CN111428063A - Image feature association processing method and system based on geographic spatial position division - Google Patents
Image feature association processing method and system based on geographic spatial position division Download PDFInfo
- Publication number
- CN111428063A CN111428063A CN202010243925.4A CN202010243925A CN111428063A CN 111428063 A CN111428063 A CN 111428063A CN 202010243925 A CN202010243925 A CN 202010243925A CN 111428063 A CN111428063 A CN 111428063A
- Authority
- CN
- China
- Prior art keywords
- vector
- sample
- image
- map table
- hash map
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5846—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using extracted text
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/587—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using geographical or spatial information, e.g. location
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- Software Systems (AREA)
- Processing Or Creating Images (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides an image feature association processing method and system based on geographic spatial position division, wherein the method comprises the following steps: establishing a first hash map table according to the geographic position of each video stream camera, extracting image text characteristics and sample vector characteristics of a target image sample, storing the image text characteristics and preset associated zone bits into a preset database, and storing the sample vector characteristics and the associated zone bits into a vector characteristic list; acquiring vector characteristics and the belonged geographic position of a target image to be searched, and retrieving the first hash map table according to the belonged geographic position to acquire a corresponding vector characteristic list; and calculating the similarity between each sample vector feature in the vector feature list and the vector feature, determining K sample vector features with the highest similarity and associated flag bits thereof, and acquiring K image text features. The technical scheme provided by the invention can solve the problems of low speed and low precision of the existing image information storage and retrieval method.
Description
Technical Field
The invention relates to the technical field of image feature processing, in particular to an image feature association processing method and system based on geographic spatial position division.
Background
With the development of AI, image intelligent processing technology has been applied in various fields, in the field of image intelligent analysis technology, there are two general ways for storing image information, one is image text label storage, the other is image visual feature storage, and the image text label storage, that is, text information of an object in an image (such as a picture of a pedestrian, text information is gender, jacket color, and the like) is obtained and stored in a corresponding database, and the text information is obtained by database information matching in the later period, so as to realize retrieval of the target image.
The image visual feature storage is that feature vectors (divided into integer and floating point data, and having low-dimensional vectors such as 128-dimensional feature vectors and high-dimensional feature vectors such as 2048-dimensional feature vectors) of a target image are obtained by a feature extraction mode and stored in a database, and image retrieval is realized by a feature vector matching mode at the later stage.
However, the image text label is only used for storage, although the storage speed is high, the later retrieval effect is poor, and the desired matching picture cannot be searched; only the image visual features are used for storage, and for some matching targets (such as pedestrian targets), the interference targets of similar vector features are more aiming at different targets, so that the accuracy of searching the image by the image is seriously influenced.
In addition, the storage and retrieval of image text labels typically uses relational or non-relational databases; the image visual characteristics can be stored by using a relational or non-relational database, but the search engine of the databases has very low retrieval efficiency under the existing large-scale visual characteristic engine, and the work efficiency is seriously influenced.
In addition, in the existing image visual feature storage process, all vector features acquired at an earlier stage are usually stored in a database (a memory or a hard disk), and along with the accumulation of time, the data volume in the database is larger and larger, so that the data matching efficiency and precision during later retrieval are seriously influenced, and the data processing speed of the whole system is further seriously increased.
Therefore, in view of several problems, a need exists for an efficient and highly accurate method for storing and retrieving image information.
Disclosure of Invention
The invention provides an image feature association processing method and system based on geographic spatial position division, and mainly aims to solve the problems of low speed and low precision of the existing image information storage and retrieval method.
