that the portrait detection model pair
The prediction classification of the first history portrait class HOG characteristic information output, with the first history portrait class HOG characteristic information
Penalty values between true classification are lower than default SVM and lose threshold value, wherein the first history portrait class HOG characteristic information pair
The ratio of pixel distance and actual distance is greater than proportion threshold value in the history space base orbital image answered；
If the ratio of pixel distance and actual distance is less than proportion threshold value in space base orbital image, with the second history portrait class
The true classification of HOG characteristic information carries out SVM algorithm training to the portrait detection model, so that the portrait detection model pair
The prediction classification of the second history portrait class HOG characteristic information output, with the second history portrait class HOG characteristic information
Penalty values between true classification are lower than default SVM and lose threshold value, wherein the second history portrait class HOG characteristic information pair
The ratio of pixel distance and actual distance is less than proportion threshold value in the history space base orbital image answered.
7. a kind of space base pedestrian of rail traffic invades boundary's detection device characterized by comprising
Anti- battery limit (BL) area image of invading obtains module, for the ratio according to pixel distance and actual distance in space base orbital image,
It is determined in the space base orbital image and prevents invading battery limit (BL) area image, the space base orbital image is space base image capture device along track
The image taken vertically downward；
Portrait class administrative division map obtains module, for the anti-battery limit (BL) area image of invading to be cut at least two subregion figure, and from institute
Stating acquisition at least two subregion figure includes the portrait class administrative division map for invading boundary pedestrian；
Position determination module, for position and the sky of the administrative division map according to the portrait class in the space base orbital image
The corresponding scene location of base orbital image, determine described in invade the position of boundary pedestrian.
8. device according to claim 7, which is characterized in that
The anti-battery limit (BL) area image of invading obtains module, specifically for judging pixel distance and actual distance in space base orbital image
Whether ratio is greater than proportion threshold value；If so, the space base orbital image is determined as preventing invading battery limit (BL) area image；If it is not, then using
It is preset to prevent invading battery limit (BL) domain detection model, the anti-selected frame for invading battery limit (BL) domain is determined in the space base orbital image, and will be described
Image in selected frame is determined as described anti-invading battery limit (BL) area image.
9. device according to claim 7 or 8, which is characterized in that
The anti-battery limit (BL) area image of invading obtains module, be also used to it is described according to pixel distance in space base orbital image with really away from
From ratio, determined in the space base orbital image it is anti-invade battery limit (BL) area image before, obtain the cross of the space base orbital image
To pixel distance and longitudinal pixel distance；The corresponding photographing information of the space base orbital image is obtained, the photographing information includes
Shooting height information and shooting visual angle information；According to the corresponding photographing information of the space base orbital image, the space base rail is obtained
The corresponding lateral actual distance of road image and longitudinal actual distance；By the horizontal pixel distance and the lateral actual distance
Ratio is determined as the first ratio；By the ratio of longitudinal pixel distance and longitudinal actual distance, it is determined as the second ratio
Example；By the minimum scale in first ratio and the second ratio, it is determined as the ratio of the pixel distance and actual distance.
