CN105320710A - Illumination variation resistant vehicle retrieval method and device - Google Patents
Illumination variation resistant vehicle retrieval method and device Download PDFInfo
- Publication number
- CN105320710A CN105320710A CN201410381921.7A CN201410381921A CN105320710A CN 105320710 A CN105320710 A CN 105320710A CN 201410381921 A CN201410381921 A CN 201410381921A CN 105320710 A CN105320710 A CN 105320710A
- Authority
- CN
- China
- Prior art keywords
- image
- vehicle
- checked
- information
- retrieval
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention provides an illumination variation resistant vehicle retrieval method and an illumination variation resistant vehicle retrieval device. The method comprises the following steps: determining vehicle model information of an image to be inquired according to the image to be inquired comprising a vehicle; determining time information about collection of each image in a target database and illumination conditions; selecting a plurality of sample images meeting the time information and the illumination conditions from a vehicle model template library corresponding to the vehicle model information, and generating an inquiry image set through the plurality of selected sample images; acquiring a retrieval result of each sample image in the inquiry image set and all images in the target database; determining vehicles similar to the vehicle in the image to be inquired in the target database according to the retrieval results of all the sample images in the inquiry image set, wherein the vehicle model template library comprises a plurality of sample images at different collection times and under different illumination conditions. According to the method, the problem of sharp drop of the vehicle retrieval performance caused by a larger illumination difference can be solved.
Description
Technical field
The present invention relates to intelligent transport technology, particularly relate to a kind of vehicle retrieval method and device of anti-illumination variation.
Background technology
Along with the fast development of China's economy, the continuous expansion of city size and increasing substantially of vehicle fleet size, the traffic system of China moves towards intelligent just gradually.Traffic Surveillance Video be public business significant data basis, social security stability maintenance, hit break laws and commit crime etc. in have vital effect.Wherein, from a large amount of monitor videos, retrieve target vehicle is a primary demand.But when light differential is larger, as daytime, cloudy day, night, query image just there will be some problems, as reflective, fuzzy, even substantially cannot see vehicle body, in these situations, retrieval performance all can sharply decline.In addition, even if query image quality is better, if but there is the large problem of light differential in a target vehicle image that should be retrieved in database, and two width images are also difficult to match, and this will cause recall rate to decline.In prior art, for lighting issues, effective solution is proposed unspecially.In similar vehicle retrieval, lighting issues is the most common in the industry and very severe problem.
Given this, how to solve the problem that the vehicle retrieval performance brought more greatly due to light differential sharply declines and become the current technical issues that need to address.
Summary of the invention
For defect of the prior art, the invention provides a kind of vehicle retrieval method and device of anti-illumination variation, the problem that the vehicle retrieval performance brought more greatly due to light differential sharply declines can be solved.
First aspect, the invention provides a kind of vehicle retrieval method of anti-illumination variation, comprising:
According to the image to be checked comprising vehicle, determine the vehicle information of described image to be checked;
Determine the temporal information of the collection of each image in target database, illumination condition;
According to described temporal information, illumination condition, from the vehicle template base that described vehicle information is corresponding, choose the multiple sample image meeting described temporal information, illumination condition, by the multiple sample image generated query image collections chosen;
Obtain the result for retrieval of all images in each sample image in described query image set and described target database;
According to the result for retrieval of all sample image in described query image set, determine the similar vehicle with the vehicle in described image to be checked in described target database;
Wherein, described vehicle template base comprises: the sample image under multiple different acquisition time and different illumination conditions under different angles, different scene.
Alternatively, the described result for retrieval according to all sample image in described query image set, determine the step with the similar vehicle of the vehicle in image to be checked in described target database, comprising:
The result for retrieval of all sample image in described query image set is sorted according to similarity size, selects similarity to be greater than the similar vehicle of the image in described target database corresponding to default first threshold as vehicle in image to be checked;
Or,
The result for retrieval of all sample image in described query image set is normalized, and the similarity after normalization is greater than the similar vehicle of the image in described target database corresponding to default Second Threshold as vehicle in image to be checked.
Alternatively, described basis comprises the image to be checked of vehicle, determines the step of the vehicle information of described image to be checked, comprising:
When described image to be checked comprises license plate number, identify the license plate number in described image to be checked, inquire about in vehicle authority database according to described license plate number, determine the vehicle information of described image to be checked;
Or,
Extract the first subimage that image to be checked comprises vehicle;
The vehicle image mated with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle image of mating with the described first subimage vehicle information as described image to be checked;
Or,
Extract the first subimage that image to be checked comprises vehicle;
The vehicle template base of mating with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle template base of mating with the described first subimage vehicle information as described image to be checked;
Wherein, described database comprises multiple vehicle template base of described image affiliated area to be checked.
Alternatively, the image of described target database is the image of the collection in specific region in multiple monitor video device in special time period;
Described temporal information comprises: the earliest time point that in described target database, image is collected, and the time point the latest that image is collected;
Described illumination condition is: described earliest time point is to described Lighting information the latest between time point;
Described vehicle template base is the vehicle template base in described specific region.
Alternatively, the step of the result for retrieval of all images in each sample image in the set of described acquisition described query image and described target database, comprising:
Obtain the visual signature similarity of each image in each sample image and described target database, by the image in this sample image, described target database, described visual signature similarity composition triplet information;
Described result for retrieval comprises: the triplet information of all sample image;
Or,
Described result for retrieval comprises: according to the triplet information of all sample image of visual signature sequencing of similarity.
