Invention content
In view of the above defects of the prior art, the present invention provide it is a kind of identify train model method, apparatus andEquipment.
An aspect of of the present present invention provides a kind of method for identifying train model, including：Acquire the picture number of train to be identifiedAccording to, and extract preset characteristic area in image data；By calculating the characteristic value of each characteristic area, acquisition and imageThe corresponding fisrt feature value set of data；The train model identification model that the input of fisrt feature value set is pre-established, obtainsThe target model corresponding with train to be identified of output.
Wherein, it is further included after described the step of obtaining the target model corresponding with train to be identified exported：Obtain targetThe corresponding second feature value set of model；Characteristic area each in fisrt feature value set and second feature value set is corresponded to respectivelyCharacteristic value be compared, obtain the similarity between fisrt feature value set and second feature value set；If similarity is more thanPredetermined threshold value then confirms the model target model of train to be identified.
Wherein, it is further included after the step of image data of the acquisition train to be identified：Image increasing is carried out to image dataStrength is managed and/or picture smooth treatment；Wherein, described image enhancing processing, which specifically includes, carries out at gray scale stretching image dataReason, histogram equalization processing and normalized；Described image smoothing processing is specifically included using the threshold value chosen, to imagePixel in data is filtered, and the threshold value of the selection is pixel and the mean-square value of surrounding ash scale.
Wherein, it is specifically included the step of preset characteristic area in the extraction image data：To described image data intoRow identification obtains the profile information of each characteristic area；And according to the profile information, extract the corresponding profile of each characteristic area.
Wherein, the step of image data of the acquisition train to be identified specifically includes：In the operation of train to be identifiedCheng Zhong is continuously shot train to be identified, obtains multiple sub-image datas；According to time sequencing, to multiple subgraph numbersAccording to splicing is carried out, image data is obtained, wherein, described image data include the complete shape data of train to be identified.
Wherein, it was also wrapped before described the step of fisrt feature value set is inputted the train model identification model pre-establishedIt includes：Using characteristic value collection and train model corresponding with characteristic value collection carries out model instruction as input value and output valvePractice, obtain the train model identification model that training is completed.
Wherein, the characteristic area includes headstock region, vehicle body region, vehicle window region, car roof area, wheel region and vehicleAt least one of join domain between compartment.
Another aspect of the present invention provides a kind of device for identifying train model, including：Extraction module is waited to know for acquiringThe image data of other train, and extract preset characteristic area in image data；Acquisition module calculates each for passing throughThe characteristic value of characteristic area obtains fisrt feature value set corresponding with image data；Identification module, for by the First EigenvalueThe train model identification model that set input pre-establishes obtains the target model corresponding with train to be identified of output.
Another aspect of the present invention provides a kind of equipment for identifying excitation model, including：At least one processor；And withAt least one processor of the processor communication connection, wherein：The memory is stored with what can be performed by the processorProgram instruction, the processor call described program instruction to be able to carry out the identification train model that the above-mentioned aspect of the present invention providesMethod, such as including：The image data of train to be identified is acquired, and extracts preset characteristic area in image data；It is logicalThe characteristic value for calculating each characteristic area is crossed, obtains fisrt feature value set corresponding with image data；By the First Eigenvalue collectionThe train model identification model that input pre-establishes is closed, obtains the target model corresponding with train to be identified of output.
Another aspect of the present invention provides a kind of non-transient computer readable storage medium storing program for executing, and the non-transient computer is readableStorage medium stores computer instruction, and the computer instruction makes the computer perform the identification that the above-mentioned aspect of the present invention providesThe method of train model, such as including：The image data of train to be identified is acquired, and extracts preset spy in image dataLevy region；By calculating the characteristic value of each characteristic area, fisrt feature value set corresponding with image data is obtained；By firstThe train model identification model that characteristic value collection input pre-establishes obtains the object type corresponding with train to be identified of outputNumber.
The method, apparatus and equipment of identification train model provided by the invention, by being based on preset characteristic areaCorresponding characteristic value collection judges train model using the difference between characteristic area, realizes to train modelIdentification, and with degree of precision.
Specific embodiment
Purpose, technical scheme and advantage to make the embodiment of the present invention are clearer, below in conjunction with the embodiment of the present inventionIn attached drawing, the technical solution in the embodiment of the present invention is explicitly described, it is clear that described embodiment be the present inventionPart of the embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not havingAll other embodiments obtained under the premise of creative work are made, shall fall within the protection scope of the present invention.
Fig. 1 is the flow diagram of the method for identification train model provided in an embodiment of the present invention, as shown in Figure 1, including：Step 101, the image data of train to be identified is acquired, and extracts preset characteristic area in image data；Step 102,By calculating the characteristic value of each characteristic area, fisrt feature value set corresponding with image data is obtained；Step 103, byThe train model identification model that the input of one characteristic value collection pre-establishes obtains the object type corresponding with train to be identified of outputNumber.
Wherein, characteristic area is the subregion in the image of collected train to be identified, which can show thisThe difference of model train and other model trains；And characteristic area should be the area with significant difference between different model trainDomain, for example, the train nose shape of different model train differs, then can be by the region where the headstock profile in image dataAs characteristic area.
Wherein, characteristic value can reflect the characteristic of characteristic area, the characteristic value of the same characteristic area of different model trainIt is different；It in practical applications, can be using the cryptographic Hash for the characteristic area being calculated as the characteristic value of characteristic area.
Wherein, train model identification model is based on trained the obtained model of neural network pre-established；The mouldThe input value of type is the corresponding characteristic value collection of a certain model train, and the output valve of the model is train model.
In a step 101, equipment is produced by images such as video camera or cameras first, train to be identified is clappedIt takes the photograph, obtains the corresponding image data of train to be identified, wherein, the train to be identified during shooting can be motion state or static shapeState；Image data should can carry out the shape of train displaying to a certain extent, that is, include all or part of trainShape；Then operation is extracted to image data, that is, extracts preset characteristic area, such as headstock region；Characteristic areaDomain can be one or more.
In a step 102, according to the characteristic area extracted in step 101, feature can be calculated to each characteristic areaValue, the characteristic values of multiple characteristic areas form a characteristic value collection, as fisrt feature value set (each spy in characteristic value collectionThe sequence of value indicative is to be obtained according to location information of each characteristic area in train image data, such as the operation in trainOn direction, each characteristic value of arrangement successively from headstock to parking stall)；The image data phase of fisrt feature value set and train to be identifiedIt is corresponding, the situation of each characteristic area of train to be identified is described, and characteristic value collection reflects the type of train to be identified on the wholeFeature on number.
In step 103, the fisrt feature value set obtained according to step 102, using fisrt feature value set as inputValue is input to the train model identification model pre-established, and train identification model exports train corresponding with fisrt feature value setModel, as target model；Train model identification model is to be trained acquisition based on training set, and training set includes mayThe train model and corresponding characteristic value collection of appearance.
The method of identification train model provided in an embodiment of the present invention, it is corresponding by being based on preset characteristic areaCharacteristic value collection judges train model using the difference between characteristic area, realizes the identification to train model, andWith degree of precision.
On the basis of any of the above-described embodiment, the step of the target model corresponding with train to be identified for obtaining outputIt is further included after rapid：Obtain the corresponding second feature value set of target model；Respectively to fisrt feature value set and Second EigenvalueThe corresponding characteristic value of each characteristic area is compared in set, is obtained between fisrt feature value set and second feature value setSimilarity；If similarity is more than predetermined threshold value, the model target model of train to be identified is confirmed.
Specifically, due to train model identification model be based on neural metwork training obtain, and training precision be withThe quantity of training sample increases and improves；And in order to ensure the accuracy of train type identifier, export target model in modelAfterwards, the accuracy of target model can be verified.
Since the corresponding second feature value set of target model in the training process of model, can be inputted, as the targetThe characteristic value collection of model standard；Therefore, after model exports object module, second as standard can further be gotCharacteristic value collection.Since fisrt feature value set and second feature value set are all based on what identical characteristic area was got,Therefore the corresponding characteristic value of each characteristic area is compared successively, it can be determined that the corresponding characteristic value phase of each characteristic areaAs degree (be greater than 80% and think that characteristic area is identical, otherwise differ)；Finally each characteristic value region is corresponded toCharacteristic value be compared respectively, so as to obtain total similarity；If total similarity is more than predetermined threshold value, it is confirmed as targetModel, otherwise it is assumed that the target model of train model identification model output is not the model of train to be identified.
In the case of any of the above-described embodiment, the step of the image data of acquisition train to be identified after further include：Image enhancement processing and/or picture smooth treatment are carried out to image data；Wherein, described image enhancing processing is specifically included to figureAs data carry out gray scale stretching processing, histogram equalization processing and normalized；Described image smoothing processing specifically includesUsing the threshold value of selection, the pixel in image data is filtered, the threshold value of the selection is pixel and surrounding grayscaleThe mean-square value of difference.