In order to achieve the above object, the present invention provides an image feature association processing method based on geospatial location division, including the following steps:
establishing a first hash map table in a system memory according to the geographic position of each video stream camera, wherein a vector feature list corresponding to each geographic position is arranged in the first hash map table;
extracting image text characteristics and sample vector characteristics of a target image sample, storing the image text characteristics and a preset associated zone bit into a preset database, and storing the sample vector characteristics and the associated zone bit into a vector characteristic list; wherein the image text features and the sample vector features are associated by the associated flag bits;
acquiring vector characteristics and a belonged geographic position of a target image to be searched, and retrieving the first hash map table according to the belonged geographic position to acquire a vector characteristic list corresponding to the belonged geographic position;
calculating the similarity between the vector features of each sample in the vector feature list and the vector features of the target image to be searched, and determining K sample vector features with the highest similarity and associated flag bits thereof;
and searching a preset database according to the determined K associated zone bits with the highest similarity so as to obtain corresponding K image text characteristics.
In another aspect, the present invention further provides an image feature association processing system based on geospatial location division, where the image feature association processing system based on geospatial location division includes:
the table building unit is used for building a first hash map table in a system memory according to the geographic position of each video stream camera, wherein a vector feature list corresponding to each geographic position is arranged in the first hash map table;
the data storage unit is used for extracting image text characteristics and sample vector characteristics of a target image sample, storing the image text characteristics and a preset associated zone bit into a preset database, and storing the sample vector characteristics and the associated zone bit into the vector characteristic list; wherein the image text features and the vector features are associated through the associated flag bits;
the vector feature list acquisition unit is used for acquiring the vector features of the target image to be searched and the belonged geographic position, and retrieving the first hash map table according to the belonged geographic position to acquire a vector feature list corresponding to the belonged geographic position;
the similarity matching unit is used for calculating the similarity between each sample vector feature in the vector feature list and the vector feature of the target image to be searched and determining K sample vector features with the highest similarity and associated flag bits thereof;
and the retrieval unit is used for retrieving the preset database according to the K associated zone bits so as to obtain corresponding K image text characteristics.
According to the image feature association processing method and system based on geographic space position division, the geographic positions of the image sources (video stream cameras) are divided, and the corresponding vector feature lists are established, so that the classification and storage of original data can be realized, the later retrieval speed is improved, and the retrieval precision is obviously improved in a mode of predetermining the geographic positions of the images to be searched; in addition, the image text features and the vector features of the original data are associated and stored in different storage positions, so that the simultaneous matching of the image text features and the vector features can be realized, the image retrieval precision is improved, and the retrieval speed can be further improved; and finally, performing early data storage and later data retrieval by using a faiss similarity search tool, so that the working efficiency can be obviously improved.
Drawings
FIG. 1 is a flowchart of an embodiment of an image feature association processing method based on geospatial location partitioning according to the present invention;
fig. 2 is a logic relationship diagram of an image feature association processing system based on geospatial location partitioning according to an embodiment of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of one or more embodiments. It may be evident, however, that such embodiment(s) may be practiced without these specific details.
Specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.
Example 1
In order to explain the image feature association processing method based on the geographic spatial location division provided by the present invention, fig. 1 shows a flow of the image feature association processing method based on the geographic spatial location division provided by the present invention.
As shown in fig. 1, the image feature association processing method based on geospatial location division provided by the present invention includes:
step S110: according to the geographic position of each video stream camera where a target image sample (namely, raw data) is located, a first hash map table for storing vector characteristics of the raw data is established in a system memory.
Specifically, in order to distinguish the geographical source of the target image sample, the total geographic space where all the video stream cameras are located may be subjected to measurement processing, and according to the geographic spatial position of each video stream camera, a unique cam id identifier is attached to the video stream camera, so as to distinguish the video stream cameras in different geographic spatial positions.
More specifically, in order to facilitate storage of original data and later-stage image retrieval, a total geographic space where all video stream cameras are located may be clustered and divided according to geographic spatial positions (i.e., longitude and latitude) of the video stream cameras, into at least 5000 (determined according to the density of the video stream cameras in actual work) spatial regions (i.e., point location classification spaces), each spatial region including a plurality of point locations of the video stream cameras, and then a mapping relationship y (f) (x) between a point and a plane is established between the geographic position (i.e., longitude and latitude x) of each video stream camera and the point location classification space (y) to which the video stream camera belongs; the mapping relation table is stored by establishing a corresponding first hash map table in a system memory, keys of the first hash map table are numbers id (y) of the point location classification space (y), and values of the first hash map table are camid and a vector feature list of all video stream cameras included in the point location classification space (y).