Second aspect, the invention provides a kind of vehicle retrieval device of anti-illumination variation, comprising:
Vehicle information determination unit, for according to the image to be checked comprising vehicle, determines the vehicle information of described image to be checked;
Condition determining unit, for determining temporal information, the illumination condition of the collection of each image in target database;
Query image set generation unit, for according to described temporal information, illumination condition, the multiple sample image meeting described temporal information, illumination condition are chosen, by the multiple sample image generated query image collections chosen from the vehicle template base that described vehicle information is corresponding;
Result for retrieval acquiring unit, for obtaining the result for retrieval of all images in each sample image in described query image set and described target database;
Similar vehicle determining unit, for the result for retrieval according to all sample image in described query image set, determines the similar vehicle with the vehicle in described image to be checked in described target database;
Wherein, described vehicle template base comprises: the different angles under multiple different acquisition time and different illumination conditions, the sample image under different scene.
Alternatively, described similar vehicle determining unit, specifically for
The result for retrieval of all sample image in described query image set is sorted according to similarity size, selects similarity to be greater than the similar vehicle of the image in described target database corresponding to first threshold as vehicle in image to be checked;
Or,
The result for retrieval of all sample image in described query image set is normalized, and the similarity after normalization is greater than the similar vehicle of the image in described target database corresponding to first threshold as vehicle in image to be checked.
Alternatively, described vehicle information determination unit, specifically for
When described image to be checked comprises license plate number, identify the license plate number in described image to be checked, inquire about in vehicle authority database according to described license plate number, determine the vehicle information of described image to be checked;
Or,
Extract the first subimage that image to be checked comprises vehicle;
The vehicle image mated with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle image of mating with the described first subimage vehicle information as described image to be checked;
Or,
Extract the first subimage that image to be checked comprises vehicle;
The vehicle template base of mating with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle template base of mating with the described first subimage vehicle information as described image to be checked;
Wherein, described database comprises multiple vehicle template base of described image affiliated area to be checked.
Alternatively, the image of described target database is the image of the collection in specific region in multiple monitor video device in special time period;
Described temporal information comprises: the earliest time point that in described target database, image is collected, and the time point the latest that image is collected;
Described illumination condition is: described earliest time point is to described Lighting information the latest between time point;
Described vehicle template base is the vehicle template base in described specific region.
Alternatively, described result for retrieval acquiring unit, specifically for
Obtain the visual signature similarity of each image in each sample image and described target database, by the image in this sample image, described target database, described visual signature similarity composition triplet information;
Described result for retrieval comprises: the triplet information of all sample image;
Or,
Described result for retrieval comprises: according to the triplet information of all sample image of visual signature sequencing of similarity.
As shown from the above technical solution, the vehicle retrieval method of anti-illumination variation of the present invention and device, by determining the vehicle information of described image to be checked, the temporal information meeting target database is chosen from the vehicle template base that described vehicle information is corresponding, multiple sample image of illumination condition, obtain query image set, obtain the result for retrieval of all images in each sample image in described query image set and described target database, according to the result for retrieval of all sample image in described query image set, determine the similar vehicle with the vehicle in described image to be checked in described target database, thus, the problem that the vehicle retrieval performance brought more greatly due to light differential sharply declines can be solved.
Accompanying drawing explanation
The schematic flow sheet of the vehicle retrieval method of the anti-illumination variation that Fig. 1 provides for one embodiment of the invention;
The schematic flow sheet of the method for building up of the vehicle template base that Fig. 2 provides for one embodiment of the invention;
The structural representation of the vehicle retrieval device of the anti-illumination variation that Fig. 3 provides for one embodiment of the invention.
Embodiment
For making the object of the embodiment of the present invention, technical scheme and advantage clearly, below in conjunction with the accompanying drawing in the embodiment of the present invention, clear, complete description is carried out to the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on embodiments of the invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.
Fig. 1 shows the schematic flow sheet of the vehicle retrieval method of the anti-illumination variation that one embodiment of the invention provides, and as shown in Figure 1, the vehicle retrieval method of the anti-illumination variation of the present embodiment is as described below.
101, according to the image to be checked comprising vehicle, the vehicle information of described image to be checked is determined.
Vehicle information in the present embodiment can comprise the model of vehicle, or other information of color or vehicle that vehicle information also can comprise vehicle are as size etc., and the present embodiment does not limit vehicle information.
For example, when described image to be checked comprises license plate number, identify the license plate number in described image to be checked, inquire about in vehicle authority database according to described license plate number, determine the vehicle information of described image to be checked.
In addition, in image to be checked, do not comprise license plate number, or when license plate number can not be identified, can refer to the content of the illustrational acquisition vehicle information of following sub-step 1011 to sub-step 1013.
102, the temporal information of the collection of each image in target database, illumination condition is determined.
In the present embodiment, the image in target database can be the database of multiple images compositions that a certain area video supervising device in section sometime gathers, for public security system needs the database of the content of searching needs in this target database.
That is, the image in target database can be the image of the collection in specific region in multiple monitor video device in special time period;
Described temporal information can comprise: the earliest time point that in described target database, image is collected, and the time point the latest that image is collected; Such as, in target database, the time point of a certain image acquisition is time point the earliest, morning on July 26th, 2014 8:00, the time point of another image acquisition is time point the latest, morning on July 26th, 2014,11:00, then can think that temporal information is 8:00 to 11:00 in morning on July 26th, 2014.