Specifically, it in order to make image that there is preferable effect, needs to optimize image data processing.
Wherein, the purpose of image enhancement processing is to reduce noise, enhances grey-scale contrast so that image is more clearReally, convenient for the accuracy of follow-up car body characteristic area thresholding extraction.Concrete processing procedure is to draw the image elder generation gray scale gotStretch and histogram equalization, then the direction of image, size and unitary of illumination, maximum possible reduce because illumination, camera quality,The noise jamming caused by image such as camera site.
Wherein, the purpose of picture smooth treatment is in order to which whole image is allowed to obtain uniform chiaroscuro effect.Specific placeReason process is to choose the pixel of whole image to handle as threshold value with the mean-square value of its periphery ash scale.
On the basis of any of the above-described embodiment, the step of image data of acquisition train to be identified, specifically includes：In the operational process of train to be identified, train to be identified is continuously shot, obtains multiple sub-image datas；According to the timeSequentially, splicing is carried out to multiple sub-image datas, obtains image data, wherein, described image data include row to be identifiedThe complete shape data of vehicle.
Specifically, since train vehicle body is longer, in the image data for obtaining train to be identified, use is needed to take the photographThe images such as camera or camera production equipment is continuously shot train；Shooting all carries out a part of region of train every timeShooting obtains sub-image data every time；Since sub-image data is chronologically-based acquisition, in order toObtain a complete image data for including the complete shape data of train, can with it is chronologically-based to sub-image data intoRow splicing.
On the basis of any of the above-described embodiment, the train model that the input of fisrt feature value set is pre-established is knownIt was further included before the step of other model：Using characteristic value collection and train model corresponding with characteristic value collection is as input valueModel training is carried out with output valve, obtains the train model identification model that training is completed.
In order to obtain train model identification model, need to be trained the model initially set up using training set.TrainingCollection acquires multiple images data, and obtain multiple characteristic value collections specifically, for a kind of train of model；Characteristic value collectionAs trained input data, corresponding train model are the output data or label of training；It can be arranged after trainedVehicle identification model.
It should be noted that in the training process, the corresponding multiple characteristic value collections of each model train can be integrated, are obtainedThe second feature value set as standard is obtained, the similarity judgement after being exported for result.
On the basis of any of the above-described embodiment, the characteristic area include headstock region, vehicle body region, vehicle window region,At least one of join domain between car roof area, wheel region and compartment.
Specifically, when setting characteristic area, the region with significant difference should be chosen；It such as can be from headstock regionExtraction drives chamber shape, goes out color from vehicle body extracted region；From other corresponding extracted regions such as airflow design feature.
During specifically used, the characteristic area and interregional geometry (position) that may be used 40 or more are madeFor matching, the accuracy of identification is improved.In the car body characteristic area obtained every time along with service life, weather, illumination etc.The influence of factor all will not be duplicate, can also mend the characteristic area in freshly harvested car body in comparison processIt fills or updates in the car body feature templates stored in advance, constantly improve car body feature templates, can thus cause car bodyIt is higher and higher to be identified by rate, accuracy rate.
Fig. 2 is the structure diagram of the device of identification train model provided in an embodiment of the present invention, as shown in Fig. 2, including：Extraction module 201 for acquiring the image data of train to be identified, and extracts preset characteristic area in image data；Acquisition module 202 for passing through the characteristic value for calculating each characteristic area, obtains the First Eigenvalue collection corresponding with image dataIt closes；Identification module 203 for the train model identification model for pre-establishing the input of fisrt feature value set, obtains outputTarget model corresponding with train to be identified.
Wherein, train to be identified is shot by extraction module 201 first, obtains the corresponding figure of train to be identifiedAs data, wherein, the train to be identified during shooting can be motion state or stationary state；Image data should can be to trainShape carry out displaying to a certain extent, that is, include all or part of shape of train；Then extraction module 201 is to imageData extract operation, that is, extract preset characteristic area, such as headstock region；Characteristic area can be one or moreIt is a.
Wherein, acquisition module 202, can to each characteristic area according to the characteristic area extracted in extraction module 201Characteristic value is calculated, the characteristic value of multiple characteristic areas forms a characteristic value collection, as fisrt feature value set (characteristic value collectionThe sequence of each characteristic value is to be obtained according to location information of each characteristic area in train image data, such as arranging in conjunctionOn the traffic direction of vehicle, each characteristic value of arrangement successively from headstock to parking stall)；The figure of fisrt feature value set and train to be identifiedAs data are corresponding, describe the situation of each characteristic area of train to be identified, and characteristic value collection reflect on the whole it is to be identifiedFeature in the model of train.
Wherein, the fisrt feature value set that identification module 203 is obtained according to acquisition module 202, by fisrt feature value setThe train model identification model pre-established, the output of train identification model and fisrt feature value set pair are input to as input valueThe model for the train answered, as target model；Train model identification model is to be trained acquisition, training set based on training setIncluding the train model being likely to occur and corresponding characteristic value collection.
The device of identification train model provided in an embodiment of the present invention, it is corresponding by being based on preset characteristic areaCharacteristic value collection judges train model using the difference between characteristic area, realizes the identification to train model, andWith degree of precision.
On the basis of any of the above-described embodiment, described device further includes comparison means, is corresponded to for obtaining target modelSecond feature value set；Respectively to the corresponding characteristic value of characteristic area each in fisrt feature value set and second feature value setIt is compared, obtains the similarity between fisrt feature value set and second feature value set；If similarity is more than predetermined threshold value,Then confirm the model target model of train to be identified.
On the basis of any of the above-described embodiment, the extraction element further includes：Processing unit, for image data intoRow image enhancement processing and/or picture smooth treatment；Wherein, described image enhancing processing is specifically included carries out ash to image dataSpend stretch processing, histogram equalization processing and normalized；Described image smoothing processing is specifically included using the threshold chosenValue, is filtered the pixel in image data, and the threshold value of the selection is pixel and the mean-square value of surrounding ash scale.
On the basis of any of the above-described embodiment, the extraction module 201 is specifically used for, and described image data are knownNot, the profile information of each characteristic area is obtained；And according to the profile information, extract the corresponding profile of each characteristic area.
On the basis of any of the above-described embodiment, the extraction module 201 is specifically used for：In the operation of train to be identifiedCheng Zhong is continuously shot train to be identified, obtains multiple sub-image datas；According to time sequencing, to multiple subgraph numbersAccording to splicing is carried out, image data is obtained, wherein, described image data include the complete shape data of train to be identified.
On the basis of any of the above-described embodiment, described device further includes training module, for by characteristic value collection andTrain model corresponding with characteristic value collection carries out model training respectively as input value and output valve, obtains the row that training is completedVehicle identification model.
On the basis of any of the above-described embodiment, the characteristic area include headstock region, vehicle body region, vehicle window region,At least one of join domain between car roof area, wheel region and compartment.
Fig. 3 is the structure diagram of the equipment of identification train model provided in an embodiment of the present invention, as shown in figure 3, this setsIt is standby to include：At least one processor 301；And at least one processor 302 with the processor 301 communication connection, wherein：The memory 302 is stored with the program instruction that can be performed by the processor 301, and the processor 301 calls described programInstruction is able to carry out the method for identification train model that the various embodiments described above are provided, such as including：Acquire train to be identifiedImage data, and extract preset characteristic area in image data；By calculating the characteristic value of each characteristic area, obtainFisrt feature value set corresponding with image data；The train type identifier mould that the input of fisrt feature value set is pre-establishedType obtains the target model corresponding with train to be identified of output.
The embodiment of the present invention also provides a kind of non-transient computer readable storage medium storing program for executing, the non-transient computer readable storageMedium storing computer instructs, which makes computer perform the side of identification train model that corresponding embodiment is providedMethod, such as including：The image data of train to be identified is acquired, and extracts preset characteristic area in image data；Pass throughThe characteristic value of each characteristic area is calculated, obtains fisrt feature value set corresponding with image data；By fisrt feature value setThe train model identification model pre-established is inputted, obtains the target model corresponding with train to be identified of output.
The embodiments such as the equipment of identification train model described above are only schematical, wherein as separating componentThe unit of explanation may or may not be physically separate, and the component shown as unit can be or can alsoIt is not physical unit, you can be located at a place or can also be distributed in multiple network element.It can be according to realityIt needs that some or all of module therein is selected to realize the purpose of this embodiment scheme.Those of ordinary skill in the art are notIn the case of paying performing creative labour, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment canIt is realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on such understanding, onTechnical solution is stated substantially in other words to embody the part that the prior art contributes in the form of software product, it shouldComputer software product can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including several fingersIt enables and (can be personal computer, server or the network equipment etc.) so that computer equipment is used to perform each implementationCertain Part Methods of example or embodiment.
Finally it should be noted that：The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；AlthoughThe present invention is described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that：It still may be usedTo modify to the technical solution recorded in foregoing embodiments or carry out equivalent replacement to which part technical characteristic；And these modification or replace, various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution spirit andRange.