It should be noted that cluster partitioning is an existing region partitioning technology, and the present invention mainly uses this cluster partitioning technology, and therefore, detailed data processing procedures of this technology are not described herein again.
Step S120: extracting image text characteristics and sample vector characteristics of a target image sample (original data), storing the image text characteristics and a preset associated zone bit into a preset database, and storing the sample vector characteristics and the associated zone bit into a vector characteristic list; it should be noted that after feature extraction, each target image sample is given its unique associated flag feature id to distinguish different target image samples, and in addition, the image text feature and the sample vector feature are also associated by the associated flag.
Specifically, in the process of extracting image text features and sample vector features of a target image sample (original data), the target image sample or a camera video stream where the target image sample is located is directly input into a preset deep convolutional neural network, feature extraction is performed on targets (such as pedestrians, vehicles and the like) in the target image sample or the camera video stream through the deep convolutional neural network, finally, image text features and vector features of the target image sample are obtained, a unique feature _ id identifier is given to the target image sample, and data association between image text features and sample vector features of the target image sample is achieved through the feature _ id identifier.
More specifically, the preset database in which the image text features and the associated flag feature _ id are stored may be a relational database so as to be used for later-stage text feature retrieval; a relational database is a data organization composed of two-dimensional tables and the relations between the two-dimensional tables. Certainly, according to the characteristics of the text features, a non-relational database can be set and stored in key value pairs, the structure is not fixed, each tuple can have different fields, each tuple can be added with some key value pairs according to needs, the method is not limited to a fixed structure, and the time and space expenses can be reduced. In practical use, the appropriate database type can be reasonably selected according to the data structure.
It should be noted that, because the number of original data (target image samples) is very large, the storage time is long, so that the storage of each original data is performed in a certain order, when the video stream cameras where the two previous and next target image samples are located are the same, if two cam id identifiers are continuously established according to the two target image samples, a phenomenon of data matching confusion occurs in the later period, in order to solve the problem, before the sample vector features and the associated flag bits are stored in the vector feature list,
determining a corresponding cam id identifier and a point location classification space according to the geographic position of the target image sample;
then judging whether the cam id identification and the number of the point location classification space both exist in the first hash map table, if so, directly storing the sample vector characteristics and the associated zone bit of the target image sample into a vector characteristic list corresponding to the geographic position (corresponding to one cam id identification) of the target image sample;
otherwise, establishing a new mapping relation according to the cam id identifier corresponding to the target image sample and the serial number of the point location classification space, establishing a new vector feature list according to the mapping relation, and storing the vector features and the associated zone bits of the target image into the new vector feature list.
Preferably, in view of a relationship between memory capacity occupation and data processing of the system, in order to ensure that the entire system has higher data processing efficiency, after the time length T that the sample vector feature and the associated flag bit of each target image sample are stored in the vector feature list, it may be determined whether the system memory exceeds a capacity threshold, and if so, the sample vector feature and the associated flag bit thereof are stored in the second hash map table of the preset hard disk, and data in the corresponding vector feature list in the system memory is deleted. By the method, the transfer of the sample vector feature storage position can be realized, and the phenomenon that the working efficiency is influenced due to the overlarge occupied system memory is avoided.
More preferably, in order to ensure the later retrieval speed of the sample vector features of the part of the storage location transfer, after deleting data in a corresponding vector feature list in a system memory, storing a storage path of the sample vector features stored in a preset hard disk into the third hash map table, and quickly finding corresponding sample feature vectors according to the storage path in the third hash map table during later deceleration; in addition, the occupied space of the storage path of the characteristic feature vector is extremely small and can be ignored, so that the working efficiency of the system is not influenced.