Described illumination condition can: described earliest time point is to described Lighting information the latest between time point.Such as, morning on July 26th, 2014 8:00 to 11:00 be rainy day/information such as fine day.
103, according to described temporal information, illumination condition, from the vehicle template base that described vehicle information is corresponding, the multiple sample image meeting described temporal information, illumination condition are chosen, by the multiple sample image generated query image collections chosen.
For example, if the database of image that the video monitoring apparatus that target database is Yu Quan Road, Haidian District 8:00 to 11:00 in morning on July 26th, 2014 gathers, then the multiple sample image chosen according to temporal information and illumination condition are in step 103 the sample image in the vehicle template base of Yu Quan Road, Haidian District under the Lighting information meeting 8:00 to 11:00 in morning.
Will be understood that, the vehicle template base in the present embodiment can comprise: the sample image of different angles, different scene under multiple different acquisition time and different illumination conditions.And these sample image are the sample image under real scene.
That is, described vehicle template base should ensure illumination condition diversity, namely the sample image under the different illumination conditions that comprises of described vehicle template base should comprise: daytime, night, wherein, comprise daytime: fine day, cloudy day, rainy day, greasy weather etc. contain the sample image in all different light situations as far as possible.
Further, described vehicle template base is the vehicle template base in specific region, and the specific region at this place refers to consistent region corresponding to the region residing for image in target database.
In the present embodiment, this region belonging to vehicle template base corresponding to place's query image set must be that the region corresponding with target database is consistent.
That is, if target database comprises the image in two regions, then query image set can be two set, and the region belonging to sample image in these two set is consistent with the region belonging to image in target database.
104, the result for retrieval of all images in each sample image in described query image set and described target database is obtained.
In a particular application, above-mentioned steps can be: the visual signature similarity obtaining each image in each sample image and described target database, by the image in this sample image, described target database, described visual signature similarity composition triplet information;
Described result for retrieval comprises: the triplet information of all sample image.In other embodiments, this result for retrieval also can comprise: according to the triplet information of all sample image of visual signature sequencing of similarity.The sortord of the triplet information that this result for retrieval exports can be arranged according to user's request, can be carry out sorting according to the size of visual signature similarity, can also be that the sequence formed according to the result for retrieval of each sample image carries out merging rear sequence.
For example, can for sort from high to low according to visual signature similarity according to visual signature sequencing of similarity.
For example, the visual signature of arbitrary image is the feature that can reflect picture material, and the extraction of Image Visual Feature mainly also calculates the characteristic of reflection picture material by computer recognizing.
In the present embodiment, before obtaining the visual signature similarity of image, need the visual signature extracting each sample image respectively, and the visual signature of each image in described target database; Such as, the mode of global characteristics descriptor can be adopted to extract the visual signature of each image in each sample image and described target database, or adopt the mode of local feature description's to extract the visual signature of each image in each sample image and described target database.
Usually, scale invariant feature can be adopted to change (Scale-invariantfeaturetransform, be called for short SIFT), rapid robust feature (Speeded-upRobustFeatures, be called for short SURF), the modes such as character gradient histogram (HistogramsofOrientedGradients is called for short HOG) extract the visual signature of each image in each sample image or described target database.
In the present embodiment, the visual signature extracting arbitrary image can be known technology in the industry, and the present embodiment is not described in detail.
In addition, after the visual signature extracting each image in each sample image and described target database, the mode of Euclidean distance or horse formula distance can be adopted to obtain the visual signature similarity of the visual signature of each image in the visual signature of each sample image and described target database.
105, according to the result for retrieval of all sample image in described query image set, the similar vehicle with the vehicle in described image to be checked in described target database is determined.
In a particular application, by the similar vehicle of acquisition according to visual signature similarity hybrid-sorting, and can export.
Such as, the result for retrieval of all sample image in described query image set can be sorted according to similarity size, select similarity to be greater than the similar vehicle of the image in described target database corresponding to default first threshold as vehicle in image to be checked.
In the implementation that another is possible, above-mentioned steps 105 also can be and is normalized by the result for retrieval of all sample image in described query image set, and the image similarity after normalization be greater than in described target database corresponding to default Second Threshold is as the similar vehicle with the vehicle in image to be checked.For example, normalization can be visual signature similarity in all result for retrieval belonged in the same area divided by the maximal value in visual signature similarities all in this region, obtains the visual signature similarity after normalization, and then compares.
It should be noted that, to result for retrieval normalization in the present embodiment, reason mainly may occur that the visual signature similarity entirety of the result for retrieval inquiring about sample image is on the low side, or partial visual characteristic similarity entirety is higher, for this reason, when hybrid-sorting exports, the overall result on the low side of visual signature similarity all can be ordered into after overall higher result, result may be caused to export not too accurate, Given this, by result for retrieval normalization, and the result for retrieval after normalization can be exported.
In a particular application, can first threshold and Second Threshold being set according to actual needs, for selecting more similar image as the similar vehicle with the vehicle in image to be checked in the present embodiment, therefore needing to set first threshold.Due to the image that image that the visual signature similarity being less than or equal to first threshold is corresponding is similarity difference, the present embodiment is not considered.
The vehicle retrieval method of the anti-illumination variation of the present embodiment, in retrieving images process, consider the impact of illumination condition, the illumination condition of combining target database in retrieving and temporal information can solve the problem that the vehicle retrieval performance brought more greatly due to light differential in prior art sharply declines preferably.
Be simply described as follows in conjunction with above-mentioned Fig. 1, the first step: identify that the vehicle obtaining target vehicle is S, so access vehicle template base corresponding to vehicle S.
Wherein, for example, camera A captures the image P of a suspected vehicles at night, by other information, public security department knows that this suspicion car is at certain day time period, as 8:00-9:00 in the morning, the region that camera B monitors may be have passed, this suspected vehicles is retrieved in the video flowing that present needs gather at camera B, because image P gathers under night scenes, huge with the vehicle image vision mode difference that daytime, scene gathered, directly utilize P to retrieve, effect is very bad.
For this reason, select to meet the acquisition time of camera B and the sample image of illumination condition from the vehicle template base that vehicle S is corresponding, add query image set Q.
Further, if this time period (8:00-9:00) is the greasy weather, can the priority of the sample in scene lower greasy weather on daytime be put into the highest.Wherein, choose meet described temporal information, the standard of multiple sample image of illumination condition determines by actual demand, be specially described temporal information, the illumination condition of target database, the sample of selection should as much as possible close to described temporal information, the illumination condition of target database.For example, need to retrieve from the video flowing of 10:00-11:00 collection on daytime, weather condition is fine day (reflective problem), when so selecting sample image, should select the sample image in sunny weather situation.And for example, need to retrieve the video flowing gathered within the time period of 20:00-21:00 at night, the sample image at night so should be selected to add query image set.In the present embodiment, the process of choosing of sample image is not limited, and the quantity of sample image is not limited, correspond to actual needs.
Then, obtain the visual signature similarity of all images in target database corresponding to each sample image and camera B in query image set Q, the visual signature similarity according to obtaining sorts, and exports result for retrieval.
In a particular application, the step 101 in the method shown in earlier figures 1 can comprise not shown step 1011 to step 1013:
1011, the first subimage that image to be checked comprises vehicle is extracted.
1012, in the database set up in advance, search the vehicle image mated with described first subimage.
In the present embodiment, by calculating the First look similarity of all sample image of all vehicle template base in the first subimage and database, using sample image corresponding for First look characteristic similarity maximal value as the vehicle image mated with the first subimage.
The present embodiment is only and illustrates, in a particular application, also searches the vehicle image mated with the first subimage in a database by alternate manner.
Database in this step can be the multiple vehicle template base comprising image affiliated area to be checked.
In this step, the visual signature similarity of described first subimage and all images in the database to set up in advance can be obtained; Determine whether some or multiple visual signature similarity is greater than predetermined threshold value, if be greater than predetermined threshold value, the images match in the first subimage and database can be determined.If all visual signature similarities are all not more than predetermined threshold value, the image mated with the first subimage in database can be thought.
The visual signature extracting the first subimage is respectively needed before obtaining visual signature similarity, and the visual signature of each image in the database set up in advance; In actual applications, the mode of global characteristics descriptor can be adopted to extract the visual signature of each image in the first subimage and the database set up in advance, or adopt the mode of local feature description's to extract the visual signature of each image in the first subimage and the database set up in advance.
1013, using the vehicle information of vehicle image of mating with the described first subimage vehicle information as described image to be checked.
In the implementation that another kind is possible, the step 101 in the method shown in earlier figures 1 can comprise not shown step 1011 " to step 1013 ":
1011 " the first subimage that image to be checked comprises vehicle, is extracted.
1012 " in the database set up in advance, search the vehicle template base of mating with described first subimage.
Data in this step can be the multiple vehicle template base comprising image affiliated area to be checked.
In this step, the visual signature similarity of described first subimage and all vehicle template base in the database to set up in advance can be obtained; Determine whether some or multiple visual signature similarity is greater than another threshold value default, if be greater than another threshold value default, can determine that the first subimage mates with the vehicle template base in database.If all visual signature similarities are all not more than another threshold value default, the vehicle template base of mating with the first subimage in database can be thought.
For example, adopt feature interpretation submode to obtain the First look characteristic similarity of all sample image in described first subimage and described database in all vehicle template base, be called for short the first similarity, obtain the first similarity set; Mathematical statistics analysis is carried out to the first similarity set of all sample image corresponding with each vehicle template base, obtains the visual signature similarity of described first subimage and each vehicle template base.
Such as, can using the similarity of the mean value of the visual signature similarity of all sample image in described vehicle template base as described first subimage and described vehicle template base; Or, using the similarity of the maximal value in the visual signature similarity of all sample image in described vehicle template base as described first subimage and described vehicle template base; Or, using the similarity of the minimum value in the visual signature similarity of all sample image in described vehicle template base as described first subimage and described vehicle template base.Also possibly, employing foreign peoples sample analysis mode removes the isolated point in the visual signature similarity of all sample image corresponding to each vehicle template base, obtain the mean value removing the visual signature similarity outside isolated point in all sample image corresponding with each vehicle template base, using the visual signature similarity of this mean value as the first subimage and each vehicle template base.
1013 ", using the vehicle information of vehicle template base of mating with the described first subimage vehicle information as described image to be checked.
Will be understood that, the visual signature of arbitrary image is the feature that can reflect picture material, current, and the extraction of Image Visual Feature mainly also calculates the characteristic of reflection picture material by computer recognizing.
The vehicle information of image to be checked can be obtained by the mode of above-mentioned citing preferably, and then good query expansion can be realized, and recall rate and the accuracy rate of vehicle image inquiry can be realized preferably.
Fig. 2 shows the schematic flow sheet of the method for building up of the vehicle template base that one embodiment of the invention provides, and as shown in Figure 2, the method for building up of the vehicle template base of the present embodiment is as described below.
201, obtain multiple vehicle image, there is in each vehicle image the license plate number that can identify.
For example, by obtaining multiple vehicle image in the video monitoring apparatus of vehicle, or, obtain multiple vehicle image by image collecting device.Those vehicle images can be sample image.
It should be noted that each vehicle image in the multiple vehicle images obtained in the present embodiment can comprise license plate number, namely in each vehicle image, there is the license plate number that can identify.
License plate number is comprised for vehicle image be mainly used for conveniently obtaining information of vehicles.Current, only have and can obtain information of vehicles by the mode of license plate number, ensure the accuracy of information of vehicles.
202, identify the license plate number in described vehicle image, and according to the license plate number of described vehicle image, from the database preset, obtain the information of vehicles corresponding with described license plate number, described information of vehicles can comprise: vehicle information.
For example, identify that license plate number can be current known technology in the industry, such as, adopt license plate recognition technology to identify license plate number from vehicle image.The license plate number in known technology identification vehicle image is adopted to be mainly used for obtaining information of vehicles corresponding to license plate number in the present embodiment.
Certainly, in actual applications, the information of vehicles of the present embodiment also can comprise: vehicle part information (as number of element types), vehicle color (as white, black), as described in purchase date etc. of vehicle, the present embodiment is only illustrated information of vehicles, does not limit the other guide that information of vehicles comprises.In addition, it should be noted that, information of vehicles described here can be the information of vehicles of vehicle authority inside.
Aforesaid default database can be the database of vehicle authority known in the industry.
For example, the database of vehicle authority comprises following information: vehicle is the information of the vehicles such as BMW X6 automobile, Audi Q7 automobile, popular v6 automobile, and what color vehicle is, is black, white or silver color, purchase date of vehicle owner's vehicle etc.
203, described information of vehicles and described vehicle image are generated candidate's vehicle template base of described vehicle information.
That is, to the vehicle image S in a width monitor video, the license plate number in vehicle image S is identified, obtain the license plate number P of vehicle image S; Obtained the information of vehicles of vehicle image S by license plate number P, license plate number P is retrieved to the vehicle T of vehicle corresponding with it, the vehicle T of vehicle image S and vehicle is joined candidate's vehicle template base TDS of vehicle T.
It should be noted that, the present embodiment can repeat aforesaid step 201 to step 203, obtain the multiple vehicle image under multiple angles of vehicle T, different light, different scene and sample image, and then these sample image are all joined candidate's vehicle template base TDS of vehicle T.
In addition, after the candidate's vehicle template base determining vehicle T, image collecting device also can be adopted to gather multiple image to this vehicle T, and those images can comprise license plate number, also can not comprise license plate number etc., multiple images of collection all belong to the image in candidate's vehicle template base of vehicle T.
It should be noted that, due to the distribution in region, candidate's vehicle template base of the vehicle T in each region can be different, and such as, candidate's vehicle template base of candidate's vehicle template base of the vehicle T of Beijing Area, candidate's vehicle template base of the vehicle T in region, Tianjin, the vehicle T in region, Nanjing can not be identical.Respective candidate's vehicle template base can be set up for zones of different in the present embodiment, facilitate the management of subsequent vehicle management organization.
Certainly, in actual applications, vehicle T also can set up candidate's vehicle template base, and this candidate's vehicle template base can comprise sub-candidate's vehicle template base of zones of different, and the present embodiment is only and illustrates, does not limit it.
204, according to pre-conditioned screening described candidate's vehicle template base, the vehicle template base of described vehicle information is obtained.
Under normal circumstances, the image-erasing that can will repeat in candidate's vehicle template base, ensures that the scene/illumination/attribute of each sample image in vehicle template base is all unique.
In actual applications, screening can be artificial screening and also can be automatic screening, preferably realize automatic screening, because the data (comprising image) in the vehicle template base of each vehicle information have a upper thousand sheets, artificial screening causes wasting time and energy, the image that the mode automatic screening by visual signature comparison repeats.
In the present embodiment, in candidate's vehicle template base, the quantity of image can be more than or equal to the quantity of image in the last vehicle template base obtained.
The diversity of the data in vehicle template base should be ensured during screening, namely comprise different angles, different scale, different colours, different coverage extent, difference block angle, different light, different weather situation etc. contain as far as possible the image under representational different situations.
In the present embodiment, by the screening to candidate's vehicle template base, good vehicle template base can be set up, contain the image of various condition, facilitate the management of vehicle authority.
In addition, also multiple vehicle sample image can be obtained from the video monitoring apparatus of road, if described vehicle image comprises: background area and vehicle viewing area.Background area in the present embodiment is the region that when using vehicle image, user does not pay close attention to.Thus, the vehicle sample image in the vehicle template base finally obtained can not comprise the background area of vehicle image.
