CN109784265A - A kind of rail level semantic segmentation method and device - Google Patents
A kind of rail level semantic segmentation method and device Download PDFInfo
- Publication number
- CN109784265A CN109784265A CN201910020741.9A CN201910020741A CN109784265A CN 109784265 A CN109784265 A CN 109784265A CN 201910020741 A CN201910020741 A CN 201910020741A CN 109784265 A CN109784265 A CN 109784265A
- Authority
- CN
- China
- Prior art keywords
- image
- rail level
- pixel
- orbital
- class probability
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
This application provides a kind of rail level semantic segmentation method and devices, wherein this method comprises: obtaining orbital image；The orbital image is input in preparatory trained rail level semantic segmentation model, obtains the corresponding class probability of each pixel in the orbital image, the class probability is for characterizing the probability that each pixel is rail level feature；According to the corresponding class probability of pixel each in the orbital image and default class probability threshold value, the rail level segmented image of the orbital image is obtained, the rail level segmented image is marked with rail level region and background area in the orbital image.The embodiment of the present application can be partitioned into rail level part by image Segmentation Technology from the image in front of train driving, promote the vision difference of rail level and other regions, whether there is barrier so as to help driver preferably to distinguish on track, improves the accuracy rate of discrimination.
Description
Technical field
This application involves technical field of rail traffic, in particular to a kind of rail level semantic segmentation method and device.
Background technique
Currently, often there are barrier, such as pedestrian, vehicle or other articles in train driving region in rail traffic,
Easily cause traffic accident.Train in the process of moving, mainly by the visual running region of driver, artificially judges on track at present
Whether barrier is had.But since the color of the color of track and other backgrounds or barrier is not much different, and train driving speed
Degree is big, and human eye sighting distance is shorter, therefore only goes to have discerned whether that barrier is easy to appear the case where failing to judge by human eye, causes to distinguish
Other accuracy rate is lower.
Summary of the invention
In view of this, the embodiment of the present application is designed to provide a kind of rail level semantic segmentation method and device, Neng Goutong
It crosses image Segmentation Technology and is partitioned into rail level part from the image in front of train driving, promote the visual difference of rail level and other regions
Not, whether there is barrier so as to help driver preferably to distinguish on track, improve the accuracy rate of discrimination.
In a first aspect, the embodiment of the present application provides a kind of rail level semantic segmentation method, comprising:
Obtain orbital image；
The orbital image is input in preparatory trained rail level semantic segmentation model, is obtained in the orbital image
The corresponding class probability of each pixel, the class probability is for characterizing the probability that each pixel is rail level feature；
According to the corresponding class probability of pixel each in the orbital image and default class probability threshold value, obtain described
The rail level segmented image of orbital image, the rail level segmented image are marked with rail level region and background in the orbital image
Region.
With reference to first aspect, the embodiment of the present application provides the first possible embodiment of first aspect, wherein logical
It crosses following manner and obtains the trained rail level semantic segmentation model in advance:
Multiple path images are obtained, and sample corresponding with the path image marks image, the sample
This mark image tagged has rail level region and background area in the path image；
The path image is input in the rail level semantic segmentation model constructed in advance, the path is obtained
The corresponding class probability of each pixel in image；
According to the corresponding class probability of pixel each in the path image and with each path image
Corresponding sample marks image, carries out epicycle training to the rail level semantic segmentation model；
By carrying out more wheel training to the rail level semantic segmentation model, the trained rail level semanteme point in advance is obtained
Cut model.
The possible embodiment of with reference to first aspect the first, the embodiment of the present application provide second of first aspect
Possible embodiment, wherein it is described according to the corresponding class probability of pixel each in the path image and with
The corresponding sample of each path image marks image, carries out epicycle training to the rail level semantic segmentation model, comprising:
For each path image, according to the corresponding class probability of pixel each in the path image
And corresponding with path image sample marks image, obtains the corresponding friendship of each pixel in the path image
Pitch entropy loss；
According to the corresponding intersection entropy loss of pixel each in the path image, the rail level semantic segmentation mould is adjusted
The parameter of type；
Until determining and completing to the rail level semantic segmentation model after the training of all path images completion epicycle
Epicycle training.
With reference to first aspect, the embodiment of the present application provides the third possible embodiment of first aspect, wherein institute
It states according to the corresponding class probability of pixel each in the orbital image and default class probability threshold value, obtains the trajectory diagram
The rail level segmented image of picture, comprising:
For each pixel in the orbital image, if the corresponding class probability of the pixel is greater than described preset
Class probability threshold value, it is determined that the pixel belongs to rail level region, and the pixel value of the pixel is arranged to default rail level picture
Element value；
In having adjusted the orbital image after the pixel value of each pixel, the first rail of the orbital image is obtained
Face segmented image.