Step S130: and acquiring the vector characteristics and the belonged geographic position of a target image to be searched, and retrieving the first hash map table according to the belonged geographic position to acquire a vector characteristic list corresponding to the belonged geographic position.
Specifically, the process of obtaining the vector features of the target image to be searched is the same as the process of obtaining the features of the sample target image, and both the process and the process are implemented by extracting the features through a preset deep convolutional neural network, which is not described herein again.
In the process of retrieving the first hash map table according to the belonged geographic position to obtain the vector feature list corresponding to the belonged geographic position, firstly, determining a cam id identifier of a video stream camera to which the target image belongs and a point location classification space to which the video stream camera belongs according to the belonged geographic position of the target image to be searched, and then retrieving in the first hash map table according to the number of the cam id identifier and the point location classification space to find the corresponding vector feature list.
It should be noted that, for original data, some sample target images of the existing video stream camera may not be acquired in the previous data storage, and to improve the coverage of the original data, vector features of target images to be searched that do not correspond to camid may be stored in the vector feature list.
Step S140: calculating the similarity between the vector features of each sample in the vector feature list and the vector features of the target image to be searched, and determining K sample vector features with the highest similarity and associated flag bits thereof; and then, retrieving the preset database according to the K associated zone bits to obtain corresponding K image text characteristics.
Specifically, during actual search, classification calculation is performed according to cam id, a corresponding ID (y) key value is mapped, if the ID (y) key value exists in the hash map, the faiss is called to perform similarity search on a vector feature list of the ID (y) key value, and similarity and feature id identifications of the most similar K targets are obtained.
More specifically, calculating the similarity between each sample vector feature in the vector feature list and the vector feature of the target image to be searched through a faiss similarity search tool or other tools with similarity calculation, then sorting the vector features in the vector feature list according to the similarity, and finally obtaining K sample vector features with the highest similarity and associated flag bits thereof; wherein, the value of K is determined according to actual requirements and can be selected as 10.
Through the step S140, K sample vector features and K image text features corresponding to the target image to be searched can be retrieved, and K sample target images with the highest similarity to the target image to be searched can be found through data reduction and other manners according to the K sample vector features and the K image text features, so that the purpose of finally searching the image by using the image is achieved, and the working efficiency and the searching accuracy are remarkably improved.
Preferably, because the sample vector features of a part of the sample target images are stored in a preset hard disk, the second hash map table in the preset hard disk can be retrieved according to the belonged geographic position to obtain a vector feature list corresponding to the belonged geographic position;
calculating the similarity between the vector features of each sample in the vector feature list and the vector features of the target image to be searched, and determining K sample vector features with the highest similarity and associated flag bits thereof;
sorting the similarity of the 2K sample vector features from the first hash map table and the second hash map table, and acquiring the K sample vector features with the highest similarity and associated flag bits thereof;
and searching the preset database according to the K associated zone bits to obtain corresponding K image text characteristics.
It should be noted that the search process of the second hash map table of the preset hard disk is the same as the search process of the second hash map table, and can be implemented by a faiss similarity search tool, which is not described herein again.
Furthermore, it should be further explained that, regarding the geographic location of the target image to be searched (i.e., the geographic location of the video stream camera of the target image to be searched, corresponding to the unique cam id identification), due to the unknown source of the target image to be searched, in most cases, the geographic position of the target image cannot be accurately determined, and at this time, that is, according to the source region (or possible source region) of the target image to be searched, the geographic positions of all the video stream cameras in the source region are determined as the geographic positions of the target image to be searched, thus forming a cam id list, then repeating the step S130 and the step S140 according to the cam id marks of the cam id list in turn, thereby forming a plurality of sets of feature sets (each set of feature set corresponds to K sample vector features and K image text features), and then obtaining a plurality of groups (K in each group) of sample target images with the highest similarity of the search target images. By the method, more effective sample target images can be obtained on the premise of ensuring the searching speed, and the working efficiency is further improved.
In the actual use process, if the established point-surface mapping relation and the search speed and accuracy of the vector feature list do not achieve the expected effect, the classification measurement can be carried out again according to the cam id identification of the original feature vector to obtain a new y ═ f (x) point-surface mapping relation, and a new hash map is established for storage. The new vector feature list file block of the corresponding item of the ID (y) number can be rewritten according to the indexing mode provided by faiss, such as inversion, vector dimension reduction and the like, and the adaptation of the search speed and the search precision is carried out on the premise of adapting to the software and hardware configuration of the system, so that the corresponding search speed and precision are improved.
Example 2
Corresponding to the method, the present application further provides an image feature association system based on geospatial location partitioning, and fig. 2 shows a logical relationship of an image feature association processing system based on geospatial location partitioning according to an embodiment of the present invention, as shown in fig. 2, the system includes:
the table building unit is used for building a first hash map table in a system memory according to the geographic position of each video stream camera, wherein a vector feature list corresponding to each geographic position is arranged in the first hash map table;
the data storage unit is used for extracting image text characteristics and sample vector characteristics of a target image sample, storing the image text characteristics and a preset associated zone bit into a preset database, and storing the sample vector characteristics and the associated zone bit into the vector characteristic list; wherein the image text features and the vector features are associated through the associated flag bits;
the vector feature list acquisition unit is used for acquiring the vector features of the target image to be searched and the belonged geographic position, and retrieving the first hash map table according to the belonged geographic position to acquire a vector feature list corresponding to the belonged geographic position;
the similarity matching unit is used for calculating the similarity between each sample vector feature in the vector feature list and the vector feature of the target image to be searched and determining K sample vector features with the highest similarity and associated flag bits thereof;
and the retrieval unit is used for retrieving the preset database according to the K associated zone bits so as to obtain corresponding K image text characteristics.
Wherein, the table building unit further comprises a space dividing unit and a mapping unit (not shown in the figure).
The space division unit is used for clustering and dividing the geographic space where the video stream camera is located to form a point location classification space;
the mapping unit is used for establishing a first hash map table according to the mapping relation between the geographic position of each video stream camera and the point location classification space to which the video stream camera belongs; and the key of the first hash map table is the serial number of the point location classification space, and the value of the first hash map table is the vector feature list of all video stream cameras contained in the point location classification space.
In another embodiment of the present invention, the data storage unit further includes a point location determining unit and a number checking unit (not shown in the figure), configured to determine a storage manner of the sample vector feature and the associated flag bit before storing the sample vector feature and the associated flag bit into the vector feature list.
Specifically, the point location determining unit is configured to determine a corresponding point location classification space according to the geographic location of the target image sample; the number checking unit is used for judging whether the number of the point location classification space exists in the first hash map table or not; if the serial number of the point location classification space exists, directly storing the sample vector characteristics and the associated flag bit of the target image sample into a vector characteristic list corresponding to the geographic position of the target image sample;
otherwise, establishing a new vector feature list according to the serial number of the point location classification space corresponding to the target image sample, and storing the vector features and the associated flag bits of the target image into the new vector feature list.
In addition, the data storage unit may further include a table capacity monitoring unit (not shown in the figure), configured to determine whether the system memory exceeds a capacity threshold after the time duration T when the sample vector feature and the associated flag bit of each target image sample are stored in the vector feature list, and if the system memory exceeds the capacity threshold, store the sample vector feature and the associated flag bit thereof in a second hash map table of a preset hard disk, and delete data in the vector feature list corresponding to the system memory.
For the image feature association system based on the geographic spatial location division, there are other specific embodiments corresponding to the image feature association method based on the geographic spatial location division, and the implementation manners thereof are all similar to those of the image feature association method based on the geographic spatial location division, and are not described in detail herein.