The structural representation of the vehicle retrieval device of the anti-illumination variation that Fig. 3 provides for another embodiment of the present invention, as shown in Figure 3, the vehicle retrieval device of the anti-illumination variation of the present embodiment, comprising: vehicle information determination unit 31, condition determining unit 32, query image set generation unit 33, result for retrieval acquiring unit 34, similar vehicle determining unit 35;
Vehicle information determination unit 31, for according to the image to be checked comprising vehicle, determines the vehicle information of described image to be checked;
Condition determining unit 32, for determining temporal information, the illumination condition of the collection of each image in target database;
Query image set generation unit 33, for according to described temporal information, illumination condition, the multiple sample image meeting described temporal information, illumination condition are chosen, by the multiple sample image generated query image collections chosen from the vehicle template base that described vehicle information is corresponding;
Result for retrieval acquiring unit 34, for obtaining the result for retrieval of all images in each sample image in described query image set and described target database;
Similar vehicle determining unit 35, for the result for retrieval according to all sample image in described query image set, determines the similar vehicle with the vehicle in described image to be checked in described target database;
Wherein, described vehicle template base comprises: the sample image under multiple different acquisition time and different illumination conditions under different angles, different scene.
In addition, in a particular application, aforesaid device also can comprise not shown result output unit, and this result output unit can be used for the similar vehicle of acquisition to export from high to low according to similarity.
In a kind of possible implementation, described similar vehicle determining unit 35, can, specifically for being sorted according to similarity size by the result for retrieval of all sample image in described query image set, similarity be selected to be greater than the similar vehicle of the image in described target database corresponding to first threshold as vehicle in image to be checked;
In the implementation that another kind is possible, described similar vehicle determining unit 35, also specifically for being normalized by the result for retrieval of all sample image in described query image set, and the similarity after normalization can be greater than the similar vehicle of the image in described target database corresponding to first threshold as vehicle in image to be checked.
In a particular application, described vehicle information determination unit 31, can specifically for when described image to be checked comprises license plate number, identify the license plate number in described image to be checked, inquire about in vehicle authority database according to described license plate number, determine the vehicle information of described image to be checked;
Or described vehicle information determination unit 31, also can comprise the first subimage of vehicle specifically for extracting image to be checked;
The vehicle image mated with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle image of mating with the described first subimage vehicle information as described image to be checked;
In possible implementation, described vehicle information determination unit 31, also can comprise the first subimage of vehicle specifically for extracting image to be checked;
The vehicle template base of mating with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle template base of mating with the described first subimage vehicle information as described image to be checked;
Wherein, above-mentioned database can comprise multiple vehicle template base of described image affiliated area to be checked.
In addition, aforesaid result for retrieval acquiring unit 34, specifically for obtaining the visual signature similarity of each image in each sample image and described target database, the image in this sample image, described target database, described visual signature similarity can be formed triplet information;
Described result for retrieval comprises: the triplet information of all sample image; Or described result for retrieval comprises: according to the triplet information of all sample image of visual signature sequencing of similarity.
Further, in the present embodiment, the image of described target database can be the image of the collection in specific region in multiple monitor video device in special time period;
Described temporal information can comprise: the earliest time point that in described target database, image is collected, and the time point the latest that image is collected;
Described illumination condition can be: described earliest time point is to described Lighting information the latest between time point;
Described vehicle template base can be the vehicle template base in described specific region.
The vehicle retrieval device of the anti-illumination variation of the present embodiment, can solve the problem that the vehicle retrieval performance brought more greatly due to light differential sharply declines.
The vehicle retrieval device of the anti-illumination variation of the present embodiment, may be used for the technical scheme performing embodiment of the method shown in earlier figures 1, it realizes principle and technique effect is similar, repeats no more herein.
Last it is noted that above each embodiment is only in order to illustrate technical scheme of the present invention, be not intended to limit; Although with reference to foregoing embodiments to invention has been detailed description, those of ordinary skill in the art is to be understood that: it still can be modified to the technical scheme described in foregoing embodiments, or carries out equivalent replacement to wherein some or all of technical characteristic; And these amendments or replacement, do not make the essence of appropriate technical solution depart from the scope of the claims in the present invention.
Claims (10)
1. a vehicle retrieval method for anti-illumination variation, is characterized in that, comprising:
According to the image to be checked comprising vehicle, determine the vehicle information of described image to be checked;
Determine the temporal information of the collection of each image in target database, illumination condition;
According to described temporal information, illumination condition, from the vehicle template base that described vehicle information is corresponding, choose the multiple sample image meeting described temporal information, illumination condition, by the multiple sample image generated query image collections chosen;
Obtain the result for retrieval of all images in each sample image in described query image set and described target database;
According to the result for retrieval of all sample image in described query image set, determine the similar vehicle with the vehicle in described image to be checked in described target database;
Wherein, described vehicle template base comprises: the sample image under multiple different acquisition time and different illumination conditions under different angles, different scene.
2. method according to claim 1, is characterized in that, the described result for retrieval according to all sample image in described query image set, determines the step with the similar vehicle of the vehicle in image to be checked in described target database, comprising:
The result for retrieval of all sample image in described query image set is sorted according to similarity size, selects similarity to be greater than the similar vehicle of the image in described target database corresponding to default first threshold as vehicle in image to be checked;
Or,
The result for retrieval of all sample image in described query image set is normalized, and the similarity after normalization is greater than the similar vehicle of the image in described target database corresponding to default Second Threshold as vehicle in image to be checked.
3. method according to claim 1, is characterized in that, described basis comprises the image to be checked of vehicle, determines the step of the vehicle information of described image to be checked, comprising:
When described image to be checked comprises license plate number, identify the license plate number in described image to be checked, inquire about in vehicle authority database according to described license plate number, determine the vehicle information of described image to be checked;
Or,
Extract the first subimage that image to be checked comprises vehicle;
The vehicle image mated with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle image of mating with the described first subimage vehicle information as described image to be checked;
Or,
Extract the first subimage that image to be checked comprises vehicle;
The vehicle template base of mating with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle template base of mating with the described first subimage vehicle information as described image to be checked;
Wherein, described database comprises multiple vehicle template base of described image affiliated area to be checked.
4. method according to claim 1, is characterized in that, the image of described target database is the image of the collection in specific region in multiple monitor video device in special time period;
Described temporal information comprises: the earliest time point that in described target database, image is collected, and the time point the latest that image is collected;
Described illumination condition is: described earliest time point is to described Lighting information the latest between time point;
Described vehicle template base is the vehicle template base in described specific region.
5. method according to claim 1, is characterized in that, the step of the result for retrieval of all images in each sample image in the set of described acquisition described query image and described target database, comprising:
Obtain the visual signature similarity of each image in each sample image and described target database, by the image in this sample image, described target database, described visual signature similarity composition triplet information;
Described result for retrieval comprises: the triplet information of all sample image;
Or,
Described result for retrieval comprises: according to the triplet information of all sample image of visual signature sequencing of similarity.
6. a vehicle retrieval device for anti-illumination variation, is characterized in that, comprising:
Vehicle information determination unit, for according to the image to be checked comprising vehicle, determines the vehicle information of described image to be checked;
Condition determining unit, for determining temporal information, the illumination condition of the collection of each image in target database;
Query image set generation unit, for according to described temporal information, illumination condition, the multiple sample image meeting described temporal information, illumination condition are chosen, by the multiple sample image generated query image collections chosen from the vehicle template base that described vehicle information is corresponding;
Result for retrieval acquiring unit, for obtaining the result for retrieval of all images in each sample image in described query image set and described target database;
Similar vehicle determining unit, for the result for retrieval according to all sample image in described query image set, determines the similar vehicle with the vehicle in described image to be checked in described target database;
Wherein, described vehicle template base comprises: the different angles under multiple different acquisition time and different illumination conditions, the sample image under different scene.
7. device according to claim 6, is characterized in that, described similar vehicle determining unit, specifically for
The result for retrieval of all sample image in described query image set is sorted according to similarity size, selects similarity to be greater than the similar vehicle of the image in described target database corresponding to default first threshold as vehicle in image to be checked;
Or,
The result for retrieval of all sample image in described query image set is normalized, and the similarity after normalization is greater than the similar vehicle of the image in described target database corresponding to default Second Threshold as vehicle in image to be checked.
8. device according to claim 6, is characterized in that, described vehicle information determination unit, specifically for
When described image to be checked comprises license plate number, identify the license plate number in described image to be checked, inquire about in vehicle authority database according to described license plate number, determine the vehicle information of described image to be checked;
Or,
Extract the first subimage that image to be checked comprises vehicle;
The vehicle image mated with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle image of mating with the described first subimage vehicle information as described image to be checked;
Or,
Extract the first subimage that image to be checked comprises vehicle;
The vehicle template base of mating with described first subimage is searched in the database set up in advance;
Using the vehicle information of vehicle template base of mating with the described first subimage vehicle information as described image to be checked;
Wherein, described database comprises multiple vehicle template base of described image affiliated area to be checked.
9. device according to claim 6, is characterized in that, the image of described target database is the image of the collection in specific region in multiple monitor video device in special time period;
Described temporal information comprises: the earliest time point that in described target database, image is collected, and the time point the latest that image is collected;
Described illumination condition is: described earliest time point is to described Lighting information the latest between time point;
Described vehicle template base is the vehicle template base in described specific region.
10. device according to claim 6, is characterized in that, described result for retrieval acquiring unit, specifically for
Obtain the visual signature similarity of each image in each sample image and described target database, by the image in this sample image, described target database, described visual signature similarity composition triplet information;
Described result for retrieval comprises: the triplet information of all sample image;
Or,
Described result for retrieval comprises: according to the triplet information of all sample image of visual signature sequencing of similarity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410381921.7A CN105320710B (en) | 2014-08-05 | 2014-08-05 | The vehicle retrieval method and device of resisting illumination variation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410381921.7A CN105320710B (en) | 2014-08-05 | 2014-08-05 | The vehicle retrieval method and device of resisting illumination variation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105320710A true CN105320710A (en) | 2016-02-10 |
| CN105320710B CN105320710B (en) | 2018-11-27 |
Family
ID=55248110
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410381921.7A Active CN105320710B (en) | 2014-08-05 | 2014-08-05 | The vehicle retrieval method and device of resisting illumination variation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105320710B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106055704A (en) * | 2016-06-22 | 2016-10-26 | 重庆中科云丛科技有限公司 | Image retrieving and matching method and system |
| CN108109164A (en) * | 2017-12-08 | 2018-06-01 | 联想(北京)有限公司 | A kind of information processing method and electronic equipment |
| CN110569794A (en) * | 2019-09-06 | 2019-12-13 | 深圳市梵高控股有限公司 | article information storage method and device and computer readable storage medium |
| CN111814690A (en) * | 2020-07-09 | 2020-10-23 | 浙江大华技术股份有限公司 | Target re-identification method and device and computer readable storage medium |
| CN112541096A (en) * | 2020-07-27 | 2021-03-23 | 广元量知汇科技有限公司 | Video monitoring method for smart city |
| CN113609319A (en) * | 2021-07-26 | 2021-11-05 | 阿里巴巴（中国）有限公司 | Commodity searching method, device and equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1365077A (en) * | 2001-01-11 | 2002-08-21 | 中华人民共和国深圳出入境边防检查总站 | Multi-media processing method for vehicle immigration data |
| US20100293181A1 (en) * | 2009-05-18 | 2010-11-18 | Autoonline Gmbh Informationssysteme | VALUEpilot - METHOD AND APPARATUS FOR ESTIMATING A VALUE OF A VEHICLE |
| CN103116988A (en) * | 2013-01-18 | 2013-05-22 | 合肥工业大学 | Traffic flow and vehicle type detecting method based on TOF (time of flight) camera |
| CN103310199A (en) * | 2013-06-17 | 2013-09-18 | 武汉大学 | Vehicle model identification method based on high-resolution remote sensing data |
-
2014
- 2014-08-05 CN CN201410381921.7A patent/CN105320710B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1365077A (en) * | 2001-01-11 | 2002-08-21 | 中华人民共和国深圳出入境边防检查总站 | Multi-media processing method for vehicle immigration data |
| US20100293181A1 (en) * | 2009-05-18 | 2010-11-18 | Autoonline Gmbh Informationssysteme | VALUEpilot - METHOD AND APPARATUS FOR ESTIMATING A VALUE OF A VEHICLE |
| CN103116988A (en) * | 2013-01-18 | 2013-05-22 | 合肥工业大学 | Traffic flow and vehicle type detecting method based on TOF (time of flight) camera |
| CN103310199A (en) * | 2013-06-17 | 2013-09-18 | 武汉大学 | Vehicle model identification method based on high-resolution remote sensing data |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106055704A (en) * | 2016-06-22 | 2016-10-26 | 重庆中科云丛科技有限公司 | Image retrieving and matching method and system |
| CN106055704B (en) * | 2016-06-22 | 2020-02-04 | 重庆中科云丛科技有限公司 | Image retrieval and matching method and system |
| CN108109164A (en) * | 2017-12-08 | 2018-06-01 | 联想(北京)有限公司 | A kind of information processing method and electronic equipment |
| CN110569794A (en) * | 2019-09-06 | 2019-12-13 | 深圳市梵高控股有限公司 | article information storage method and device and computer readable storage medium |
| CN110569794B (en) * | 2019-09-06 | 2022-07-19 | 深圳市恒天伟焱科技股份有限公司 | Article information storage method and device and computer readable storage medium |
| CN111814690A (en) * | 2020-07-09 | 2020-10-23 | 浙江大华技术股份有限公司 | Target re-identification method and device and computer readable storage medium |
| CN111814690B (en) * | 2020-07-09 | 2023-09-01 | 浙江大华技术股份有限公司 | Target re-identification method, device and computer readable storage medium |
| CN112541096A (en) * | 2020-07-27 | 2021-03-23 | 广元量知汇科技有限公司 | Video monitoring method for smart city |
| CN113609319A (en) * | 2021-07-26 | 2021-11-05 | 阿里巴巴（中国）有限公司 | Commodity searching method, device and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105320710B (en) | 2018-11-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105320705A (en) | Retrieval method and device for similar vehicle | |
| CN105320710A (en) | Illumination variation resistant vehicle retrieval method and device | |
| Liu et al. | Deep relative distance learning: Tell the difference between similar vehicles | |
| CN102207966B (en) | Video content quick retrieving method based on object tag | |
| Wang et al. | A survey of vehicle re-identification based on deep learning | |
| CN105320923A (en) | Vehicle type recognition method and apparatus | |
| Morishita et al. | Sakurasensor: quasi-realtime cherry-lined roads detection through participatory video sensing by cars | |
| CN104303193A (en) | Clustering-based object classification | |
| CN107103314A (en) | A kind of fake license plate vehicle retrieval system based on machine vision | |
| GB2493580A (en) | Method of searching for a target within video data | |
| CN108830254B (en) | Fine-grained vehicle type detection and identification method based on data balance strategy and intensive attention network | |
| CN103336957A (en) | Network coderivative video detection method based on spatial-temporal characteristics | |
| CN110163109B (en) | Lane line marking method and device | |
| CN104133875A (en) | Face-based video labeling method and face-based video retrieving method | |
| CN104537360A (en) | Method and system for detecting vehicle violation of not giving way | |
| CN105138525A (en) | Traffic video processing device and method, and retrieval device and method | |
| CN105320703A (en) | Query expansion method and device in retrieval of similar vehicle | |
| CN102103609A (en) | Information retrieval method and system | |
| CN105760844A (en) | Video stream data processing method, apparatus and system | |
| CN102098449A (en) | Method for realizing automatic inside segmentation of TV programs by utilizing mark detection | |
| CN105554456A (en) | Video processing method and apparatus | |
| CN102244769A (en) | Object and key person monitoring system and method thereof | |
| Zaman et al. | A robust deep networks based multi-object multi-camera tracking system for city scale traffic | |
| CN113221760A (en) | Expressway motorcycle detection method | |
| CN105320704A (en) | Cross-regional similar vehicle retrieval method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |