CN105718597A - Data retrieving method and system thereof - Google Patents

Abstract

The embodiment of the invention provides a data retrieving method and system. The method comprises the following steps: acquiring a target object used as retrieval reference; acquiring target metadata; extracting target retrieval information corresponding to a target object from the target metadata; retrieving in a preset hierarchical tree-shaped data index structure based on various attributes included by the target retrieving information to acquire a retrieval result corresponding to the target object. According to the data retrieving method and the system thereof, the target retrieval information includes target space attribute, target time attribute and target foreground image attribute in the target metadata while the preset hierarchical tree-shaped data index structure is built according to the various attributes included in the target retrieval information in a plurality of preset metadata; therefore, due to the application of the data retrieving method and the system thereof, the retrieval efficiency is improved; meanwhile, the storage pressure of an intelligent traffic monitoring system is relieved.

Description

A kind of data retrieval method and system

Technical field

The present invention relates to wireless communications application field, particularly to a kind of data retrieval method and system.

Background technology

In recent years, along with the lifting of the fast development of auto manufacturing and national life quality, automobile has been increasingly becoming the vehicles of most main flow instantly, and Traffic Administration Bureau of the Ministry of Public Security announces result display cut-off in February, 2015, and the vehicles number of China has reached 2.64 hundred million.Quick growth with this vehicles number, the incidence rate of vehicle accident and break in traffic rules and regulations also increases, in order to realize the automatically identification to traffic object and all kinds of traffic abnormal incident be retrieved, intelligent traffic monitoring system (IntelligentTrafficSurveillanceSystem), because of the feature that its recognition speed is fast, retrieval precision is high, personnel attrition is few, is used in road traffic video monitoring system more.

At present, it is generally all use metadata concept when traffic surveillance videos is stored and retrieves by intelligent traffic monitoring system, wherein, metadata is for recording the information of each generic attribute of data, for example: for video image, the metadata of its correspondence generally comprises space attribute, time attribute, color attribute, vehicle attribute and vehicle foreground image attributes.Traditional index structure is according to the data attribute comprised in metadata, multiple metadata to be grouped, same metadata will be in different groups according to contained different attribute, and be respectively independent between group data, the multi-group data obtained is deposited in data base, when target metadata is retrieved, retrieve according in each corresponding attribute packet in data base of the attribute in target metadata, retrieval is only capable of an attribute is retrieved every time, so effectiveness of retrieval is relatively low, further, owing to the rule of classification of data makes the databases preserved the increase along with metadata sharply be increased, this will bring very big storage pressure to intelligent traffic monitoring system.

Summary of the invention

The embodiment of the invention discloses a kind of data retrieval method and system, to improve recall precision, alleviate the storage pressure of intelligent traffic monitoring system simultaneously.

In order to achieve the above object, embodiments providing a kind of data retrieval method, be applied to intelligent monitor system, method includes step:

Obtain the destination object as retrieval foundation；

Determine the target metadata of described destination object；

Extracting the target retrieval information corresponding to described destination object from described target metadata, wherein, described target retrieval information includes: object space attribute in described target metadata, object time attribute and target prospect image attributes；

According to each generic attribute included by described target retrieval information, retrieve in default hierarchical tree data directory structure, it is thus achieved that the retrieval result corresponding to described destination object；

Wherein, described default hierarchical tree data directory structure is that each generic attribute included by the target retrieval information in multiple default metadata is set up, the plurality of default metadata is as the metadata in the default metadatabase of default collection of metadata, and, in described default hierarchical tree data directory structure, root node is determined according to the space attribute in the plurality of default metadata, the second layer is determined according to the time attribute in the plurality of default metadata, and leaf node is determined according to the foreground image attribute in the plurality of default metadata.

Optionally, described determine that the mode of leaf node includes according to the foreground image attribute in the plurality of default metadata:

Foreground image attribute in each default metadata is carried out Visual Feature Retrieval Process, it is thus achieved that the visual vocabulary of foreground image attribute；

Calculate the described visual vocabulary visual vocabulary weight for default metadatabase of each default metadata, and the visual vocabulary weight obtained is ranked up by preset rules, obtain visual vocabulary weight table；

According to described visual vocabulary weight table, set up the leaf node in described default hierarchical tree data directory structure.

Optionally, described each generic attribute included by described target retrieval information, retrieve in default hierarchical tree data directory structure, it is thus achieved that the retrieval result corresponding to described destination object, including:

Described target prospect image attributes is carried out Visual Feature Retrieval Process, it is thus achieved that the target visual vocabulary of described target prospect image；

Calculate the described target visual vocabulary target visual term weight for described default metadatabase；

Each generic attribute included by described target retrieval information and described target visual term weight, retrieve, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure.

Optionally, the described visual vocabulary of each default metadata of described calculating passes through equation below for the visual vocabulary weight of default metadatabase:

$w_{i,d}={\mathrm{tf}}_{i,d}\×log\left(\frac{N}{{\mathrm{df}}_{i,d}}\right)$

In formula, tf_i,dFor the visual vocabulary i frequency occurred in current preset metadata, df_i,dThe frequency that the metadata comprising visual vocabulary i occurs in default metadatabase, N presets the number of all metadata in metadatabase.

Optionally, described each generic attribute included by described target retrieval information and described target visual term weight, retrieve in default hierarchical tree data directory structure, it is thus achieved that the retrieval result corresponding to described destination object, including:

Utilize each generic attribute included by described target retrieval information, each respective layer in described default hierarchical tree data directory structure is successively retrieved, wherein, successively carry out the mode retrieved: in default hierarchical tree data directory structure, determine target root node according to the space attribute in described target retrieval information, retrieve under described target root node further according to the time attribute in described target retrieval information, determine this node layer, retrieve further according under the determined node of last layer corresponding with this attribute in described default hierarchical tree data directory structure of other each attributes in described target retrieval information, finally determine leaf node；

The target visual term weight in described target metadata included by described target retrieval information carries out Similarity Measure under leaf node described in described default hierarchical tree data directory structure, and the Similarity value calculated is arranged in descending order, take the foreground image corresponding to front predetermined number metadata as the retrieval result corresponding to described destination object.

Optionally, carry out Similarity Measure under the leaf node described in described default hierarchical tree data directory structure of the target visual term weight in the described target metadata included by described target retrieval information and pass through below equation:

$score(q,p)=\frac{V_q\×V_p}{\left|V_q\right|\left|V_p\right|}=\frac{{\mathrm{\Σ}}_{i=1}^n{w}_{i,q}\×w_{i,p}}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}_{i,q}^2}*\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}_{i,p}^2}}$

In formula, q is the target metadata of destination object, and p is the metadata in described default hierarchical tree data directory structure, V_qFor the visual vocabulary weight vectors in target retrieval information, V_pFor the visual vocabulary weight vectors of metadata, w in default hierarchical tree data directory structure_i,qFor V_qComponent, w_i,pFor V_pComponent.

In order to achieve the above object, embodiments providing a kind of data retrieval system, be applied to intelligent monitor system, system includes:

Destination object obtains module, for obtaining the destination object as retrieval foundation；

Target metadata determines module, for determining the target metadata of described destination object；

Target retrieval information acquisition module, for extracting the target retrieval information corresponding to described destination object from described target metadata, wherein, described target retrieval information includes: object space attribute in described target metadata, object time attribute and target prospect image attributes；

Retrieval result obtains module, for each generic attribute included by described target retrieval information, retrieves, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure；

Wherein, described default hierarchical tree data directory structure is that index structure sets up the module each generic attribute foundation included by the target retrieval information in multiple default metadata, the plurality of default metadata is as the metadata in the default metadatabase of default collection of metadata, and, in described default hierarchical tree data directory structure, root node is determined according to the space attribute in the plurality of default metadata, the second layer is determined according to the time attribute in the plurality of default metadata, leaf node is determined according to the foreground image attribute in the plurality of default metadata.

Optionally, described index structure is set up module and is determined that the detailed process of leaf node includes according to the foreground image attribute in the plurality of default metadata:

Foreground image attribute in each default metadata is carried out Visual Feature Retrieval Process, it is thus achieved that the visual vocabulary of foreground image attribute；

Calculate the described visual vocabulary visual vocabulary weight for default metadatabase of each default metadata, and the visual vocabulary weight obtained is ranked up by preset rules, obtain visual vocabulary weight table；

According to described visual vocabulary weight table, set up the leaf node in described default hierarchical tree data directory structure.

Optionally, described retrieval result acquisition module includes:

Target visual vocabulary obtains submodule, for described target prospect image attributes is carried out Visual Feature Retrieval Process, it is thus achieved that the target visual vocabulary of described target prospect image；

Target visual term weight calculating sub module, for calculating the described target visual vocabulary target visual term weight for described default metadatabase；

Retrieval result obtains submodule, for each generic attribute included by described target retrieval information and described target visual term weight, retrieves, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure.

Optionally, described retrieval result obtains submodule, including:

Layering retrieval unit, for each generic attribute included by described target retrieval information, each respective layer in described default hierarchical tree data directory structure is successively retrieved, wherein, successively carry out the mode retrieved: in default hierarchical tree data directory structure, determine target root node according to the space attribute in described target retrieval information, retrieve under described target root node further according to the time attribute in described target retrieval information, determine this node layer, retrieve further according under the determined node of last layer corresponding with this attribute in described default hierarchical tree data directory structure of other each attributes in described target retrieval information, finally determine leaf node；

Retrieval result determines unit, for carrying out Similarity Measure under the leaf node described in described default hierarchical tree data directory structure of the target visual term weight in the described target metadata included by described target retrieval information, and the Similarity value calculated is arranged in descending order, take the foreground image corresponding to front predetermined number metadata as the retrieval result corresponding to described destination object.

Embodiments provide a kind of data retrieval method and system, it is thus achieved that as the destination object of retrieval foundation；And obtain target metadata；The target retrieval information corresponding to described destination object is extracted from target metadata；Based on each generic attribute included by described target retrieval information, retrieve in default hierarchical tree data directory structure, it is thus achieved that the retrieval result corresponding to described destination object；The object space attribute in target metadata, object time attribute and target prospect image attributes is included due to target retrieval information, and the hierarchical tree data directory structure preset is each generic attribute foundation included by the target retrieval information in multiple default metadata, therefore, the application embodiment of the present invention, improve retrieval and improve recall precision, alleviate the storage pressure of intelligent traffic monitoring system simultaneously.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, the accompanying drawing used required in embodiment or description of the prior art will be briefly described below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the premise not paying creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.