As can be seen from the foregoing embodiments, the image feature association processing method and system based on geographic spatial location division provided by the present invention have at least the following advantages:
1. by dividing the geographical position of the source of the picture and establishing a corresponding vector feature list, the method can realize classified storage of original data, improve the later retrieval speed, and obviously improve the retrieval precision in a mode of predetermining the geographical position of the image to be searched;
2. by associating the image text features and the vector features of the original data and storing the image text features and the vector features to different storage positions, the simultaneous matching of the image text features and the vector features can be realized, the image retrieval precision is improved, and the retrieval speed can be further improved.
3. The first hash map table, the second hash map table, the third hash map table and the data storage and data retrieval of the preset database are realized through a faiss tool, and therefore the working efficiency of the system can be remarkably improved.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. An image feature association processing method based on geographic spatial position division is characterized by comprising the following steps:
establishing a first hash map table in a system memory according to the geographic position of each video stream camera, wherein a vector feature list corresponding to each geographic position is arranged in the first hash map table;
extracting image text characteristics and sample vector characteristics of a target image sample, storing the image text characteristics and a preset associated zone bit into a preset database, and storing the sample vector characteristics and the associated zone bit into a vector characteristic list; wherein the image text features and the sample vector features are associated by the associated flag bits;
acquiring vector characteristics and a belonged geographic position of a target image to be searched, and retrieving the first hash map table according to the belonged geographic position to acquire a vector characteristic list corresponding to the belonged geographic position;
calculating the similarity between the vector features of each sample in the vector feature list and the vector features of the target image to be searched, and determining K sample vector features with the highest similarity and associated flag bits thereof;
and searching a preset database according to the determined K associated zone bits with the highest similarity so as to obtain corresponding K image text characteristics.
2. The image feature association processing method based on geospatial location partitioning according to claim 1, wherein the process of establishing the first hash map table in the system memory according to the geographic location of each video stream camera comprises:
clustering and dividing the geographic space where the video stream camera is located to form a point location classification space;
and establishing a first hash map table according to a mapping relation between the geographic position of each video stream camera and the point location classification space to which the video stream camera belongs, wherein keys of the first hash map table are serial numbers of the point location classification space, and values of the first hash map table are vector feature lists of all the video stream cameras contained in the point location classification space.
3. The image feature association processing method based on geospatial location division according to claim 2, wherein before storing the sample vector features and the association flag into the vector feature list, the method further comprises:
determining a corresponding point location classification space according to the geographic position of the target image sample;
judging whether the serial number of the point location classification space exists in the first hash map table or not, if so, directly storing the sample vector characteristics and the associated flag bit of the target image sample into a vector characteristic list corresponding to the geographic position of the target image sample;
otherwise, establishing a new vector feature list according to the serial number of the point location classification space corresponding to the target image sample, and storing the vector features and the associated flag bits of the target image into the new vector feature list.
4. The image feature association processing method based on geographic spatial location division according to claim 3,
after the time length T of the sample vector characteristics and the associated flag bits of each target image sample are stored in a vector characteristic list, judging whether the system memory exceeds a capacity threshold value, if so, storing the sample vector characteristics and the associated flag bits in a second hash map table of a preset hard disk, and deleting data in the corresponding vector characteristic list in the system memory.
5. The image feature correlation processing method based on geographic spatial position division according to claim 4, wherein a third hash map table is further arranged in the system memory; and the number of the first and second electrodes,
and after deleting the data in the corresponding vector feature list in the system memory, storing the storage path of the sample vector features stored in the preset hard disk into the third hash map table.