10. a kind of space base pedestrian of rail traffic invades boundary's detection system, which is characterized in that the space base pedestrian including rail traffic invades
Boundary's detection device and at least one space base image capture device；
The space base image capture device, for acquiring space base orbital image；
The space base pedestrian of the rail traffic invades boundary's detection device, for obtaining from space base image capture device described at least one
The space base orbital image, and perform claim requires the space base pedestrian of 1~6 any rail traffic to invade boundary's detection method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810710554.9A CN109145715B (en) | 2018-07-02 | 2018-07-02 | Air-based pedestrian boundary-crossing detection method, device and system for rail transit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810710554.9A CN109145715B (en) | 2018-07-02 | 2018-07-02 | Air-based pedestrian boundary-crossing detection method, device and system for rail transit |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109145715A true CN109145715A (en) | 2019-01-04 |
| CN109145715B CN109145715B (en) | 2020-10-30 |
Family
ID=64799662
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810710554.9A Active CN109145715B (en) | 2018-07-02 | 2018-07-02 | Air-based pedestrian boundary-crossing detection method, device and system for rail transit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109145715B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101101204A (en) * | 2006-07-05 | 2008-01-09 | 三星电子株式会社 | System and method for detecting moving object using structured light, and mobile robot including system thereof |
| US20130148850A1 (en) * | 2011-12-13 | 2013-06-13 | Fujitsu Limited | User detecting apparatus, user detecting mehtod, and computer-readable recording medium storing a user detecting program |
| CN105759834A (en) * | 2016-03-09 | 2016-07-13 | 中国科学院上海微系统与信息技术研究所 | System and method of actively capturing low altitude small unmanned aerial vehicle |
| CN106741008A (en) * | 2016-12-29 | 2017-05-31 | 北京交通大学 | Rail track method for recognizing impurities and system |
| CN107169401A (en) * | 2017-02-10 | 2017-09-15 | 北京交通大学 | Object detecting method is invaded based on the track that track visual signature is composed |
| CN107204965A (en) * | 2016-03-18 | 2017-09-26 | 阿里巴巴集团控股有限公司 | The hold-up interception method and system of a kind of password cracking behavior |
| CN108009506A (en) * | 2017-12-07 | 2018-05-08 | 平安科技（深圳）有限公司 | Intrusion detection method, application server and computer-readable recording medium |
-
2018
- 2018-07-02 CN CN201810710554.9A patent/CN109145715B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101101204A (en) * | 2006-07-05 | 2008-01-09 | 三星电子株式会社 | System and method for detecting moving object using structured light, and mobile robot including system thereof |
| US20130148850A1 (en) * | 2011-12-13 | 2013-06-13 | Fujitsu Limited | User detecting apparatus, user detecting mehtod, and computer-readable recording medium storing a user detecting program |
| CN105759834A (en) * | 2016-03-09 | 2016-07-13 | 中国科学院上海微系统与信息技术研究所 | System and method of actively capturing low altitude small unmanned aerial vehicle |
| CN107204965A (en) * | 2016-03-18 | 2017-09-26 | 阿里巴巴集团控股有限公司 | The hold-up interception method and system of a kind of password cracking behavior |
| CN106741008A (en) * | 2016-12-29 | 2017-05-31 | 北京交通大学 | Rail track method for recognizing impurities and system |
| CN107169401A (en) * | 2017-02-10 | 2017-09-15 | 北京交通大学 | Object detecting method is invaded based on the track that track visual signature is composed |
| CN108009506A (en) * | 2017-12-07 | 2018-05-08 | 平安科技（深圳）有限公司 | Intrusion detection method, application server and computer-readable recording medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109145715B (en) | 2020-10-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| RU2484531C2 (en) | Apparatus for processing video information of security alarm system | |
| US9846946B2 (en) | Objection recognition in a 3D scene | |
| CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
| US8509526B2 (en) | Detection of objects in digital images | |
| CN109145708B (en) | Pedestrian flow statistical method based on RGB and D information fusion | |
| US20150054824A1 (en) | Object detection method, object detection device, and image pickup device | |
| CN105260749B (en) | Real-time target detection method based on direction gradient binary pattern and soft cascade SVM | |
| CN108197604A (en) | Fast face positioning and tracing method based on embedded device | |
| CN109033950A (en) | Vehicle based on multiple features fusion cascade deep model, which is disobeyed, stops detection method | |
| CN107909033A (en) | Suspect's fast track method based on monitor video | |
| CN104301630B (en) | A kind of video image joining method and device | |
| CN106446150A (en) | Method and device for precise vehicle retrieval | |
| CN111414807A (en) | Tidal water identification and crisis early warning method based on YO L O technology | |
| CN113111727A (en) | Method for detecting rotating target in remote sensing scene based on feature alignment | |
| CN114926747A (en) | Remote sensing image directional target detection method based on multi-feature aggregation and interaction | |
| CN109558790B (en) | Pedestrian target detection method, device and system | |
| CN114140745A (en) | Method, system, device and medium for detecting personnel attributes of construction site | |
| CN111652035A (en) | Pedestrian re-identification method and system based on ST-SSCA-Net | |
| CN109214254A (en) | A kind of method and device of determining robot displacement | |
| Neuhausen et al. | Automatic window detection in facade images | |
| Xiao et al. | Geo-spatial aerial video processing for scene understanding and object tracking | |
| Wang et al. | Real-time damaged building region detection based on improved YOLOv5s and embedded system from UAV images | |
| Xing et al. | Compact one-stage object detection network | |
| CN109145715A (en) | The space base pedestrian of rail traffic invades boundary's detection method, device 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 |