The third possible embodiment with reference to first aspect, the embodiment of the present application provide the 4th kind of first aspect
Possible embodiment, wherein described according to the corresponding class probability of pixel each in the orbital image and default classification
Probability threshold value obtains the rail level segmented image of the orbital image, comprising:
For each pixel in the orbital image, if the corresponding class probability of the pixel is no more than described pre-
If class probability threshold value, it is determined that the pixel belongs to background area, and the pixel value of the pixel is arranged to preset non-rail
Face pixel value；
In having adjusted the orbital image after the pixel value of each pixel, the second rail of the orbital image is obtained
Face segmented image.
With reference to first aspect, the embodiment of the present application provides the 5th kind of possible embodiment of first aspect, wherein institute
State method further include: will be on the rail level area maps to the orbital image in the rail level segmented image.
Second aspect, the embodiment of the present application provide a kind of rail level semantic segmentation device, comprising:
Orbital image obtains module, for obtaining orbital image；
Pixel categorization module, for the orbital image to be input to preparatory trained rail level semantic segmentation model
In, the corresponding class probability of each pixel in the orbital image is obtained, the class probability is for characterizing each pixel
It is the probability of rail level feature；
Rail level segmented image obtains module, for according to the corresponding class probability of pixel each in the orbital image and
Default class probability threshold value, obtains the rail level segmented image of the orbital image, the rail level segmented image is marked with the rail
Rail level region and background area in road image.
In conjunction with second aspect, the embodiment of the present application provides the first possible embodiment of second aspect, wherein also
Include:
Rail level semantic segmentation model training module, for obtaining multiple path images, and with the path
The corresponding sample of image marks image, the sample mark image tagged have rail level region in the path image and
Background area；
The path image is input in the rail level semantic segmentation model constructed in advance, the path is obtained
The corresponding class probability of each pixel in image；
According to the corresponding class probability of pixel each in the path image and with each path image
Corresponding sample marks image, carries out epicycle training to the rail level semantic segmentation model；
By carrying out more wheel training to the rail level semantic segmentation model, the trained rail level semanteme point in advance is obtained
Cut model.
In conjunction with the first possible embodiment of second aspect, the embodiment of the present application provides second of second aspect
Possible embodiment, wherein the rail level semantic segmentation model training module is specifically used for using following manner according to
Each corresponding class probability of pixel and sample corresponding with each path image mark figure in path image
Picture carries out epicycle training to the rail level semantic segmentation model:
For each path image, according to the corresponding class probability of pixel each in the path image
And corresponding with path image sample marks image, obtains the corresponding friendship of each pixel in the path image
Pitch entropy loss；
According to the corresponding intersection entropy loss of pixel each in the path image, the rail level semantic segmentation mould is adjusted
The parameter of type；
Until determining and completing to the rail level semantic segmentation model after the training of all path images completion epicycle
Epicycle training.
In conjunction with second aspect, the embodiment of the present application provides the third possible embodiment of second aspect, wherein institute
It states rail level segmented image and obtains module, specifically for each pixel being directed in the orbital image, if the pixel pair
The class probability answered is greater than the default class probability threshold value, it is determined that the pixel belongs to rail level region, and by the pixel
Pixel value be arranged to default rail level pixel value；
In having adjusted the orbital image after the pixel value of each pixel, the first rail of the orbital image is obtained
Face segmented image.
In conjunction with the third possible embodiment of second aspect, the embodiment of the present application provides the 4th kind of second aspect
Possible embodiment, wherein the rail level segmented image obtains module, also particularly useful for for every in the orbital image
A pixel, if the corresponding class probability of the pixel is not more than the default class probability threshold value, it is determined that the pixel
Belong to background area, and the pixel value of the pixel is arranged to preset non-rail level pixel value；
In having adjusted the orbital image after the pixel value of each pixel, the second rail of the orbital image is obtained
Face segmented image.
In conjunction with second aspect, the embodiment of the present application provides the 5th kind of possible embodiment of second aspect, wherein also
Include:
Mapping block, for by the rail level area maps in the rail level segmented image to the orbital image.
The third aspect, the embodiment of the present application also provide a kind of computer equipment, comprising: processor, memory and bus, institute
State memory and be stored with the executable machine readable instructions of the processor, when computer equipment operation, the processor with
By bus communication between the memory, the machine readable instructions execute above-mentioned first party when being executed by the processor
The possible embodiment of the first of face or first aspect any possible embodiment party into the 5th kind of possible embodiment
Step in formula.
Fourth aspect, the embodiment of the present application also provide a kind of computer readable storage medium, the computer-readable storage medium
Computer program is stored in matter, which executes above-mentioned in a first aspect, or first aspect when being run by processor
Step of the first possible embodiment into the 5th kind of possible embodiment in any possible embodiment.
Rail level semantic segmentation method and device provided by the embodiments of the present application will acquire when carrying out rail level segmentation
Orbital image be input in preparatory trained rail level semantic segmentation model, obtain each pixel pair in the orbital image
The class probability answered, the class probability is for characterizing the probability that each pixel is rail level feature；According to the orbital image
In each corresponding class probability of pixel and default class probability threshold value, obtain the rail level segmented image of the orbital image,
The rail level segmented image is marked with rail level region and background area in the orbital image.It can be seen that the application is real
It applies in example and rail level part is partitioned into from the image in front of train driving by image Segmentation Technology, promote rail level and other regions
Vision difference, whether have barrier so as to help driver preferably to distinguish on track, improve the accuracy rate of discrimination.
To enable the above objects, features, and advantages of the application to be clearer and more comprehensible, preferred embodiment is cited below particularly, and cooperate
Appended attached drawing, is described in detail below.
Detailed description of the invention
Technical solution in ord to more clearly illustrate embodiments of the present application, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only some embodiments of the application, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 shows a kind of flow chart of rail level semantic segmentation method provided by the embodiment of the present application；
Fig. 2 shows in rail level semantic segmentation method provided by the embodiment of the present application, rail level semantic segmentation model training
The flow chart of method；
Fig. 3 is shown in rail level semantic segmentation method provided by the embodiment of the present application, rail level semantic segmentation model epicycle
Trained flow chart；
Fig. 4 shows a kind of structural schematic diagram of rail level semantic segmentation device provided by the embodiment of the present application；
Fig. 5 shows a kind of structural schematic diagram of computer equipment provided by the embodiment of the present application.
Specific embodiment
To keep the purposes, technical schemes and advantages of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application
Middle attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment is only
It is some embodiments of the present application, instead of all the embodiments.The application being usually described and illustrated herein in the accompanying drawings is real
The component for applying example can be arranged and be designed with a variety of different configurations.Therefore, below to the application's provided in the accompanying drawings
The detailed description of embodiment is not intended to limit claimed scope of the present application, but is merely representative of the selected reality of the application
Apply example.Based on embodiments herein, those skilled in the art institute obtained without making creative work
There are other embodiments, shall fall in the protection scope of this application.
Currently, often there are barrier, such as pedestrian, vehicle or other articles in train driving region in rail traffic,
Easily cause traffic accident.Train in the process of moving, mainly by the visual running region of driver, artificially judges on track at present
Whether barrier is had.But since the color of the color of track and other backgrounds or barrier is not much different, and train driving speed
Degree is big, and human eye sighting distance is shorter, therefore only goes to have discerned whether that barrier is easy to appear the case where failing to judge by human eye, causes to distinguish
Other accuracy rate is lower.Based on this, a kind of rail level semantic segmentation method and device provided by the present application can pass through image segmentation
Technology is partitioned into rail level part from the image in front of train driving, promotes the vision difference of rail level and other regions, so as to
Whether there is barrier to help driver preferably to distinguish on track, improves the accuracy rate of discrimination.
For convenient for understanding the present embodiment, first to a kind of rail level semantic segmentation side disclosed in the embodiment of the present application
Method describes in detail.Wherein, rail level refers to two tracks and in-between region.
Shown in Figure 1, rail level semantic segmentation method includes S101~S104 provided by the embodiment of the present application:
S101: orbital image is obtained.
When specific implementation, image collecting device is installed in train head position, such as can be monocular camera, leads to
Cross the orbital image that image collecting device obtains train frontal scene in real time.
S102: orbital image is input in preparatory trained rail level semantic segmentation model, is obtained every in orbital image
The corresponding class probability of a pixel.
Wherein, class probability is for characterizing the probability that each pixel is rail level feature.
In a kind of possible situation, the possible size of the orbital image of acquisition is larger, if directly by larger-size rail
Road image is input in rail level semantic segmentation model, and it is lower to will lead to calculating speed.It therefore can be by larger-size trajectory diagram
As being first converted into suitable small-sized image, precision is not only guaranteed but also has guaranteed computational efficiency.
Optionally, rail level semantic segmentation model can be the semantic segmentation model based on autocoder, be also possible to base
In the semantic segmentation model of full convolutional neural networks (Fully Convolutional networks, FCN).
Specifically, shown in Figure 2, the embodiment of the present application obtains trained rail level semanteme point in advance by following manner
Cut model:
S201: obtaining multiple path images, and sample corresponding with path image marks image.
Path image refers to the trajectory diagram obtained under several scenes by the image collecting device of train Chinese herbaceous peony
Picture.Sample mark image tagged has rail level region and background area in path image, such as sample mark image can
Think that bianry image, the pixel value in rail level region are 1, the pixel value of background area is 0.
S202: path image being input in the rail level semantic segmentation model constructed in advance, obtains path figure
The corresponding class probability of each pixel as in.
The rail level semantic segmentation model constructed in advance includes that characteristic pattern extracts network and classifier, and characteristic pattern extracts network
For by a series of coding and decodings process and intermediate features figure fusion process is extracted from path image and
The size of the corresponding characteristic pattern of path image, characteristic pattern is consistent with the size of path image, each of characteristic pattern
The pixel that the point of position is used to characterize corresponding position in path image belongs to the fractional value in rail level region.Classifier is used for
The fractional value of each pixel in characteristic pattern is normalized, the pixel category of corresponding position in path image is obtained
Probability in rail level region.
S203: according to the corresponding class probability of pixel each in path image and with each path image
Corresponding sample marks image, carries out epicycle training to rail level semantic segmentation model.
Shown in Figure 3 when specific implementation, the embodiment of the present application is by following step to rail level semantic segmentation model
Carry out epicycle training:
S301: being directed to each path image, general according to the corresponding classification of pixel each in the path image
Rate and sample corresponding with the path image mark image, and it is corresponding to obtain each pixel in the path image
Intersect entropy loss.
It during model training, is trained using stochastic gradient descent method, each pixel in path image
The corresponding intersection entropy loss of point, as shown in formula (1):
Formula (1):
Wherein, y be the corresponding label of pixel, the value of y be 1 or 0,1 indicate the pixel be rail level region, 0 table
Show that the pixel is background area；For the corresponding class probability of pixel.
S302: according to the corresponding intersection entropy loss of pixel each in the path image, rail level semantic segmentation is adjusted
The parameter of model.
S303: it until after the training of all path images completion epicycle, determines and completes to rail level semantic segmentation model
Epicycle training.
Optionally, after completing the epicycle training to rail level semantic segmentation model, the embodiment of the present application can pass through following two
Any one method in kind mode detects whether deconditioning:
Mode one:
Whether detection epicycle reaches default wheel number；
If it is, stop the training to rail level semantic segmentation model, the rail level semanteme that last training in rotation is got point
Model is cut as preparatory trained rail level semantic segmentation model.
Mode two:
Whether the intersection entropy loss of detection epicycle is less than default loss threshold value；
If it is, stop the training to rail level semantic segmentation model, the rail level semanteme that last training in rotation is got point
Model is cut as preparatory trained rail level semantic segmentation model.
S204: by carrying out more wheel training to rail level semantic segmentation model, preparatory trained rail level semantic segmentation is obtained
Model.
Orbital image is input to preparatory trained rail level semantic segmentation first when specific implementation by step S102
Characteristic pattern in model extracts in network, obtains the characteristic pattern of orbital image, and the point of each position in characteristic pattern is for characterizing
The pixel of corresponding position belongs to the fractional value in rail level region in orbital image.Then characteristic pattern is input in classifier, is obtained
The corresponding class probability of each pixel into orbital image, class probability are the general of rail level feature for characterizing each pixel
Rate.
S103: according to the corresponding class probability of pixel each in orbital image and default class probability threshold value, rail is obtained
The rail level segmented image of road image.
Wherein, rail level segmented image is marked with rail level region and background area in orbital image.
When specific implementation, a class probability threshold value can be preset, class probability threshold value can be according to specific
Situation is configured, such as can be set to 0.9, is also possible to 0.8.The embodiment of the present application can obtain rail by following manner
The rail level segmented image of road image:
For each pixel in orbital image, if the corresponding class probability of the pixel is greater than default class probability
Threshold value, it is determined that the pixel belongs to rail level region, and the pixel value of the pixel is arranged to default rail level pixel value；It is adjusting
In whole complete orbital image after the pixel value of each pixel, the first rail level segmented image of orbital image is obtained.
Illustratively, rail level pixel value is preset, can be the gray value 1 of white, be also possible to color gray.By upper
The process of stating can be obtained by the first rail level segmented image of orbital image.
As can be seen that the first rail level segmented image that the embodiment of the present application obtains only has adjusted and belongs to rail in orbital image
The pixel value of the pixel in face region can retain the original background region of orbital image in this way.
It optionally,, can also be to trajectory diagram after obtaining the first rail level segmented image in another embodiment of the application
Picture is further adjusted, and the second rail level segmented image is obtained:
For each pixel in orbital image, if the corresponding class probability of the pixel is general no more than default classification
Rate threshold value, it is determined that the pixel belongs to background area, and the pixel value of the pixel is arranged to preset non-rail level pixel value；
In having adjusted orbital image after the pixel value of each pixel, the second rail level segmented image of orbital image is obtained.
Illustratively, non-rail level pixel value is preset, can be the gray value 0 of black.It can be obtained by by the above process
Second rail level segmented image of orbital image.
As can be seen that the second rail level segmented image that the embodiment of the present application obtains is on the basis of the first rail level segmented image
On, and the pixel value that the pixel in rail level region is not belonging in orbital image is had adjusted, it in this way can be by rail level region and background
Region is further discriminated between, bigger on vision difference.
In a further embodiment, pixel value that can also only to the pixel for being not belonging to rail level region in rail level image
It is adjusted, i.e., only adjusts the pixel value of background area.But the rail level segmented image that this partitioning scheme obtains does not have first two
Rail level segmented image good visual effect obtained in embodiment.
Optionally, the embodiment of the present application is after obtaining the rail level segmented image of orbital image by the above process, in order to
Rail level segmented image is preferably presented on train operator at the moment, further includes step S104:
S104: will be on the rail level area maps to orbital image in rail level segmented image.
It, can not be directly by the first rail level in order to which image segmentation result is preferably presented when specific implementation
Segmented image or the second rail level segmented image are presented on the display screen before driver eye, can be by the first rail level segmented image or
On the orbital image in front of rail level area maps to the train obtained in real time in two rail level segmented images.What mapping herein referred to
It is on the orbital image that only the rail level region in rail level segmented image is individually added in front of the train obtained in real time.Such as rail
The color in face region can be the bright single colour of translucent, such as green, and be added to the column obtained in real time in this way
On the orbital image of front side, original orbital image can not only be presented, but also the orbital region that can give prominence to the key points, so that effect is presented
Fruit is relatively good.
In a kind of possible situation, the orbital image that obtains in real time may size it is larger, in step s101 incited somebody to action
Larger-size orbital image has been first converted into suitable small-sized image, the rail level segmentation figure obtained in step s 103 in this way
Image as being also small size.Before executing step S104, need for the rail level segmented image of small size to be converted into in real time
The orbital image of acquisition image of a size in this way could obtain the rail level area maps in rail level segmented image to real-time
Train in front of orbital image on.
After on the rail level area maps to orbital image in rail level segmented image, train operator can be according to presentation
Image out judges whether there is barrier on track.For example, rail level region should be complete in the case where no barrier
Single colour stripe region, when there is barrier, it is possible that biggish cavity, i.e. barrier on colored rail level region
The color in the colored rail level region of the color and surrounding in the region at place has biggish difference, therefore can be biggish based on this
Color distinction judges whether there is barrier on rail level, reduces the probability that accident occurs.
Rail level semantic segmentation method provided by the embodiments of the present application, the orbital image that will acquire first are input to preparatory training
In good rail level semantic segmentation model, the corresponding class probability of each pixel in orbital image is obtained, class probability is used for table
Levy the probability that each pixel is rail level feature；According to the corresponding class probability of pixel each in orbital image and default classification
Probability threshold value, obtains the rail level segmented image of orbital image, rail level segmented image be marked with the rail level region in orbital image with
And background area.It can be seen that being divided from the image in front of train driving in the embodiment of the present application by image Segmentation Technology
Derailed face part, promotes the vision difference of rail level and other regions, is on track so as to help driver preferably to distinguish
It is no to have barrier, improve the accuracy rate of discrimination.
Based on the same inventive concept, rail level language corresponding with rail level semantic segmentation method is additionally provided in the embodiment of the present application
Adopted segmenting device, the principle solved the problems, such as due to the device in the embodiment of the present application and the above-mentioned rail level semanteme of the embodiment of the present application point
Segmentation method is similar, therefore the implementation of device may refer to the implementation of method, and overlaps will not be repeated.
It is shown in Figure 4, rail level semantic segmentation device provided by the embodiment of the present application, comprising:
Orbital image obtains module 41, for obtaining orbital image；
Pixel categorization module 42, for the orbital image to be input to preparatory trained rail level semantic segmentation model
In, the corresponding class probability of each pixel in the orbital image is obtained, the class probability is for characterizing each pixel
It is the probability of rail level feature；
Rail level segmented image obtains module 43, for according to the corresponding class probability of pixel each in the orbital image
With default class probability threshold value, the rail level segmented image of the orbital image is obtained, the rail level segmented image is marked with described
Rail level region and background area in orbital image.
Optionally, rail level semantic segmentation device provided by the embodiments of the present application, further includes:
Rail level semantic segmentation model training module 44, for obtaining multiple path images, and with the sample rail
Image corresponding sample in road marks image, the sample mark image tagged have the rail level region in the path image with
And background area；
The path image is input in the rail level semantic segmentation model constructed in advance, the path is obtained
The corresponding class probability of each pixel in image；
According to the corresponding class probability of pixel each in the path image and with each path image
Corresponding sample marks image, carries out epicycle training to the rail level semantic segmentation model；
By carrying out more wheel training to the rail level semantic segmentation model, the trained rail level semanteme point in advance is obtained
Cut model.
Optionally, the rail level semantic segmentation model training module 44 is specifically used for using following manner according to the sample
Each corresponding class probability of pixel and sample corresponding with each path image mark image in this orbital image,
Epicycle training is carried out to the rail level semantic segmentation model:
For each path image, according to the corresponding class probability of pixel each in the path image
And corresponding with path image sample marks image, obtains the corresponding friendship of each pixel in the path image
Pitch entropy loss；
According to the corresponding intersection entropy loss of pixel each in the path image, the rail level semantic segmentation mould is adjusted
The parameter of type；
Until determining and completing to the rail level semantic segmentation model after the training of all path images completion epicycle
Epicycle training.
Optionally, the rail level segmented image obtains module 43, specifically for for each picture in the orbital image
Vegetarian refreshments, if the corresponding class probability of the pixel is greater than the default class probability threshold value, it is determined that the pixel belongs to rail
Face region, and the pixel value of the pixel is arranged to default rail level pixel value；
In having adjusted the orbital image after the pixel value of each pixel, the first rail of the orbital image is obtained
Face segmented image.
Optionally, the rail level segmented image obtains module 43, also particularly useful for for each of described orbital image
Pixel, if the corresponding class probability of the pixel is not more than the default class probability threshold value, it is determined that the pixel category
In background area, and the pixel value of the pixel is arranged to preset non-rail level pixel value；
In having adjusted the orbital image after the pixel value of each pixel, the second rail of the orbital image is obtained
Face segmented image.
Optionally, rail level semantic segmentation device provided by the embodiments of the present application, further includes:
Mapping block 45, for by the rail level area maps in the rail level segmented image to the orbital image.
Rail level semantic segmentation device provided by the embodiments of the present application will acquire first when carrying out rail level segmentation
Orbital image is input in preparatory trained rail level semantic segmentation model, and it is corresponding to obtain each pixel in the orbital image
Class probability, the class probability is for characterizing the probability that each pixel is rail level feature；According in the orbital image
Each corresponding class probability of pixel and default class probability threshold value, obtain the rail level segmented image of the orbital image, institute
State the rail level region and background area that rail level segmented image is marked in the orbital image.It can be seen that the application is implemented
Rail level part is partitioned into from the image in front of train driving by image Segmentation Technology in example, promotes rail level and other regions
Whether vision difference has barrier so as to help driver preferably to distinguish on track, improves the accuracy rate of discrimination.
The embodiment of the present application also provides a kind of computer readable storage medium, stored on the computer readable storage medium
There is computer program, which executes above-mentioned rail level semantic segmentation method when being run by processor the step of.
Specifically, which can be general storage medium, such as mobile disk, hard disk, on the storage medium
Computer program when being run, be able to carry out above-mentioned rail level semantic segmentation method, so as to by image Segmentation Technology from
It is partitioned into rail level part in image in front of train driving, the vision difference of rail level and other regions is promoted, so as to help
Driver preferably distinguishes on track whether there is barrier, improves the accuracy rate of discrimination.
Corresponding to rail level semantic segmentation method provided by the embodiments of the present application, the embodiment of the present application also provides a kind of meters
Machine equipment is calculated, as shown in figure 5, the equipment includes memory 1000, processor 2000 and is stored on the memory 1000 and can
The computer program run on the processor 2000, wherein realization when above-mentioned processor 2000 executes above-mentioned computer program
The step of above-mentioned rail level semantic segmentation method.
Specifically, above-mentioned memory 1000 and processor 2000 can be general memory and processor, not do here
It is specific to limit, when the computer program of 2000 run memory 1000 of processor storage, it is able to carry out above-mentioned rail level semanteme point
Segmentation method promotes rail level so as to be partitioned into rail level part from the image in front of train driving by image Segmentation Technology
It improves and distinguishes so as to help driver preferably to distinguish whether there is barrier on track with the vision difference in other regions
Accuracy rate.
The computer program product of rail level semantic segmentation method and device provided by the embodiment of the present application, including store
The computer readable storage medium of program code, the instruction that said program code includes can be used for executing in previous methods embodiment
The method, specific implementation can be found in embodiment of the method, and details are not described herein.
In all examples being illustrated and described herein, any occurrence should be construed as merely illustratively, without
It is as limitation, therefore, other examples of exemplary embodiment can have different values.
It is apparent to those skilled in the art that for convenience and simplicity of description, the device of foregoing description
Specific work process, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.Provided herein
Several embodiments in, it should be understood that disclosed device and method may be implemented in other ways.It is above to be retouched
The Installation practice stated is only schematical, for example, the division of the unit, only a kind of logical function partition is practical
There may be another division manner when realization.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, each functional unit in each embodiment of the application can integrate in one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in the executable non-volatile computer-readable storage medium of a processor.Based on this understanding, the application
Technical solution substantially the part of the part that contributes to existing technology or the technical solution can be with software in other words
The form of product embodies, which is stored in a storage medium, including some instructions use so that
One computer equipment (can be personal computer, server or the network equipment etc.) executes each embodiment institute of the application
State all or part of the steps of method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (Read-Only
Memory, ROM), random access memory (Random Access Memory, RAM), magnetic or disk etc. is various to deposit
Store up the medium of program code.
Finally, it should be noted that embodiment described above, the only specific embodiment of the application, to illustrate the application
Technical solution, rather than its limitations, the protection scope of the application is not limited thereto, although with reference to the foregoing embodiments to this Shen
It please be described in detail, those skilled in the art should understand that: anyone skilled in the art
Within the technical scope of the present application, it can still modify to technical solution documented by previous embodiment or can be light
It is readily conceivable that variation or equivalent replacement of some of the technical features；And these modifications, variation or replacement, do not make
The essence of corresponding technical solution is detached from the spirit and scope of the embodiment of the present application technical solution, should all cover the protection in the application
Within the scope of.Therefore, the protection scope of the application shall be subject to the protection scope of the claim.
Claims (10)
1. a kind of rail level semantic segmentation method characterized by comprising
Obtain orbital image；
The orbital image is input in preparatory trained rail level semantic segmentation model, is obtained each in the orbital image
The corresponding class probability of pixel, the class probability is for characterizing the probability that each pixel is rail level feature；
According to the corresponding class probability of pixel each in the orbital image and default class probability threshold value, the track is obtained
The rail level segmented image of image, the rail level segmented image are marked with rail level region and background area in the orbital image
Domain.
2. the method according to claim 1, wherein obtaining the trained rail level in advance by following manner
Semantic segmentation model:
Multiple path images are obtained, and sample corresponding with the path image marks image, the sample mark
Note image tagged has rail level region and background area in the path image；
The path image is input in the rail level semantic segmentation model constructed in advance, the path image is obtained
In the corresponding class probability of each pixel；
According to the corresponding class probability of pixel each in the path image and corresponding with each path image
Sample mark image, to the rail level semantic segmentation model carry out epicycle training；
By carrying out more wheel training to the rail level semantic segmentation model, the trained rail level semantic segmentation mould in advance is obtained
Type.
3. according to the method described in claim 2, it is characterized in that, described according to each pixel in the path image
Corresponding class probability and sample corresponding with each path image mark image, to the rail level semantic segmentation model
Carry out epicycle training, comprising:
For each path image, according to the corresponding class probability of pixel each in the path image and
Corresponding with path image sample marks image, obtains the corresponding cross entropy of each pixel in the path image
Loss；
According to the corresponding intersection entropy loss of pixel each in the path image, the rail level semantic segmentation model is adjusted
Parameter；
Until determining the epicycle completed to the rail level semantic segmentation model after the training of all path images completion epicycle
Training.
4. the method according to claim 1, wherein described corresponding according to pixel each in the orbital image
Class probability and default class probability threshold value, obtain the rail level segmented image of the orbital image, comprising:
For each pixel in the orbital image, if the corresponding class probability of the pixel is greater than the default classification
Probability threshold value, it is determined that the pixel belongs to rail level region, and the pixel value of the pixel is arranged to default rail level pixel value；
In having adjusted the orbital image after the pixel value of each pixel, the first rail level point of the orbital image is obtained
Cut image.
5. according to the method described in claim 4, it is characterized in that, described corresponding according to pixel each in the orbital image
Class probability and default class probability threshold value, obtain the rail level segmented image of the orbital image, comprising:
For each pixel in the orbital image, if the corresponding class probability of the pixel is no more than described default point
Class probability threshold value, it is determined that the pixel belongs to background area, and the pixel value of the pixel is arranged to preset non-rail level picture
Element value；
In having adjusted the orbital image after the pixel value of each pixel, the second rail level point of the orbital image is obtained
Cut image.
6. the method according to claim 1, wherein the method also includes: will be in the rail level segmented image
Rail level area maps to the orbital image on.
7. a kind of rail level semantic segmentation device characterized by comprising
Orbital image obtains module, for obtaining orbital image；
Pixel categorization module is obtained for the orbital image to be input in preparatory trained rail level semantic segmentation model
The corresponding class probability of each pixel in the orbital image is taken, the class probability is rail level for characterizing each pixel
The probability of feature；
Rail level segmented image obtains module, is used for according to the corresponding class probability of pixel each in the orbital image and presets
Class probability threshold value, obtains the rail level segmented image of the orbital image, and the rail level segmented image is marked with the trajectory diagram
Rail level region and background area as in.
8. device according to claim 7, which is characterized in that further include:
Rail level semantic segmentation model training module, for obtaining multiple path images, and with the path image
Corresponding sample marks image, and the sample mark image tagged has rail level region and background in the path image
Region；
The path image is input in the rail level semantic segmentation model constructed in advance, the path image is obtained
In the corresponding class probability of each pixel；
According to the corresponding class probability of pixel each in the path image and corresponding with each path image
Sample mark image, to the rail level semantic segmentation model carry out epicycle training；
By carrying out more wheel training to the rail level semantic segmentation model, the trained rail level semantic segmentation mould in advance is obtained
Type.
9. device according to claim 7, which is characterized in that the rail level segmented image obtains module, is specifically used for needle
To each pixel in the orbital image, if the corresponding class probability of the pixel is greater than the default class probability threshold
Value, it is determined that the pixel belongs to rail level region, and the pixel value of the pixel is arranged to default rail level pixel value；
In having adjusted the orbital image after the pixel value of each pixel, the first rail level point of the orbital image is obtained
Cut image；
The rail level segmented image obtains module, also particularly useful for each pixel being directed in the orbital image, if should
The corresponding class probability of pixel is not more than the default class probability threshold value, it is determined that and the pixel belongs to background area, and
The pixel value of the pixel is arranged to preset non-rail level pixel value；
In having adjusted the orbital image after the pixel value of each pixel, the second rail level point of the orbital image is obtained
Cut image.
10. device according to claim 7, which is characterized in that further include:
Mapping block, for by the rail level area maps in the rail level segmented image to the orbital image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910020741.9A CN109784265A (en) | 2019-01-09 | 2019-01-09 | A kind of rail level semantic segmentation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910020741.9A CN109784265A (en) | 2019-01-09 | 2019-01-09 | A kind of rail level semantic segmentation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109784265A true CN109784265A (en) | 2019-05-21 |
Family
ID=66500134
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910020741.9A Pending CN109784265A (en) | 2019-01-09 | 2019-01-09 | A kind of rail level semantic segmentation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109784265A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110795779A (en) * | 2019-08-29 | 2020-02-14 | 西南交通大学 | Method and device for establishing rail surface mapping model |
| CN113052048A (en) * | 2021-03-18 | 2021-06-29 | 北京百度网讯科技有限公司 | Traffic incident detection method and device, road side equipment and cloud control platform |
| CN114005097A (en) * | 2020-07-28 | 2022-02-01 | 株洲中车时代电气股份有限公司 | Train operation environment real-time detection method and system based on image semantic segmentation |
| CN114694125A (en) * | 2022-05-31 | 2022-07-01 | 杭州蓝芯科技有限公司 | Data processing method for track foreign matter identification |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101021417A (en) * | 2007-03-21 | 2007-08-22 | 于起峰 | Long straight rail geometric parameter photogrammetry |
| CN108596184A (en) * | 2018-04-25 | 2018-09-28 | 清华大学深圳研究生院 | Training method, readable storage medium storing program for executing and the electronic equipment of image, semantic parted pattern |
-
2019
- 2019-01-09 CN CN201910020741.9A patent/CN109784265A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101021417A (en) * | 2007-03-21 | 2007-08-22 | 于起峰 | Long straight rail geometric parameter photogrammetry |
| CN108596184A (en) * | 2018-04-25 | 2018-09-28 | 清华大学深圳研究生院 | Training method, readable storage medium storing program for executing and the electronic equipment of image, semantic parted pattern |
Non-Patent Citations (2)
| Title |
|---|
| DAVY NEVEN 等: "Towards End-to-End Lane Detection：an Instance Segmentation Approach", 《ARXIV》 * |
| 韩江洪 等: "基于空间卷积神经网络的井下轨道检测方法", 《电子测量与仪器学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110795779A (en) * | 2019-08-29 | 2020-02-14 | 西南交通大学 | Method and device for establishing rail surface mapping model |
| CN114005097A (en) * | 2020-07-28 | 2022-02-01 | 株洲中车时代电气股份有限公司 | Train operation environment real-time detection method and system based on image semantic segmentation |
| CN113052048A (en) * | 2021-03-18 | 2021-06-29 | 北京百度网讯科技有限公司 | Traffic incident detection method and device, road side equipment and cloud control platform |
| CN113052048B (en) * | 2021-03-18 | 2024-05-10 | 阿波罗智联(北京)科技有限公司 | Traffic event detection method and device, road side equipment and cloud control platform |
| CN114694125A (en) * | 2022-05-31 | 2022-07-01 | 杭州蓝芯科技有限公司 | Data processing method for track foreign matter identification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109784265A (en) | A kind of rail level semantic segmentation method and device | |
| US10133941B2 (en) | Method, apparatus and device for detecting lane boundary | |
| CN106295646B (en) | A kind of registration number character dividing method and device based on deep learning | |
| US20190122059A1 (en) | Signal light detection | |
| EP3767520B1 (en) | Method, device, equipment and medium for locating center of target object region | |
| CN109635656A (en) | Vehicle attribute recognition methods, device, equipment and medium neural network based | |
| US20160117562A1 (en) | Traffic sign recognizing apparatus and operating method thereof | |
| KR20210078530A (en) | Lane property detection method, device, electronic device and readable storage medium | |
| CN105260713A (en) | Method and device for detecting lane line | |
| CN110723432A (en) | Garbage classification method and augmented reality equipment | |
| CN105389553A (en) | Living body detection method and apparatus | |
| CN111259878A (en) | Method and equipment for detecting text | |
| CN108537782A (en) | A method of building images match based on contours extract with merge | |
| CN110390314B (en) | Visual perception method and equipment | |
| DE102013222322A1 (en) | Method and device for providing Augmented Reality | |
| CN109034136A (en) | Image processing method, device, picture pick-up device and storage medium | |
| CN105426863A (en) | Method and device for detecting lane line | |
| CN114037966A (en) | High-precision map feature extraction method, device, medium and electronic equipment | |
| CN101369312B (en) | Method and equipment for detecting intersection in image | |
| CN111079675A (en) | Driving behavior analysis method based on target detection and target tracking | |
| CN111950345A (en) | Camera identification method and device, electronic equipment and storage medium | |
| US20120189161A1 (en) | Visual attention apparatus and control method based on mind awareness and display apparatus using the visual attention apparatus | |
| CN114419603A (en) | Automatic driving vehicle control method and system and automatic driving vehicle | |
| CN110110590A (en) | The determination method and Related product in congestion lane | |
| CN104809438B (en) | A kind of method and apparatus for detecting electronic eyes |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190521 |
|
| RJ01 | Rejection of invention patent application after publication |