6. The image feature association processing method based on geospatial location division according to claim 5, after obtaining the vector features of the target image to be searched and the geographic location, further comprising:
retrieving a second hash map table in the preset hard disk according to the belonged geographic position to obtain a vector feature list corresponding to the belonged geographic position;
calculating the similarity between the vector features of each sample in the vector feature list and the vector features of the target image to be searched, and determining K sample vector features with the highest similarity and associated flag bits thereof;
sorting the similarity of the 2K sample vector features from the first hash map table and the second hash map table, and acquiring the K sample vector features with the highest similarity and associated flag bits thereof;
and searching the preset database according to the K associated zone bits to obtain corresponding K image text characteristics.
7. The image feature association processing method based on geographic spatial location division according to any one of claims 5 or 6,
and the data storage and the data retrieval of the first hash map table, the second hash map table, the third hash map table and the preset database are realized by a faiss tool.
8. An image feature association processing system based on geospatial location partitioning, the system comprising:
the table building unit is used for building a first hash map table in a system memory according to the geographic position of each video stream camera, wherein a vector feature list corresponding to each geographic position is arranged in the first hash map table;
the data storage unit is used for extracting image text characteristics and sample vector characteristics of a target image sample, storing the image text characteristics and a preset associated zone bit into a preset database, and storing the sample vector characteristics and the associated zone bit into the vector characteristic list; wherein the image text features and the vector features are associated through the associated flag bits;
the vector feature list acquisition unit is used for acquiring the vector features of the target image to be searched and the belonged geographic position, and retrieving the first hash map table according to the belonged geographic position to acquire a vector feature list corresponding to the belonged geographic position;
the similarity matching unit is used for calculating the similarity between each sample vector feature in the vector feature list and the vector feature of the target image to be searched and determining K sample vector features with the highest similarity and associated flag bits thereof;
and the retrieval unit is used for retrieving the preset database according to the K associated zone bits so as to obtain corresponding K image text characteristics.
9. The image feature association processing system based on geographic spatial location division according to claim 8, wherein the table building unit comprises:
the space dividing unit is used for clustering and dividing the geographic space where the video stream camera is located to form a point location classification space;
and the mapping unit is used for establishing a first hash map table according to the mapping relation between the geographic position of each video stream camera and the point location classification space to which the video stream camera belongs, wherein keys of the first hash map table are serial numbers of the point location classification space, and values of the first hash map table are vector feature lists of all the video stream cameras contained in the point location classification space.
10. The geo-spatial location segmentation based image feature association processing system of claim 9, wherein the data storage unit further comprises:
the point location determining unit is used for determining a corresponding point location classification space according to the geographic position of the target image sample;
a number checking unit, configured to determine whether a number of the point location classification space already exists in the first hash map table; wherein the content of the first and second substances,
if the sample vector characteristics of the target image sample exist, directly storing the sample vector characteristics and the associated zone bits of the target image sample into a vector characteristic list corresponding to the geographic position of the target image sample;
otherwise, establishing a new vector feature list according to the serial number of the point location classification space corresponding to the target image sample, and storing the vector features and the associated flag bits of the target image into the new vector feature list.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010243925.4A CN111428063B (en) | 2020-03-31 | 2020-03-31 | Image feature association processing method and system based on geographic space position division |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010243925.4A CN111428063B (en) | 2020-03-31 | 2020-03-31 | Image feature association processing method and system based on geographic space position division |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111428063A true CN111428063A (en) | 2020-07-17 |