The schematic flow sheet of a kind of data retrieval method that Fig. 1 provides for the embodiment of the present invention；

A kind of schematic diagram setting up hierarchical tree data directory structure that Fig. 2 provides for the embodiment of the present invention；

A kind of concrete hierarchical tree data directory structural representation that Fig. 3 provides for the embodiment of the present invention；

The hierarchical tree data directory structural representation that another kind that Fig. 4 provides for the embodiment of the present invention is concrete；

The schematic diagram of a kind of retrieving that Fig. 5 provides for the embodiment of the present invention；

The structural representation of a kind of data retrieval system that Fig. 6 provides for the embodiment of the present invention；

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, the every other embodiment that those of ordinary skill in the art obtain under not making creative work premise, broadly fall into the scope of protection of the invention.

Embodiments provide a kind of data retrieval method and system, it is thus achieved that as the destination object of retrieval foundation；And obtain target metadata；The target retrieval information corresponding to described destination object is extracted from target metadata；Based on each generic attribute included by described target retrieval information, retrieve in default hierarchical tree data directory structure, it is thus achieved that the retrieval result corresponding to described destination object；The object space attribute in target metadata, object time attribute and target prospect image attributes is included due to target retrieval information, and the hierarchical tree data directory structure preset is each generic attribute foundation included by the target retrieval information in multiple default metadata, therefore, the application embodiment of the present invention, improve retrieval and improve recall precision, alleviate the storage pressure of intelligent traffic monitoring system simultaneously.

Below by specific embodiment, the present invention will be described in detail.The schematic flow sheet of a kind of data retrieval method that Fig. 1 provides for the embodiment of the present invention；Comprise the steps:

S100: obtain the destination object as retrieval foundation；

It will be appreciated that, need when data are retrieved what the content first knowing retrieval is, namely described destination object, it is understandable that, intelligent traffic monitoring system carrys out recording traffic monitor video typically by the photographic head disposed, and being preserved to the memorizer of intelligent traffic monitoring system by the video data recorded, conventional memorizer can be network hard disk video recorder (NetworkVideoRecorder).Certainly, for memorizer, its form has multiple, for instance cloud disk, hard disk etc..Any one image frame of stored monitor video is all can as searched targets object; but generally we are to choose the image frame containing information of vehicles in retrieval; what deserves to be explained is, the concrete form of memorizer and the searched targets object chosen are not done and clearly limit by the embodiment of the present invention.

S200: determine the target metadata of described destination object；

It will be understood by those skilled in the art that, when the video that memorizer preserves in intelligent traffic monitoring system carries out data analysis, usually introduce the concept of metadata (Metadata), metadata is the data (dataaboutotherdata) describing data, or perhaps for providing the structured data (structureddata) for information about of certain resource.Metadata is the data describing the object such as information resources or data, and metadata may identify which resource, evaluates resource, follows the trail of resource change in use；Realize simply and efficiently managing mass data；Realize the effectively discovery of data resource, lookup, integration organization and the effective management to use resource.And metadata one is set up, just can share.Also it is data due to metadata, therefore can carry out storing and obtaining in data base by the method for class likelihood data.User can first look at its metadata so as to obtain oneself required information when using data.

In the embodiment of the present invention, in intelligent traffic monitoring system, each image of stored monitor video comprises many attribute, for instance: a target metadata M can be expressed as:

M={L, T, C, VT, I}

Wherein, the space attribute that L (Location) is metadata, represent the camera collection node location in traffic route network, such as camera collection node 1, camera collection node 2, camera collection node 3 etc., space attribute is exactly geographical position attribute, when image acquisition, different camera collection point acquired images can be different, can find this image location of collection according to space attribute information；T (Time) is time attribute, represents target by the time of camera collection, such as 11:00am, 2:00pm etc.；C (Color) is color attribute, represents the colouring information of foreground image, such as red, yellow, green, black, in vain, blue etc.；VT (VehicleType) is then vehicle vehicle attribute, such as car, SUV, pick up, lorry etc.；I (Image) is target prospect image, is mainly used in extracting characteristics of image and carries out images match.Above each attribute is only the concrete example in embodiment, the attribute that image is not comprised by the embodiment of the present invention does and clearly limits, in actual retrieving, each attribute of image time image to be retrieved is carried out characteristic attribute extraction, can be extracted according to concrete search condition.It is emphasized that in embodiments of the present invention, being extracted the metadata obtained needs to include: the target prospect image attributes of the object space attribute of destination object, the time attribute of destination object and destination object.

S300: extract the target retrieval information corresponding to described destination object from described target metadata, wherein, described target retrieval information includes: object space attribute in described target metadata, object time attribute and target prospect image attributes；

The target metadata of the destination object of acquisition in above-mentioned steps S200 is carried out the extraction of target retrieval information, and then it is follow-up according to target retrieval information and executing search operaqtion, wherein, it is understandable that, the target retrieval information extracted is: the retrieval foundation more refined relative to the destination object as the retrieval foundation in image aspect is exactly the foundation retrieved, and the retrieval information according to extracting is retrieved.Understandable, the target retrieval information extracted from target metadata is also required to include the target prospect image attributes of the object space attribute of destination object, the time attribute of destination object and destination object.Certainly, the target retrieval information extracted can also include some other attribute, such as: the retrieval information after extraction can also include the color attribute of destination object, the vehicle attribute of target vehicle, the car mark attribute etc. of target vehicle, therefore, the attribute that retrieval information is not comprised by the application is defined.

Additionally, it is emphasized that, the embodiment of the present invention only determines the object space attribute in retrieval information, the sorted order of object time attribute and target prospect figure, the i.e. concrete level at place, the order of other attributes for comprising in retrieval information does not do specific requirement, such as: the attribute comprised in target retrieval information has object space attribute, object time attribute, color of object attribute, target vehicle vehicle attribute and target prospect image attributes, order between each attribute can be: object space attribute, object time attribute, color of object attribute, target vehicle vehicle attribute, target prospect image attributes；Order between above-mentioned each attribute can also be: object space attribute, object time attribute, target vehicle vehicle attribute, color of object attribute, target prospect image attributes.

S400: according to each generic attribute included by described target retrieval information, retrieve in default hierarchical tree data directory structure, it is thus achieved that the retrieval result corresponding to described destination object.

Wherein, described default hierarchical tree data directory structure is that each generic attribute included by the target retrieval information in multiple default metadata is set up, the plurality of default metadata is as the metadata in the default metadatabase of default collection of metadata, and, in described default hierarchical tree data directory structure, root node is determined according to the space attribute in the plurality of default metadata, the second layer is determined according to the time attribute in the plurality of default metadata, and leaf node is determined according to the foreground image attribute in the plurality of default metadata.

Known is, when target retrieval information is retrieved, need to establish the individual-layer data index structure corresponding with target retrieval information when metadata memory phase, stage in metadata storage, the feature of each attribute comprised in metadata is extracted, set up the hierarchical tree data directory structure corresponding to different attribute combination and different order respectively, process of setting up such as Fig. 2 of its hierarchical tree data directory structure, it show a kind of schematic diagram setting up hierarchical tree data directory structure that the embodiment of the present invention provides, in the process that hierarchical tree data directory structure is set up, Hadoop is used to carry out distributed nature index, first in the Map stage of Hadoop, feature extraction is carried out for the foreground image attribute in metadata attributes, the space attribute that will comprise in the feature of extraction and metadata, time attribute is mapped, just the Reduce stage is entered afterwards.In the Reduce stage, using the root layer as hierarchical tree data directory structure of the space attribute in metadata, the root node of the corresponding tree of each space attribute；Using corresponding for the root node residing for the time attribute second layer as hierarchical tree data directory structure corresponding foreground image；Using the foreground image leaf node as hierarchical tree data directory structure.

Setting up in process in actual hierarchical tree data directory structure, leaf node is usually set up according to the foreground image attribute in multiple metadata attributes, concrete, determines that the mode of leaf node may include that according to the foreground image in multiple default metadata

Step a1, the foreground image attribute in each default metadata is carried out Visual Feature Retrieval Process, it is thus achieved that the visual vocabulary of foreground image attribute；It is understandable that, foreground image attribute is a characteristic set, it includes various features, feature generally includes the features such as textural characteristics LBP (LocalBinaryPattern), SIFT feature, color histogram, according to the method that the SIFT feature extracted uses KMeans (K-average), these SIFT feature points are clustered, obtain " visual vocabulary ", one section of document can be equal to for a foreground image, but what it had is not general vocabulary, but " visual vocabulary ", these vocabulary just represent a characteristics of image corresponding to metadata.

Step b1, calculate the described visual vocabulary visual vocabulary weight for default metadatabase of each default metadata, and the visual vocabulary weight obtained is ranked up by preset rules, obtain visual vocabulary weight table；

The quantity occurred in collection of document according to each visual vocabulary calculates the importance of this vocabulary, the invention provides a kind of calculation expression, is expressed as:

$w_{i,d}={\mathrm{tf}}_{i,d}\×log\left(\frac{N}{{\mathrm{df}}_{i,d}}\right)$

In formula, tf_i,dFor the visual vocabulary i frequency occurred in current preset metadata, df_i,dThe frequency that the metadata comprising visual vocabulary i occurs in default metadatabase, N presets the number of all metadata in metadatabase.Tf_i,dAnd df_i,dThe main thought of the two parameter is exactly that a word is more many more important to the document in document occurrence number, if but word occurrence number in all documents is a lot, then this word is exactly a common vocabulary in full text shelves, is just less than important.By this formula, it is possible to the foreground image attribute in metadata is converted into visual vocabulary weight, and is undertaken being arranged in a tables of data by preset rules by visual vocabulary weight corresponding for multiple metadata, obtain a visual vocabulary weight table.Such as, visual vocabulary weight corresponding for multiple metadata is arranged in a tables of data according to the mode of descending, obtains a vision weight table.It is noted that the sortord of visual vocabulary weight is not defined by the application.

Step c1, according to described visual vocabulary weight table, set up the leaf node in described default hierarchical tree data directory structure.After step a1 and b1, it is possible to using the visual vocabulary weight table of above-mentioned acquisition as the leaf node being divided into tree shaped data index structure.

What deserves to be explained is, when foundation is divided into tree shaped data index structure can each attribute of combination as much as possible, so can meet different search condition, finally the hierarchical tree data directory structure of foundation is saved in index structure storehouse.

Such as, it is example that the embodiment of the present invention takes the video gathered 24 hours, every TInterval hour is section between a time, such as TInterval=2, 0:00:00-1:59:59am is included under camera collection point (root node) of this correspondence, 2:00:00am-3:59:59am, 4:00:00am-5:59:59am, ..., the 10:00:00pm-11:59:59pm time period, the time attribute of each metadata is only able to find a corresponding time period, hypothetical target foreground image gathers acquisition when 10:32:33am, time attribute in the metadata that then this image is corresponding will be based upon in hierarchical tree data directory structure in the 10:00:00am-11:59:59am time period 6 of the second layer and under corresponding root node, certainly, between the time here, section TInterval can be configured according to the fineness of retrieval, its concrete numerical value is not defined by the application.After the root node establishing hierarchical tree data directory structure and the second layer, can according to the visual signature extracted, set up leaf node, concrete, according to the method that the SIFT feature extracted uses KMeans (K-average), these SIFT feature points are clustered, obtain " visual vocabulary ", one section of document can be equal to for a foreground image, but what it had is not general vocabulary, and be " visual vocabulary ", these vocabulary just represent a characteristics of image corresponding to metadata.Then will calculate the importance of this vocabulary according to vocabulary quantity in collection of document, the invention provides a kind of calculation expression, be expressed as:

$w_{i,d}={\mathrm{tf}}_{i,d}\×log\left(\frac{N}{{\mathrm{df}}_{i,d}}\right)$

In formula, tf_i,dFor the visual vocabulary i frequency occurred in current preset metadata, df_i,dFor the frequency that the metadata comprising visual vocabulary i occurs in default metadatabase, N presets the number of all metadata in metadatabase.Tf_i,dAnd df_i,dThe main thought of the two parameter is exactly that a word is more many more important to the document in document occurrence number, if but word occurrence number in all documents is a lot, then this word is exactly a common vocabulary in full text shelves, is just less than important.By this formula, it is possible to the foreground image attribute in metadata is converted into visual vocabulary weight, and is undertaken being arranged in a tables of data by preset rules by visual vocabulary weight corresponding for multiple metadata, obtain a visual vocabulary weight table.According to described visual vocabulary weight table, set up the leaf node in described default hierarchical tree data directory structure.As it is shown on figure 3, a kind of concrete hierarchical tree data directory structural representation provided for the embodiment of the present invention；Based on the hierarchical tree data directory structure shown in Fig. 3, multiple different order structure or the hierarchical tree data directory structure of the different number of plies can also be set up, such as, third layer can be set up according to the color attribute in metadata under the time attribute path that the second layer is corresponding after two-layer before foundation, the 4th layer can be set up under the path that third layer is corresponding again according to vehicle vehicle attribute, finally resettle leaf node, as shown in Figure 4, the concrete hierarchical tree data directory structural representation of another kind provided for the embodiment of the present invention.

In the process that destination object is retrieved, extract with the above-mentioned feature needing first foreground image to destination object setting up hierarchical tree data directory structure corresponding, obtain the target visual vocabulary of described target prospect image, then proceed to follow-up detection and process.Therefore, concrete, described each generic attribute included by described target retrieval information, retrieve in default hierarchical tree data directory structure, it is thus achieved that the retrieval corresponding to described destination object is as a result, it is possible to include:

Step a2, described target prospect image attributes is carried out Visual Feature Retrieval Process, it is thus achieved that the target visual vocabulary of described target prospect image；

Step b2, calculate the described target visual vocabulary target visual term weight for described default metadatabase；

Step c2, each generic attribute included by described target retrieval information and described target visual term weight, retrieve, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure.

It is understood that, SIFT feature in target prospect image is extracted, and use the SIFT feature point that the method for KMeans (K-average) is extracted to cluster, obtain " visual vocabulary ", one section of document can be equal to for target prospect image, but what it had is not general vocabulary, but for the target visual vocabulary of destination object, these target visual vocabulary just represent the characteristics of image that target metadata is corresponding.

Concrete, the target visual vocabulary of acquisition is carried out weight calculation for metadatabase, can according to formula:

$w_{i,d}={\mathrm{tf}}_{i,d}\×log\left(\frac{N}{{\mathrm{df}}_{i,d}}\right)$

In formula, tf_{I, d}For the target visual vocabulary i frequency occurred in target metadata, df_i,dFor the frequency that the metadata comprising target visual vocabulary i occurs in default metadatabase, N presets the number of all metadata in metadatabase.Tf_i,dAnd df_i,dThe main thought of the two parameter is exactly that a word is more many more important to the document in document occurrence number, if but word occurrence number in all documents is a lot, then this word is exactly a common vocabulary in full text shelves, is just less than important.By this formula, it is possible to the foreground image attribute in destination object metadata is converted into target visual term weight.

Further, the each generic attribute included by described target retrieval information described in step c2 and described target visual term weight, default hierarchical tree data directory structure is retrieved, it is thus achieved that the retrieval corresponding to described destination object is as a result, it is possible to include:

Utilize each generic attribute included by described target retrieval information, each respective layer in described default hierarchical tree data directory structure is successively retrieved, wherein, successively carry out the mode retrieved: in default hierarchical tree data directory structure, determine target root node according to the space attribute in described target retrieval information, retrieve under described target root node further according to the time attribute in described target retrieval information, determine this node layer, retrieve further according under the determined node of last layer corresponding with this attribute in described default hierarchical tree data directory structure of other each attributes in described target retrieval information, finally determine leaf node；

The target visual term weight in described target metadata included by described target retrieval information carries out Similarity Measure under leaf node described in described default hierarchical tree data directory structure, and the Similarity value calculated is arranged in descending order, take the foreground image corresponding to front predetermined number metadata as the retrieval result corresponding to described destination object.Concrete, included by target retrieval information the target visual term weight in target metadata, carries out Similarity Measure under leaf node described in default hierarchical tree data directory structure, passes through below equation:

$score(q,p)=\frac{V_q\×V_p}{\left|V_q\right|\left|V_p\right|}=\frac{{\mathrm{\Σ}}_{i=1}^n{w}_{i,q}\×w_{i,p}}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}_{i,q}^2}*\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}_{i,p}^2}}$

In formula, q is the target metadata of destination object, and p is the metadata in described default hierarchical tree data directory structure, V_qFor the visual vocabulary weight vectors in target retrieval information, V_pFor the visual vocabulary weight vectors of metadata, w in default hierarchical tree data directory structure_i,qFor V_qComponent, w_i,pFor V_pComponent.Determining the metadata similar to destination object, as it is shown in figure 5, the schematic diagram of a kind of retrieving provided for the embodiment of the present invention, wherein the search condition of destination object is: place occur: camera collection point 3；Time of occurrence: 10:32:33am；Color: black, vehicle: the foreground image of car and destination object.Its retrieving is, the appearance place according in search condition: camera collection point 3 can be retrieved by the root node level in corresponding hierarchical tree data directory structure, is not difficult to determine that root node is camera collection point 3, i.e. determine the appearance place of target；Appearance place further according in target retrieval information: 10:32:33am retrieves under the path of camera collection point 3 correspondence, time period residing for the above-mentioned foundation known target time of occurrence of rule is the time period 6, according to this search rule, it is possible to determine easily and meet color: black；Vehicle: all foreground images of car, the foreground image of destination object foreground image after retrieval will carry out Similarity Measure again, obtain Similarity value, and Similarity value is arranged in descending order, take the metadata that front predetermined number Similarity value is corresponding, if predetermined number is 5, then Similarity value is come the foreground image output that the metadata of first 5 is corresponding, the foreground image of output is aiming at the result that the search condition of destination object is retrieved in hierarchical tree data directory structure, what deserves to be explained is, the above-mentioned concrete example for the embodiment of the present invention, this predetermined number is not limited by the application further.

Visible, applying embodiment illustrated in fig. 1 of the present invention, the process of retrieval is that carrying out successively is retrieved in tree, and the range of search of target retrieval condition is in layer reduced, the result that final acquisition is the most similar to target retrieval condition, greatly improves recall precision.

Corresponding with above-mentioned embodiment of the method, the embodiment of the present invention additionally provides a kind of data retrieval system.

The structural representation of a kind of data retrieval system that Fig. 6 provides for the embodiment of the present invention, it is possible to including: destination object obtains module 100, target metadata determines module 200, target retrieval information acquisition module 300, and retrieval result obtains module 400.

Wherein, destination object obtains module 100, for obtaining the destination object as retrieval foundation；

Target metadata determines module 200, for determining the target metadata of described destination object；

In actual applications, target metadata shown in the embodiment of the present invention determines module 200, specifically may be used for: determine that in intelligent traffic monitoring system, each image of stored monitor video comprises many attribute, for instance: a target metadata M can be expressed as:

M={L, T, C, VT, I}

Wherein, the space attribute that L (Location) is metadata, represent the camera collection node location in traffic route network, such as camera collection node 1, camera collection node 2, camera collection node 3 etc., space attribute is exactly geographical position attribute, when image acquisition, different camera collection point acquired images can be different, can find this image location of collection according to space attribute information；T (Time) is time attribute, represents target by the time of camera collection, and such as 11:00am, 2:00pm etc., C (Color) is color attribute, represents the colouring information of foreground image, such as red, yellow, green, black, in vain, blue etc.；VT (VehicleType) is then vehicle vehicle attribute, such as car, SUV, pick up, lorry etc.；I (Image) is target prospect image, is mainly used in extracting characteristics of image and carries out images match.

Target retrieval information acquisition module 300, for extracting the target retrieval information corresponding to described destination object from described target metadata, wherein, described target retrieval information includes: object space attribute in described target metadata, object time attribute and target prospect image attributes；

Retrieval result obtains module 400, for each generic attribute included by described target retrieval information, retrieves, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure.

Wherein, described default hierarchical tree data directory structure is that index structure sets up the module each generic attribute foundation included by the target retrieval information in multiple default metadata, the plurality of default metadata is as the metadata in the default metadatabase of default collection of metadata, and, in described default hierarchical tree data directory structure, root node is determined according to the space attribute in the plurality of default metadata, the second layer is determined according to the time attribute in the plurality of default metadata, leaf node is determined according to the foreground image attribute in the plurality of default metadata.

Index structure sets up module in the process setting up hierarchical tree data directory structure, it is determined that the detailed process of leaf node includes:

Foreground image attribute in each default metadata is carried out Visual Feature Retrieval Process, it is thus achieved that the visual vocabulary of foreground image attribute；

Calculate the described visual vocabulary visual vocabulary weight for default metadatabase of each default metadata, and the visual vocabulary weight obtained is ranked up by preset rules, obtain visual vocabulary weight table；

According to described visual vocabulary weight table, set up the leaf node in described default hierarchical tree data directory structure.

In actual applications, the retrieval result shown in the embodiment of the present invention obtains module 400, specifically can also include:

Target visual vocabulary obtains submodule, for described target prospect image attributes is carried out Visual Feature Retrieval Process, it is thus achieved that the target visual vocabulary of described target prospect image；

In practical application, target visual vocabulary obtains submodule, specifically for: according to the visual signature extracted, set up leaf node, concrete, according to the method that the SIFT feature extracted uses KMeans (K-average), these SIFT feature points are clustered, obtains " visual vocabulary ", target prospect image can be equal to one section of document, but what it had is not general vocabulary, but " visual vocabulary ", these vocabulary just represent the characteristics of image that target metadata is corresponding.

Target visual term weight calculating sub module, for calculating the described target visual vocabulary target visual term weight for described default metadatabase；

In practical application, target visual term weight calculating sub module, specifically for: the importance of this vocabulary will be calculated according to vocabulary quantity in collection of document, the invention provides a kind of calculation expression, be expressed as:

$w_{i,d}={\mathrm{tf}}_{i,d}\×log\left(\frac{N}{{\mathrm{df}}_{i,d}}\right)$

In formula, tf_i,dFor the target visual vocabulary i frequency occurred in current preset metadata, its value this target visual vocabulary of more big explanation is more important；Df_i,dFor the frequency that the metadata comprising target visual vocabulary i occurs in default metadatabase, its value this target visual vocabulary of more big explanation is more inessential, and N presets the number of all metadata in metadatabase.

Retrieval result obtains submodule, for each generic attribute included by described target retrieval information and described target visual term weight, retrieves, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure.

In practical application, utilize each generic attribute included by target retrieval information and described target visual term weight, default hierarchical tree data directory structure is retrieved, wherein, carry out Similarity Measure under leaf node described in default hierarchical tree data directory structure, pass through below equation:

$score(q,p)=\frac{V_q\×V_p}{\left|V_q\right|\left|V_p\right|}=\frac{{\mathrm{\Σ}}_{i=1}^n{w}_{i,q}\×w_{i,p}}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}_{i,q}^2}*\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}_{i,p}^2}}$

In formula, q is the target metadata of destination object, and p is the metadata in described default hierarchical tree data directory structure, V_qFor the visual vocabulary weight vectors in target retrieval information, V_pFor the visual vocabulary weight vectors of metadata, w in default hierarchical tree data directory structure_i,qFor V_qComponent, w_o,pFor V_pComponent.Determine the metadata similar to destination object, it is thus achieved that retrieval result.

Visible, applying embodiment illustrated in fig. 6 of the present invention, the process of retrieval is that carrying out successively is retrieved in tree, and the range of search of target retrieval condition is in layer reduced, the result that final acquisition is the most similar to target retrieval condition, greatly improves recall precision.

It should be noted that, in this article, the relational terms of such as first and second or the like is used merely to separate an entity or operation with another entity or operating space, and not necessarily requires or imply the relation that there is any this reality between these entities or operation or sequentially.And, term " includes ", " comprising " or its any other variant are intended to comprising of nonexcludability, so that include the process of a series of key element, method, article or equipment not only include those key elements, but also include other key elements being not expressly set out, or also include the key element intrinsic for this process, method, article or equipment.When there is no more restriction, statement " including ... " key element limited, it is not excluded that there is also other identical element in including the process of described key element, method, article or equipment.

Each embodiment in this specification all adopts relevant mode to describe, between each embodiment identical similar part mutually referring to, what each embodiment stressed is the difference with other embodiments.Especially for system embodiment, owing to it is substantially similar to embodiment of the method, so what describe is fairly simple, relevant part illustrates referring to the part of embodiment of the method.

The foregoing is only presently preferred embodiments of the present invention, be not intended to limit protection scope of the present invention.All make within the spirit and principles in the present invention any amendment, equivalent replacement, improvement etc., be all contained in protection scope of the present invention.

Claims (10)

1. a data retrieval method, is applied to intelligent monitor system, it is characterised in that described method includes step:

Obtain the destination object as retrieval foundation；

Determine the target metadata of described destination object；

Extracting the target retrieval information corresponding to described destination object from described target metadata, wherein, described target retrieval information includes: object space attribute in described target metadata, object time attribute and target prospect image attributes；

According to each generic attribute included by described target retrieval information, retrieve in default hierarchical tree data directory structure, it is thus achieved that the retrieval result corresponding to described destination object；

Wherein, described default hierarchical tree data directory structure is that each generic attribute included by the target retrieval information in multiple default metadata is set up, the plurality of default metadata is as the metadata in the default metadatabase of default collection of metadata, and, in described default hierarchical tree data directory structure, root node is determined according to the space attribute in the plurality of default metadata, the second layer is determined according to the time attribute in the plurality of default metadata, and leaf node is determined according to the foreground image attribute in the plurality of default metadata.

2. method according to claim 1, it is characterised in that described determine that the mode of leaf node includes according to the foreground image attribute in the plurality of default metadata:

Foreground image attribute in each default metadata is carried out Visual Feature Retrieval Process, it is thus achieved that the visual vocabulary of foreground image attribute；

Calculate the described visual vocabulary visual vocabulary weight for default metadatabase of each default metadata, and the visual vocabulary weight obtained is ranked up by preset rules, obtain visual vocabulary weight table；

According to described visual vocabulary weight table, set up the leaf node in described default hierarchical tree data directory structure.

3. method according to claim 2, it is characterised in that described each generic attribute included by described target retrieval information, retrieves, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure, including:

Described target prospect image attributes is carried out Visual Feature Retrieval Process, it is thus achieved that the target visual vocabulary of described target prospect image；

Calculate the described target visual vocabulary target visual term weight for described default metadatabase；

Each generic attribute included by described target retrieval information and described target visual term weight, retrieve, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure.

4. method according to Claims 2 or 3, it is characterised in that the described visual vocabulary of each default metadata of described calculating passes through equation below for the visual vocabulary weight of default metadatabase:

$w_{i,d}={\mathrm{tf}}_{i,d}\×log\left(\frac{N}{{\mathrm{df}}_{i,d}}\right)$

In formula, tf_i,dFor the visual vocabulary i frequency occurred in current preset metadata, df_i,dFor the frequency that the metadata comprising visual vocabulary i occurs in default metadatabase, N presets the number of all metadata in metadatabase.

5. method according to claim 3, it is characterized in that described each generic attribute included by described target retrieval information and described target visual term weight are retrieved in default hierarchical tree data directory structure, obtain the retrieval result corresponding to described destination object, including:

Utilize each generic attribute included by described target retrieval information, each respective layer in described default hierarchical tree data directory structure is successively retrieved, wherein, successively carry out the mode retrieved: in default hierarchical tree data directory structure, determine target root node according to the space attribute in described target retrieval information, retrieve under described target root node further according to the time attribute in described target retrieval information, determine this node layer, retrieve further according under the determined node of last layer corresponding with this attribute in described default hierarchical tree data directory structure of other each attributes in described target retrieval information, finally determine leaf node；

The target visual term weight in described target metadata included by described target retrieval information carries out Similarity Measure under leaf node described in described default hierarchical tree data directory structure, and the Similarity value calculated is arranged in descending order, take the foreground image corresponding to front predetermined number metadata as the retrieval result corresponding to described destination object.

6. method according to claim 5, it is characterized in that, the target visual term weight in described target metadata included by described target retrieval information carries out Similarity Measure under leaf node described in described default hierarchical tree data directory structure and passes through below equation:

$score(q,p)=\frac{V_q\×V_p}{\left|V_q\right|\left|V_p\right|}=\frac{{\mathrm{\Σ}}_{i=1}^n{w}_{i,q}\×w_{i,p}}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}_{i,q}^2}*\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}_{i,p}^2}}$

In formula, q is the target metadata of destination object, and p is the metadata in described default hierarchical tree data directory structure, V_qFor the visual vocabulary weight vectors in target retrieval information, V_pFor the visual vocabulary weight vectors of metadata, w in default hierarchical tree data directory structure_i,qFor V_qComponent, w_i,pFor V_pComponent.

7. a data retrieval system, is applied to intelligent monitor system, it is characterised in that described system includes:

Destination object obtains module, for obtaining the destination object as retrieval foundation；

Target metadata determines module, for determining the target metadata of described destination object；

Target retrieval information acquisition module, for extracting the target retrieval information corresponding to described destination object from described target metadata, wherein, described target retrieval information includes: object space attribute in described target metadata, object time attribute and target prospect image attributes；

Retrieval result obtains module, for each generic attribute included by described target retrieval information, retrieves, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure；

Wherein, described default hierarchical tree data directory structure is that index structure sets up the module each generic attribute foundation included by the target retrieval information in multiple default metadata, the plurality of default metadata is as the metadata in the default metadatabase of default collection of metadata, and, in described default hierarchical tree data directory structure, root node is determined according to the space attribute in the plurality of default metadata, the second layer is determined according to the time attribute in the plurality of default metadata, leaf node is determined according to the foreground image attribute in the plurality of default metadata.

8. system according to claim 7, it is characterised in that described index structure is set up module and determined that the detailed process of leaf node includes according to the foreground image attribute in the plurality of default metadata:

Foreground image attribute in each default metadata is carried out Visual Feature Retrieval Process, it is thus achieved that the visual vocabulary of foreground image attribute；

Calculate the described visual vocabulary visual vocabulary weight for default metadatabase of each default metadata, and the visual vocabulary weight obtained is ranked up by preset rules, obtain visual vocabulary weight table；

According to described visual vocabulary weight table, set up the leaf node in described default hierarchical tree data directory structure.

9. system according to claim 8, it is characterised in that described retrieval result obtains module and includes:

Target visual vocabulary obtains submodule, for described target prospect image attributes is carried out Visual Feature Retrieval Process, it is thus achieved that the target visual vocabulary of described target prospect image；

Target visual term weight calculating sub module, for calculating the described target visual vocabulary target visual term weight for described default metadatabase；

Retrieval result obtains submodule, for each generic attribute included by described target retrieval information and described target visual term weight, retrieves, it is thus achieved that the retrieval result corresponding to described destination object in default hierarchical tree data directory structure.

10. system according to claim 9, it is characterised in that described retrieval result obtains submodule, including:

Layering retrieval unit, for each generic attribute included by described target retrieval information, each respective layer in described default hierarchical tree data directory structure is successively retrieved, wherein, successively carry out the mode retrieved: in default hierarchical tree data directory structure, determine target root node according to the space attribute in described target retrieval information, retrieve under described target root node further according to the time attribute in described target retrieval information, determine this node layer, retrieve further according under the determined node of last layer corresponding with this attribute in described default hierarchical tree data directory structure of other each attributes in described target retrieval information, finally determine leaf node；

Retrieval result determines unit, for carrying out Similarity Measure under the leaf node described in described default hierarchical tree data directory structure of the target visual term weight in the described target metadata included by described target retrieval information, and the Similarity value calculated is arranged in descending order, take the foreground image corresponding to front predetermined number metadata as the retrieval result corresponding to described destination object.