| CN111428063B CN111428063B (en) | 2023-06-30 |
Family
ID=71556177
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010243925.4A Active CN111428063B (en) | 2020-03-31 | 2020-03-31 | Image feature association processing method and system based on geographic space position division |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111428063B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112104730A (en) * | 2020-09-11 | 2020-12-18 | 杭州海康威视***技术有限公司 | Scheduling method and device of storage tasks and electronic equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106776849A (en) * | 2016-11-28 | 2017-05-31 | 西安交通大学 | A kind of method and guide system to scheme quick-searching sight spot |
| CN109033308A (en) * | 2018-07-16 | 2018-12-18 | 安徽江淮汽车集团股份有限公司 | A kind of image search method and device |
| CN110019889A (en) * | 2017-12-01 | 2019-07-16 | 北京搜狗科技发展有限公司 | Training characteristics extract model and calculate the method and relevant apparatus of picture and query word relative coefficient |
| CN110069650A (en) * | 2017-10-10 | 2019-07-30 | 阿里巴巴集团控股有限公司 | A kind of searching method and processing equipment |
| CN110851629A (en) * | 2019-10-14 | 2020-02-28 | 信阳农林学院 | Image retrieval method |
-
2020
- 2020-03-31 CN CN202010243925.4A patent/CN111428063B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106776849A (en) * | 2016-11-28 | 2017-05-31 | 西安交通大学 | A kind of method and guide system to scheme quick-searching sight spot |
| CN110069650A (en) * | 2017-10-10 | 2019-07-30 | 阿里巴巴集团控股有限公司 | A kind of searching method and processing equipment |
| CN110019889A (en) * | 2017-12-01 | 2019-07-16 | 北京搜狗科技发展有限公司 | Training characteristics extract model and calculate the method and relevant apparatus of picture and query word relative coefficient |
| CN109033308A (en) * | 2018-07-16 | 2018-12-18 | 安徽江淮汽车集团股份有限公司 | A kind of image search method and device |
| CN110851629A (en) * | 2019-10-14 | 2020-02-28 | 信阳农林学院 | Image retrieval method |
Non-Patent Citations (1)
| Title |
|---|
| 叶巍;龚建华;郭娜;路梅;赵向军;: "基于流形结构的图像地理信息标注方法" * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112104730A (en) * | 2020-09-11 | 2020-12-18 | 杭州海康威视***技术有限公司 | Scheduling method and device of storage tasks and electronic equipment |
| CN112104730B (en) * | 2020-09-11 | 2023-03-28 | 杭州海康威视***技术有限公司 | Scheduling method and device of storage tasks and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111428063B (en) | 2023-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3709184B1 (en) | Sample set processing method and apparatus, and sample querying method and apparatus | |
| US6507669B1 (en) | Method of selecting clusters of items using a fuzzy histogram analysis | |
| CN108388656B (en) | Image searching method based on mark correlation | |
| CN103577418B (en) | Magnanimity Document distribution formula retrieval re-scheduling system and method | |
| CN103617163B (en) | Quick target association method based on cluster analysis | |
| CN106033443B (en) | A kind of expanding query method and device in vehicle retrieval | |
| CN106874421A (en) | Image search method based on self adaptation rectangular window | |
| CN106570166B (en) | Video retrieval method and device based on multiple locality sensitive hash tables | |
| CN111353062A (en) | Image retrieval method, device and equipment | |
| CN110888880A (en) | Proximity analysis method, device, equipment and medium based on spatial index | |
| Kipf et al. | Approximate geospatial joins with precision guarantees | |
| CN106777350B (en) | Method and device for searching pictures with pictures based on bayonet data | |
| CN116126864A (en) | Index construction method, data query method and related equipment | |
| CN111428063B (en) | Image feature association processing method and system based on geographic space position division | |
| CN108694411B (en) | Method for identifying similar images | |
| Schoier et al. | A methodology for dealing with spatial big data | |
| Abdel-Mottaleb | Image retrieval based on edge representation | |
| CN116721246B (en) | Continuous frame point cloud rapid labeling method and system | |
| Le et al. | Improving logo spotting and matching for document categorization by a post-filter based on homography | |
| Pfoser et al. | Trajectory indexing using movement constraints | |
| Feng et al. | Real-time SLAM relocalization with online learning of binary feature indexing | |
| CN115129915A (en) | Repeated image retrieval method, device, equipment and storage medium | |
| Zhang et al. | Clustering in dynamic spatial databases | |
| Manzo et al. | Attributed relational sift-based regions graph for art painting retrieval | |
| CN115757896A (en) | Vector retrieval method, device, equipment and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |