US20220138893A9 - Distributed image analysis method and system, and storage medium - Google Patents

Distributed image analysis method and system, and storage medium Download PDF

Info

Publication number
US20220138893A9
US20220138893A9 US17/350,205 US202117350205A US2022138893A9 US 20220138893 A9 US20220138893 A9 US 20220138893A9 US 202117350205 A US202117350205 A US 202117350205A US 2022138893 A9 US2022138893 A9 US 2022138893A9
Authority
US
United States
Prior art keywords
server
image
images
index
candidate images
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
US17/350,205
Other languages
English (en)
Other versions
US20210312587A1 (en
Inventor
Ting Lei
Chao Zou
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huawei Technologies Co Ltd
Original Assignee
Huawei Technologies Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huawei Technologies Co Ltd filed Critical Huawei Technologies Co Ltd
Publication of US20210312587A1 publication Critical patent/US20210312587A1/en
Publication of US20220138893A9 publication Critical patent/US20220138893A9/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
    • 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
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/60—Memory management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/56—Provisioning of proxy services
    • H04L67/565—Conversion or adaptation of application format or content
    • H04L67/5651—Reducing the amount or size of exchanged application data
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network

Definitions

  • the present invention relates to the field of image analysis technologies, and specifically, relates to a distributed image analysis method and system, and a storage medium for distributed image analysis.
  • image collision analysis is deeply applied to the field of security, thereby improving efficiency of investigating and cracking major and key criminal cases, reducing workload of investigators in analyzing case information, and improving efficiency of case handling.
  • investigators may perform image collision analysis on surveillance images obtained by respective web cameras, to search for an object who frequently appears in a specific period of time and in a specific place.
  • An application scenario is, for example, analysis of a scalper in a railway station.
  • a video surveillance system is usually a multi-level domain formed from a provincial level (a center) to prefectural levels (edges).
  • a center a center
  • prefectural levels edges
  • a state a plurality of prefectures are linked.
  • the provincial center initiates an image collision task, and the image collision task is executed by the prefectures.
  • Embodiments of the present invention provide a distributed image analysis method and system, and a storage medium for distributed image analysis.
  • a first server is enabled to send image collision analysis results to a second server in a form of a result set, and then the second server performs second image collision analysis on the received result set to sift result objects that frequently appear,
  • the first server may be a prefectural server
  • the second server may be a provincial central server, so that computing power of the prefectural server is fully utilized, resource pressure of the provincial central server is reduced, and time consumption of image collision analysis is decreased.
  • a volume of the data set is smaller than that of surveillance images, so that pressure of service data transmission is reduced.
  • a first aspect of this application provides a distributed image analysis method, including: obtaining, by a first server, a result set through image collision analysis, where the result set includes an index image, the index image records a result object, the index image corresponds to an object frequency, and the object frequency is used to indicate a quantity of times that the result object is photographed; and sending, by the first server, the result set to a second server, where the result set is used for the second server to perform second image collision analysis.
  • the distributed image analysis method provided by the embodiments of this application is applied to an image analysis system, and the image analysis system includes a plurality of first servers and a second server connected to the plurality of first servers.
  • the first server sends image collision analysis results to the second server in a form of a result set, and each result object in the result set corresponds to only one index image, so that a data transmission volume is reduced.
  • the index image corresponds to the object frequency, so that the data transmission volume is reduced without affecting presentation of the image collision analysis results of the first server, and in addition, computing power of the first server is fully utilized, resource pressure of the second server is reduced, and computing efficiency of an image collision analysis algorithm is improved.
  • the obtaining, by a first server, a result set through image collision analysis includes: obtaining, by the first server, a plurality of candidate images, where the candidate images are images that include the result object among images stored in the first server; determining, by the first server, the index image from the plurality of candidate images; and determining, by the first server, a quantity of the candidate images as the object frequency.
  • the first server may obtain the index image from the candidate images randomly or based on a preset rule, to generate the result set.
  • the index image can be obtained from the candidate images quickly, and computing power is saved.
  • the first server adds the quantity of the candidate images to the index image as the object frequency, so that the index image can represent all candidate images recording a same result object in the first server, for the second server to perform the second image collision analysis.
  • the method before the obtaining, by a first server, a result set through image collision analysis, the method further includes: receiving, by the first server, a target instruction sent by the second server, where the target instruction includes a confidence threshold; and the obtaining, by the first server, a plurality of candidate images includes: performing, by the first server, feature extraction on stored images, to obtain a feature value of each of the stored images, where the stored images are all images stored in the first server, performing, by the first server, image collision analysis on feature values of the stored images, to obtain a confidence of each stored image; and obtaining, by the first server, stored images having a confidence greater than or equal to the confidence threshold as the candidate images that collectively include the same result object.
  • the target instruction includes the confidence threshold
  • the first server selects images having a confidence greater than the confidence threshold from the stored images as the candidate images. That is, if a confidence between every two result objects recorded in images is higher than the confidence threshold, the result objects are determined as the same result object.
  • the performing, by the first server, feature extraction on stored images includes: performing, by the first server, feature extraction on the stored images in the target time period, and/or performing, by the first server, feature extraction on the stored images photographed by a camera that is identified by the target camera identifier.
  • the determining, by the first server, the index image from the plurality of candidate images includes: obtaining, by the first server, image quality of each candidate image through an image analysis algorithm; and selecting, by the first server, an image having the highest image quality from the candidate images as the index image.
  • the first server obtains the image having the highest image quality among the candidate images as the index image through the image analysis algorithm, where the index image accurately represents the result object collectively recorded by the candidate images, and is provided for the second server to perform the second image collision analysis.
  • the highest image quality may be that lighting conditions are the best when the image is taken, or the like, or the photographed object is not blocked.
  • the method further includes: obtaining, by the first server, a resource address of the candidate images; and obtaining, by the first server, the result set, where the result set includes the resource address, the resource address is used for the second server to access the candidate images through the resource address.
  • the resource address of the candidate images is added to the result set sent by the first server, so that the second server can access the candidate images stored in the first server through the resource address.
  • the resource address saves more resources in terms of format and does not occupy a transmit resource and a storage resource of the second server while enabling the second server to access all candidate images, so that all candidate images recording the result object of interest are obtained.
  • a second aspect of this application provides a distributed image analysis method, including: receiving, by a second server, result sets that are separately sent by a plurality of first servers and that are each obtained by each first server, where each result set includes an index image, the index image records a result object, the index image corresponds to an object frequency, and the object frequency is used to indicate a quantity of times that the result object is photographed; performing, by the second server, feature extraction on an index image in each result set, to obtain a feature value of the index image; performing, by the second server, image collision analysis on extracted feature values of all the index images, to obtain a confidence of each index image; and obtaining, by the second server, a sum of object frequencies corresponding to the index images when determining that a confidence between the index images in the result sets sent by the plurality of first servers is greater than or equal to a preset value.
  • the result set received by the second server includes the index image and the object frequency corresponding to the index image. Therefore, the second server only needs to receive one image, that is, the index image, to obtain all information of a result object, thereby greatly improving transmission efficiency.
  • the index image is an ordinary image file and is compatible with algorithms of different vendors
  • the first server and the second server are compatible with each other even if algorithms of different vendors are installed on the first server and the second server, so that computing power of servers at all levels is fully utilized, thereby improving computing efficiency.
  • the result set includes a resource address of the candidate images
  • the candidate images are images that record the result object among images stored in the first server
  • the stored images are all images stored in the first server
  • the method further includes: accessing, by the second server, the candidate images through the resource address.
  • the second server can access the candidate images stored in the first server through the resource address.
  • the method before the second server receives the result sets sent by the plurality of first servers, the method further includes: sending, by the second server, a target instruction to the first servers, where the target instruction includes a confidence threshold, and a confidence between any candidate images in the first server is greater than or equal to the confidence threshold.
  • the second server can control, through the target instruction, the first server to compare locally stored images through image collision analysis when the first server executes the image collision algorithm. If a confidence between two stored images is greater than or equal to the confidence threshold, it is determined that the two stored images record a same result object, so the two stored images are classified as candidate images recording the same result object. Therefore, the second server can control, through the target instruction, conditions through which the first server sifts the candidate images, thereby regulating and controlling a working condition of the first server.
  • the target instruction further includes a target time period and/or a target camera identifier
  • the target instruction is used for the first server to perform feature extraction on the stored images in the target time period, and/or the first server performs feature extraction on the stored images photographed by a camera that is identified by the target camera identifier.
  • the second server ma control the first server to obtain candidate images photographed in the target time period; and if the target instruction includes the target camera identifier, the first server obtains candidate images photographed by a camera that is identified by the target camera identifier, so that the second server can sift conditions of image collision analysis through the target instruction.
  • a third aspect of this application provides an image analysis server, including: a first obtaining unit, configured to obtain a result set through image collision analysis, where the result set includes an index image, the index image records a result object, the index image corresponds to an object frequency, and the object frequency is used to indicate a quantity of times that the result object is photographed; and a first sending unit, configured to send the result set obtained by the first obtaining unit to a second server, where the result set is used for the second server to perform second image collision analysis.
  • the server sends the image collision analysis results to the second server in a form of a result set, and the result set uses the index image to represent a result object, so that data transmission volume is reduced.
  • the index image corresponds to the object frequency, so that the second server can perform frequency calculation, and in addition, computing power of servers at all levels is fully utilized, and computing efficiency of an image collision analysis algorithm is improved.
  • the first obtaining unit is further configured to: obtain a plurality of candidate images, where the candidate images are images that include the result object among images stored in the first server, determine the index image from the plurality of candidate images; and determine a quantity of the candidate images as the object frequency.
  • the server may select, specifically randomly or based on a preset rule, at least one of the candidate images stored locally as the index image, to generate the result set.
  • the index image can be obtained from the candidate images quickly, and computing power is saved.
  • the first server adds the quantity of the candidate images to the index image as the object frequency, so that the index image can represent all candidate images recording a same result object in the first server, for the second server to perform the second image collision analysis.
  • the server further includes a first receiving unit, configured to receive a target instruction sent by the second server, where the target instruction includes a confidence threshold; and the first obtaining unit is configured to: perform feature extraction on stored images, to obtain a feature value of each of the stored images, where the stored images are all images stored in the first server; perform image collision analysis on feature values of the stored images, to obtain a confidence of each stored image; and obtain stored images having a confidence greater than or equal to the confidence threshold as the candidate images that collectively include the same result object.
  • the server compares locally stored images through image collision analysis. If a confidence between two stored images is greater than or equal to the confidence threshold, it is determined that the two stored images record the same result object, so the two stored images are classified as candidate images recording the same result object. Therefore, the second server can control, through the target instruction, conditions through which the server sifts the candidate images, thereby regulating and controlling a working condition of the server.
  • the first receiving unit is further configured to receive the target instruction sent by the second server, where the target instruction further includes a target time period and/or a target camera identifier, and the first obtaining unit is further configured to perform feature extraction on the stored images in the target time period, and/or perform feature extraction on the stored images photographed by a camera that is identified by the target camera identifier.
  • the server obtains candidate images photographed in the target time period; and if the target instruction includes the target camera identifier, the first server obtains candidate images photographed by a camera that is identified by the target camera identifier, so that the first obtaining unit can obtain surveillance images of a designated location as required by the second server.
  • the first obtaining unit is further configured to: perform feature extraction on each of the all candidate images, to obtain a feature value of the candidate image; perform image collision analysis on the feature values of the candidate images, to obtain a confidence level of each of the candidate images; and obtain a candidate image having the highest confidence from the candidate images as the index image.
  • the server after obtaining the candidate images, the server obtains the image having the highest confidence among the candidate images as the index image, where the index image accurately represents the result object collectively recorded by the candidate images, and is provided for the second server to perform the second image collision analysis.
  • the first obtaining unit is further configured to: obtain a resource address of the candidate images; and obtain the result set, where the result set includes the resource address, the resource address is used for the second server to access the candidate images through the resource address.
  • the resource address of the candidate images is added to the result set sent by the server, so that the second server can access the candidate images stored in the server through the resource address.
  • the resource address saves more resources in terms of format and does not occupy a transmit resource and a storage resource of the second server while enabling the second server to access all candidate images, so that all candidate images recording the result object of interest are obtained.
  • a fourth aspect of this application provides an image analysis server, including:
  • a second obtaining unit configured to receive result sets that are separately sent by the plurality of first servers and that are each obtained by each first server, where each result set includes an index image, the index image records a result object, the index image corresponds to an object frequency, and the object frequency is used to indicate a quantity of times that the result object is photographed; an analysis unit, configured to: perform feature extraction on an index image in each result set obtained by the second obtaining unit, to obtain a feature value of the index image; and perform image collision analysis on extracted feature values of all the index images, to obtain a confidence of each of the index images; and an output unit, configured to obtain a sum of object frequencies corresponding to the index images when determining that a confidence between the index images in the result sets sent by the plurality of first servers is greater than or equal to a preset value.
  • the result set received by the server includes the index image and the object frequency corresponding to the index image.
  • the server only needs to perform image collision analysis on the index image to obtain information of a frequency at which a result object appears, thereby greatly improving transmission efficiency.
  • the index image is an ordinary image file and is compatible with algorithms of different vendors, servers are compatible with each other even if algorithms of different vendors are installed on the servers, so that computing power of servers at all levels is fully utilized, thereby improving computing efficiency.
  • the result set includes a resource address of the candidate images
  • the candidate images are images that record the result object among images stored in the first server
  • the stored images are all images stored in the first server
  • the second obtaining unit is further configured to access the candidate images through the resource address.
  • the server can access the candidate images stored in the first server through the resource address.
  • the server further includes a second sending unit, configured to send a target instruction to the first server, where the target instruction includes a confidence threshold, and a confidence between any candidate images in the first server is greater than or equal to the confidence threshold.
  • the server may control sifting conditions of the image collision analysis of the first server through the target instruction.
  • the second sending unit is further configured to send the target instruction to the first server, where the target instruction further includes a target time period and/or a target camera identifier, where the target instruction is used for the first server to perform feature extraction on the stored images in the target time period, and/or the first server performs feature extraction on the stored images photographed by a camera that is identified by the target camera identifier.
  • the server may control the first server to obtain candidate images photographed in the target time period, thereby sifting conditions of image collision analysis.
  • a fifth aspect of this application provides an image analysis server, including: a processor and a memory, where the memory is configured to store a computer-executable instruction; the processor executes the computer-executable instruction stored in the memory when the image analysis server is run, to cause the image analysis server to perform the distributed image analysis method according to the first aspect or any example embodiment of the first aspect.
  • a sixth aspect of this application provides an image analysis server, including: a processor and a memory, where the memory is configured to store a computer-executable instruction; the processor executes the computer-executable instruction stored in the memory when the image analysis server is run, to cause the image analysis server to perform the distributed image analysis method according to the second aspect or any example embodiment of the second aspect.
  • a seventh aspect of this application provides a computer-readable storage medium, including one or more instructions, where the one or more instructions, when run on a computer device, causes the computer device to perform the method according to the first aspect or any implementation of the first aspect.
  • An eighth aspect of this application provides a computer-readable storage medium, including one or more instructions, where the one or more instructions, when run on a computer device, causes the computer device to perform the method according to the second aspect or any implementation of the second aspect.
  • the embodiments of the present invention provide a distributed image analysis method and system, and a storage medium for distributed image analysis, including: obtaining, by a first server, a result set through image collision analysis, where the result set includes an index image, the index image records a result object, the index image corresponds to an object frequency, and the object frequency is used to indicate a quantity of times that the result object is photographed; separately sending, by a plurality of first servers, a result set obtained by each of the plurality of first servers to a second server; performing, by the second server, feature extraction on an index image in each result set, to obtain a feature value of the index image; performing, by the second server, image collision analysis on extracted feature values of all the index images, to obtain a confidence of each of the index images; and obtaining, by the second server, a sum of object frequencies corresponding to the index images when determining that a confidence between the index images in the result sets sent by the plurality of first servers is greater than or equal to a preset value.
  • the first server may be a prefectural server, and the second server may be a provincial central server.
  • the format of the data set is not limited by a type of AI service algorithm. Computing power of the first servers and the second server is fully utilized in two times of image collision analysis, so that resource pressure of the second server is reduced, and a speed of image collision analysis is increased. In addition, a data set volume is smaller, and pressure of service data transmission is reduced.
  • FIG. 1 is an architectural topology diagram of image analysis of two-level, provincial and prefectural servers according to an embodiment of this application;
  • FIG. 2 is a schematic diagram of an exemplary distributed image analysis method according to an embodiment of this application.
  • FIG. 3 is a schematic diagram of another exemplary distributed image analysis method according to an embodiment of this application.
  • FIG. 4 is a schematic diagram of another exemplary distributed image analysis method according to an embodiment of this application.
  • FIG. 5 is a schematic diagram of another exemplary distributed image analysis method according to an embodiment of this application.
  • FIG. 6A and FIG. 6B are a schematic diagram of another exemplary distributed image analysis method according to an embodiment of this application.
  • FIG. 7 is a schematic diagram of a server in an exemplary distributed image analysis system according to an embodiment of this application.
  • FIG. 8 is a schematic diagram of an exemplary apparatus of a distributed image analysis system according to an embodiment of this application.
  • Embodiments of the present invention provide a distributed image analysis method and system, and a storage medium for distributed image analysis.
  • a first server is enabled to send image collision analysis results to a second server in a form of a result set, and the second server performs second image collision analysis after obtaining result sets sent by a plurality of first servers, so that results of different algorithms are compatible with each other, and computing power of servers at all levels is fully utilized.
  • the big data-based image collision analysis technology is deeply applied to the field of security, thereby improving efficiency of investigating and cracking major and important cases, reducing workload of investigators in analyzing case information, and improving efficiency of case handling.
  • investigators may perform image collision analysis on surveillance images obtained by respective web cameras, to search for an object who frequently appears in a specific period of time and in a specific place.
  • An application scenario is, for example, analysis of a scalper in a railway station.
  • An example execution manner of the foregoing image collision analysis algorithm is to sequentially compare a stored target image with other stored images one by one.
  • the target image is any stored image.
  • a similarity between a result object recorded by an image and a result object recorded by the target image is higher than a threshold, it is determined that the result objects recorded by the image and the target image are a same object. In this case, because there are two images recording the same result object, it is determined that the frequency for the result object to be photographed is two.
  • the foregoing steps are repeated, and the target image is continuously compared with another image until an occurrence frequency of the same result object in all stored images is obtained.
  • the similarity is referred to as a confidence in the field of image collision analysis technologies. A high confidence indicates that an image records a result object more clearly.
  • the image may be a photo or a video, and the result object may be a car.
  • a video surveillance system is usually a multi-level domain formed from a provincial level (a center) to prefectural levels (edges).
  • a center a center
  • prefectural levels edges
  • a state a plurality of prefectures are linked.
  • the provincial center initiates an image collision task, and the image collision task is executed by the prefectures.
  • a frequency of being photographed of a result object can be searched for in the whole province, and top N result objects having the highest frequencies of being photographed in the whole province can be obtained.
  • Prefectural systems have computing and analysis capacities. In such a solution, computing power resources of the lower-level prefectures are not used, and instead, prefectural systems are merely used as a data storage system, resulting in a waste of computing power.
  • the provincial center and the prefectures need to transmit a large quantity of image streams and video streams, resulting in high requirements on network bandwidth and latency, and high construction costs.
  • due to a large volume of surveillance images it takes a long time for the provincial center to carry out an image collision analysis service.
  • the embodiments of this application provide a distributed image analysis method and system, and a storage medium for a distributed image analysis.
  • the image collision analysis results of the first server can be sent to the second server in a form of a result set, thereby overcoming differences in algorithms, sharing image collision analysis results between servers of all levels, fully utilizing computing power of servers at all levels in a multi-level domain system, reducing information transmission pressure, and improving computing efficiency.
  • an exemplary distributed image analysis method includes:
  • a first server obtains a result set through image collision analysis.
  • the result set includes an index image
  • the index image records a result object
  • the index image corresponds to an object frequency
  • the object frequency is used to indicate a quantity of times that the result object is photographed.
  • a file name of the index image records information of the object frequency.
  • the first server is a Shenzhen server
  • a result set obtained by the Shenzhen server through image collision analysis includes two result sets A and B.
  • the result set A includes an index image A
  • the result set B includes an index image B.
  • Two result objects, an Object A and an Object B, are recorded separately in the index images.
  • the index image A corresponds to that a quantity of times that the Object A is photographed in Shenzhen is 100
  • index image B corresponds to that a quantity of times that the Object B is photographed in Shenzhen is 200.
  • a plurality of first servers separately send a result set obtained by each first server to a second server.
  • each first server may send one result set or a plurality of result sets to the second server, where only one result object is recorded in each result set, and the result object may be a car.
  • the result object is not specifically limited in embodiments of the present invention.
  • the second server is a Guangdong server
  • a first server namely, a Shenzhen server
  • another first server namely, a Dongguan server
  • the result set C includes an index image C
  • the result set D includes an index image D
  • the result set E includes an index image E.
  • the index image C records that a quantity of times that a result object, an Object C, is photographed in Dongguan is 100
  • the index image D records that a quantity of times that a result object, the Object A, is photographed in Dongguan is 10
  • the index image E records that a quantity of times that a result object, the Object B, is photographed in Dongguan is 20.
  • the second server performs feature extraction on an index image in each of the result sets received from the plurality of first servers, to obtain a feature value of the index image.
  • the second server performs feature extraction through an image analysis algorithm, to obtain the feature value of each index image.
  • a specific calculation process of the image analysis algorithm belongs to the prior art, and details are not described in embodiments of the present invention.
  • the Guangdong server has received the foregoing five index images A, B, C, D, and E.
  • a feature value of the Object A is obtained in the index image A
  • a feature value of the Object B is obtained in the index image B
  • a feature value of the Object C is obtained in the index image C
  • a feature value of the Object A is obtained in the index image D
  • a feature value of the Object B is obtained in the index image E.
  • the second server performs image collision analysis on extracted feature values of all the index images, to obtain a confidence of each index image.
  • the second server obtains confidences between all index images through image collision analysis.
  • the Guangdong server compares the five index images A, B, C, D, and E through the image analysis algorithm, to obtain that a confidence between the image A and the image D is 90%, a confidence between the image B and the image E is 93%, and a confidence threshold between the remaining images is less than 30%.
  • the second server obtains a sum of object frequencies corresponding to the index images when determining that a confidence between the index images in the result sets sent by the plurality of first servers is greater than or equal to a preset value.
  • the second server determines, through the preset value, whether a same result object is recorded between the index images.
  • the Guangdong server presets a confidence to 90%, it is determined that the index image A and the index image D collectively record the result object the Object A, and the index image B and the index image E collectively record the index object—the Object B.
  • the Guangdong server can learn that the Object A's total frequency of being photographed in Shenzhen and Dongguan is 110, and the Object B's total frequency of being photographed in Shenzhen and Dongguan is 220.
  • the Guangdong server may learn of the Object A's total frequency of being photographed and the Object B's total frequency of being photographed in Guangdong in the foregoing manner.
  • the preset value is specified by a user, and a function thereof is to determine, when a confidence between two index images is greater than or equal to the preset value, that the two index images record the same result object.
  • object frequencies corresponding to the two index images are obtained, so that a quantity of times that the result object is photographed is obtained.
  • the second server outputs top N result objects having the highest object frequencies.
  • N is a positive integer greater than or equal to 1, and a specific value of N is set by a user, so that the user may obtain top N result objects having the largest quantities of times of being photographed within the range of the second server.
  • the user sets a value of N to 1, that is, the user requests the Guangdong server to output a result object having the highest object frequency.
  • an output result object having the highest quantity of times of being photographed is the Object B whose total frequency of being photographed in Shenzhen and Dongguan is 220.
  • an objective for which the Guangdong server cooperates with the Shenzhen server and the Dongguan server to perform the distributed image analysis method is to find an unspecified object who appears most frequently in the two cities, Shenzhen and Dongguan. That is, the second server cannot determine a to-be-obtained result object before the distributed image analysis method is completed.
  • the result set sent by the first server includes an index image and an object frequency corresponding to the index image. Therefore, in this embodiment, the first server only needs to upload one image, that is, the index image, to report all information of a result object to the second server, thereby greatly improving transmission efficiency.
  • the index image is an ordinary image file and is compatible with algorithms of different vendors
  • the first server and the second server are compatible with each other even if algorithms of different vendors are installed on the first server and the second server, so that computing power of servers at all levels is fully utilized, thereby improving computing efficiency.
  • a specific manner of obtaining the index image in the result set is not limited. Further, two manners for obtaining the index image are provided in the following: a manner of randomly obtaining the index image from the candidate images and a manner of obtaining the candidate image having the highest confidence as the index image.
  • An Index Image is Randomly Obtained from Candidate Images.
  • the first server obtains all images recording a same result object as candidate images of a result set, and then determines the index image from the candidate images. For ease of understanding, a specific procedure of this case is described below.
  • an exemplary distributed image analysis method according to an embodiment of this application includes the following steps.
  • a first server obtains a plurality of candidate images.
  • the candidate images are images recording the result object.
  • the foregoing Shenzhen server performs image collision analysis on the stored images, to obtain 100 images recording the Object A, and the 100 images are candidate images of the result object—the Object A.
  • the first server determines an index image from the plurality of candidate images.
  • the first server determines the index image from the plurality of candidate images in a manner of randomly selecting at least one of the plurality of candidate images as the index image, or selecting the index image from the candidate images based on a preset rule.
  • the Shenzhen server determines one of the 100 images recording the Object A as an index image for representing the Object A.
  • a determining manner may be randomly selecting the index image from the 100 images, or may be selecting the index image from the 100 images based on a preset rule.
  • the first server determines a quantity of the candidate images as an object frequency.
  • a quantity of candidate images corresponding to a result object is a frequency of being photographed of the result object, that is, the object frequency.
  • the Shenzhen server obtains that a quantity of candidate images recording the Object A is 100, it is determined that an object frequency corresponding to the index image of the Object A is 100.
  • the first server obtains a result set.
  • the result set includes the foregoing index image, and the index image corresponds to the object frequency.
  • a result set obtained by the Shenzhen server includes a result set A, and the result set A includes an index image representing the Object A.
  • An object frequency of the Object A being photographed corresponding to the index image representing the Object A is 100.
  • steps 305 to 309 refer to steps 202 to 206 . Details are not repeated herein.
  • the first server obtains, through the image analysis algorithm, all images recording the same result object as the candidate images, then randomly selects at least one of the candidate images as an index image, and also records a quantity of the candidate images as an object frequency in the index image, to generate a result set.
  • the index image can be quickly obtained from the candidate images, so that computing power is saved.
  • this embodiment of the present invention provides a manner of obtaining top N candidate images having the highest confidences as index images.
  • the first server obtains the candidate image having the highest confidence as the index image, thereby ensuring that the index image has a reliable confidence, and ensuring the accuracy of image analysis.
  • an exemplary distributed image analysis method includes:
  • a first server obtains image quality of each candidate image through an image analysis algorithm.
  • the image analysis algorithm is a well-known algorithm, and details are not repeated herein.
  • image quality of a photographed image can be obtained.
  • the image quality includes light conditions when the image is photographed, for example, whether a photographed object is occluded, and so on. Higher image quality indicates that a photographed object recorded by the image is clearer and more reliable.
  • the Shenzhen server performs the image analysis algorithm on 100 candidate images recording the Object A, to obtain image quality of each candidate image recording the Object A. Because each time the Object A is photographed, photographing conditions, such as an angle of photographing the Object A, photographing light, differ, image quality of each candidate image recording the Object A differs.
  • the first server selects an image having the highest image quality from candidate images as an index image.
  • the first server selects the candidate image having the highest image quality as the index image, and the index image can best represent the result object collectively recorded by the candidate images, thereby helping increase accuracy of second image collision analysis performed by the second server.
  • the Shenzhen server selects an image having the highest image quality as an index image representing the Object A. Because the index image is the candidate image having the highest image quality, and the lighting conditions are the best during photographing, so that accuracy of the Guangdong server in performing the second image collision analysis is ensured.
  • steps 403 to 409 refer to steps 303 to 309 . Details are not repeated herein again.
  • the first server after obtaining the candidate images, obtains the candidate image having the highest confidence as the index image.
  • the method can ensure that the index image is the image having the highest confidence, accurately represents the result object collectively recorded by the candidate images, and is provided for a higher-level server to perform the second image collision analysis.
  • an exemplary distributed image analysis method includes the following steps.
  • a first server receives a target instruction sent by a second server.
  • the target instruction includes a confidence threshold, and/or a target time period, and/or a target camera identifier.
  • the confidence threshold is used by the first server to determine that stored images having a confidence greater than or equal to the confidence threshold record the same result object when the first server performs the image collision analysis.
  • the Shenzhen server receives a target instruction from the Guangdong server.
  • the target instruction includes a confidence threshold of 90%, a target time period of from 0:00 on Dec. 1, 2018 to 24:00 on Dec. 7, 2018, and a target camera identifier of from 1 to 100.
  • the first server performs feature extraction on stored images.
  • the first server performs feature extraction on the stored images based on a target time period and/or a target camera identifier preset in the target instruction, to obtain feature values of the stored images.
  • the target instruction includes the target time period and the target camera identifier.
  • the foregoing Shenzhen server extracts stored images photographed by cameras having a target camera identifier of from 1 to 100 in a time period of from 0:00 on Dec. 1, 2018 to 24:00 on Dec. 7, 2018, and performs feature extraction.
  • the first server performs image collision analysis on feature values of the stored images, to obtain a confidence of each stored image.
  • the first server compares the feature values of the stored images pairwise through image collision analysis, to obtain confidences between all the stored images.
  • the first server obtains stored images having a confidence greater than or equal to the confidence threshold as the candidate images that collectively include the same result object.
  • the confidence threshold included in the target instruction sent by the second server is provided for the first server to perform sifting, and the same result object is recorded between two candidate images having a confidence greater than the confidence threshold.
  • the Shenzhen server determines that the two candidate images collectively record a result object.
  • steps 505 to 511 refer to steps 303 to 309 . Details are not repeated herein.
  • the first server selects images having a confidence greater than the confidence threshold from the locally stored images as the candidate images when executing the image collision algorithm. That is, if a confidence between every two result objects recorded in images is higher than the confidence threshold, the result objects are determined as a same result object. If the target instruction includes a target time period, the first server obtains candidate images photographed during the target time period. If the target instruction includes a target camera identifier, the first server obtains candidate images photographed by a camera that is identified by the target camera identifier. Therefore, sifting conditions of the image collision analysis of the first server are controlled through the target instruction of the second server. In this way, in the distributed image analysis method, the first server and the second server can work cooperatively.
  • the index image and object frequency are set in the result set, for the higher-level server to perform the second image collision analysis on the result set, to find a result object having the highest frequency of being photographed.
  • the user After obtaining the result object having the highest frequency of being photographed, the user is interested in learning all images recording the result object. In this case, because each result object in the result set corresponds to only one index image, this cannot be achieved.
  • an exemplary distributed image analysis method includes the following steps.
  • steps 601 to 605 refer to steps 501 to 505 . Details are not repeated herein.
  • a first server obtains a resource address of the candidate images.
  • the resource address is a resource address of the candidate images stored in the first server, and the resource address may be a URL (Uniform Resource Locator).
  • the resource address further records identification codes IDs of the corresponding candidate images, so that the second server can distinguish different candidate images through the IDs when accessing the candidate images through the resource address.
  • steps 607 to 612 refer to steps 506 to 511 . Details are not repeated herein.
  • the second server accesses the candidate images through the resource address.
  • the second server may access, through the resource address, all candidate images that record the result object having the highest frequency of being photographed, or may access, through the resource address, all candidate images that record each result object.
  • the resource address of the candidate images is added to the result set sent by the first server, so that the second server can access the candidate images stored in the first server through the resource address.
  • the resource address saves more resources in terms of format and does not occupy a transmit resource and a storage resource of the second server while enabling the second server to access all candidate images, so that all candidate images recording the result object of interest are obtained.
  • the foregoing first server is a sub-level server of the second server.
  • the second server triggers, through indication information, the first server to perform, based on conditions indicated by the indication information, a first image collision analysis task and output the result set.
  • the second server performs the second image collision analysis based on the result set to obtain the result object having the highest frequency of being photographed in all the first servers.
  • the first server may be a prefectural server
  • the second server may be a national central server
  • the first server may be a prefectural server
  • the second server may be a district server.
  • the architecture of the foregoing distributed server may be a two-level architecture including a first server and a second server, or may be a multi-level architecture, for example, including a first server, a second server, and a third server.
  • the second server After performing image collision analysis based on result sets sent by a plurality of first servers, the second server obtains the result object having the highest frequency of being photographed in all the first servers connected to the second server. In this case, the second server generates a new result set.
  • the new result set includes an image having the highest confidence of the result object having the highest frequency of being photographed as an index image, and the index image correspondingly has a total frequency of being photographed of the result object.
  • a manner in which the second server generates the new result set is the same as the manner in which the first server generates the result set.
  • the second server sends the new result set to the third server for third image collision analysis.
  • the above distributed image analysis method can effectively reduce a collision frequency of image collision analysis and reduce a quantity of server calculations while maintaining same image identification accuracy, thereby improving calculation efficiency.
  • a higher similarity coefficient indicates that higher performance of the distributed image analysis method compared with that of a conventional centralized image frequency algorithm.
  • the foregoing similarity coefficient refers to a proportion of images recording a same result object in a same result set. For example, if 50 images out of 100 images record a same result object, the similarity coefficient of the result set is 50%.
  • the foregoing centralized image frequency algorithm refers to an algorithm in the prior art that a first server sends all stored images to a second server for unified image collision analysis.
  • a calculation formula for a quantity of calculations is explained. It is assumed that a data volume D in the result set is 10,000 (that is, there are 10,000 images in the result set, and the samples are evenly distributed), a similarity coefficient is n (n images out of 10,000 images record a same result object), and a calculation frequency is m (there are m servers at a same level in the distributed image analysis method, and in the centralized image frequency algorithm, m is equal to 1)
  • a quantity of first servers is m, and a data volume processed by each first server is:
  • a first server When performing image collision analysis, a first server performs pairwise comparison on images, so that a calculation frequency of the pairwise comparison performed by the first server is:
  • the first server performs pairwise comparison on an image A and an image B. After the image A has been compared with the image B once, the image B does not need to be repeatedly compared with the image A. Therefore, an actual quantity of times that the first server performs image comparison is:
  • a second server obtains result sets sent by all first servers and performs second image collision analysis. Therefore, for candidate images that have been obtained through analysis in the first servers, the second server does not repeat the analysis, so that a data volume that the second server needs to process is:
  • the second server When performing image collision analysis, the second server performs pairwise comparison on the images, so that a calculation frequency of the pairwise comparison performed by the second server is:
  • a second server performs pairwise collision on all received images, and a total frequency of the pairwise collision is:
  • the foregoing distributed image analysis system includes a corresponding hardware structure and/or software module for performing each function.
  • modules and algorithms steps may be implemented by hardware or a combination of hardware and computer software. Whether a function is performed by hardware or hardware driven by computer software depends on particular applications and design constraints of the technical solutions.
  • a person of ordinary skill in the art may use different methods to implement the described functions for each particular application, but it should not be considered that the implementation goes beyond the scope of this application.
  • the foregoing distributed image analysis method may be implemented by a physical device, or may be jointly implemented by a plurality of physical devices, or may be implemented by a logical functional module inside a physical device. This is not specifically limited in the embodiments of this application.
  • FIG. 7 is a schematic diagram of a hardware structure of an exemplary distributed image analysis server according to an embodiment of this application.
  • the communications device includes at least one processor 701 , a communications line 702 , a memory 703 , and at least one communications interface 704 .
  • the processor 701 may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling program execution of the solutions of this application.
  • CPU central processing unit
  • ASIC application-specific integrated circuit
  • the communications line 702 may include a path for transmitting information between the foregoing components.
  • the communications interface 704 that uses any apparatus such as a transceiver is configured to communicate with another device or a communications network, for example, the Ethernet, a radio access network (RAN), or a wireless local area network (WLAN).
  • a communications network for example, the Ethernet, a radio access network (RAN), or a wireless local area network (WLAN).
  • RAN radio access network
  • WLAN wireless local area network
  • the memory 703 may be a read-only memory (ROM) or another type of static storage device capable of storing static information and instructions, a random access memory (RAM) or another type of dynamic storage device capable of storing information and instructions, or may be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or another compact disc storage, an optical disc storage (including a compressed optical disc, a laser disc, an optical disc, a digital versatile disc, a Blu-ray optical disc, and the like), a magnetic disk storage medium or another magnetic storage device, or any other medium that is capable of carrying or storing expected program code in a form of instructions or data structures and capable of being accessed by a computer, but is not limited thereto.
  • the memory may exist independently and is connected to the processor through the communications line 702 .
  • the memory may be alternatively integrated with the processor.
  • the memory 703 is configured to store one or more computer-executable instructions for executing the solutions of this application, and the processor 701 controls the execution.
  • the processor 701 is configured to execute the one or more computer-executable instructions stored in the memory 703 , to implement the charging management method provided in the following embodiments of this application.
  • the one or more computer-executable instructions in the embodiments of this application may also be referred to as application program code. This is not specifically limited in the embodiments of this application.
  • the processor 701 may include one or more CPUs, for example, a CPU 0 and a CPU 1 in FIG. 7 .
  • the communications device may include a plurality of processors, for example, the processor 701 and a processor 707 in FIG. 7 .
  • Each of the processors may be a single-core (single-CPU) processor, or may be a multi-core (multi-CPU) processor.
  • the processor herein may be one or more devices, circuits, and/or processing cores configured to process data (for example, one or more computer program instructions).
  • the communications device may further include an output device 705 and an input device 706 .
  • the output device 705 communicates with the processor 701 , and may display information in a plurality of manners.
  • the output device 705 may be a liquid crystal display (LCD), a light emitting diode (LED) display device, a cathode ray tube (CRT) display device, a projector, or the like.
  • the input device 706 communicates with the processor 701 , and may receive an input from a user in a plurality of manners.
  • the input device 706 may be a mouse, a keyboard, a touchscreen device, a sensor device, or the like.
  • the foregoing communications device may be a general-purpose device or a dedicated device.
  • the communications device may be a server, a wireless terminal device, an embedded device, or a device having a structure similar to that in FIG. 7 .
  • a type of the communications device is not limited in the embodiments of this application.
  • the distributed image analysis server may be divided into functional units based on the foregoing method examples.
  • functional modules may be obtained through division based on corresponding functions, or two or more functions may be integrated into one processing unit.
  • the integrated unit may be implemented in a form of hardware, or may be implemented in a form of a software functional unit. It should be noted that division into the units in the embodiments of this application is an example, is merely logical function division and may be other division during actual implementation.
  • FIG. 8 is a schematic structural diagram of an exemplary distributed image analysis system according to an embodiment of the present disclosure.
  • the distributed image analysis system provided in an embodiment of this application includes a first server and a second server.
  • the first server includes:
  • a first obtaining unit 801 configured to obtain a result set through image collision analysis, where the result set includes an index image, the index image records a result object, the index image corresponds to an object frequency, and the object frequency is used to indicate a quantity of times that the result object is photographed;
  • a first sending unit 802 configured to send the result set obtained by the first obtaining unit 801 to the second server, where the result set is used for the second server to perform second image collision analysis.
  • the first obtaining unit 801 is further configured to:
  • candidate images are images that include the result object among images stored in the first server, and determine the index image from the plurality of candidate images
  • the first server further includes a first receiving unit 803 , where the first receiving unit 803 is configured to:
  • the target instruction includes a confidence threshold.
  • the first obtaining unit 801 is configured to:
  • the first receiving unit 803 is further configured to:
  • the target instruction further includes a target time period and/or a target camera identifier.
  • the first obtaining unit 801 is further configured to perform feature extraction on the stored images in the target time period, and/or perform feature extraction on the stored images photographed by a camera that is identified by the target camera identifier.
  • the first obtaining unit 801 is further configured to:
  • the first obtaining unit 801 is further configured to:
  • the result set includes the resource address, where the resource address is used for the second server to access the candidate images through the resource address.
  • the second server includes:
  • a second obtaining unit 804 configured to receive a result set that is separately sent by a plurality of first servers and that is obtained by each first server, where each result set includes an index image, the index image records a result object, the index image corresponds to an object frequency, and the object frequency is used to indicate a quantity of times that the result object is photographed;
  • an analysis unit 805 configured to: perform feature extraction on an index image in each result set, to obtain a feature value of the index image, and perform image collision analysis on extracted feature values of all index images, to obtain a confidence of each index image;
  • an output unit 806 configured to: when determining that a confidence between the index images in the result sets sent by the plurality of first servers is greater than or equal to a preset value, obtain a sum of object frequencies corresponding to the index images.
  • the result set includes a resource address of the candidate images, where the candidate images are images that record the result object among images stored in the first server, and the stored images are all images stored in the first server.
  • the second obtaining unit 804 is further configured to access the candidate images through the resource address.
  • the second server further includes a second sending unit 807 , configured to send a target instruction to the first server, where the target instruction includes a confidence threshold, and a confidence of any candidate image in the first server is greater than or equal to the confidence threshold.
  • a second sending unit 807 configured to send a target instruction to the first server, where the target instruction includes a confidence threshold, and a confidence of any candidate image in the first server is greater than or equal to the confidence threshold.
  • the second sending unit 807 is further configured to send a target instruction to the first server, where the target instruction further includes a target time period and/or a target camera identifier.
  • the target instruction is used for the first server to perform feature extraction on the stored images in the target time period, and/or perform feature extraction on the stored images photographed by a camera that is identified by the target camera identifier.
  • All or some of the foregoing embodiments may be implemented by using software, hardware, firmware, or any combination thereof.
  • software is used to implement the embodiments, all or some of the embodiments may be implemented in a form of a computer program product.
  • the computer program product includes one or more computer instructions.
  • the computer may be a general-purpose computer, a dedicated computer, a computer network, or another programmable apparatus.
  • the computer instructions may be stored in a computer-readable storage medium, or may be transmitted from a computer-readable storage medium to another computer-readable storage medium.
  • the computer instructions may be transmitted from a website, computer, server, or data center to another website, computer, server, or data center in a wired (for example, a coaxial cable, an optical fiber, or a digital subscriber line (DSL)) or wireless (for example, infrared, radio, or microwave) manner.
  • a wired for example, a coaxial cable, an optical fiber, or a digital subscriber line (DSL)
  • wireless for example, infrared, radio, or microwave
  • the computer-readable storage medium may be any usable medium accessible by a computer, or a data storage device, such as a server or a data center, integrating one or more usable media.
  • the usable medium may be a magnetic medium (for example, a floppy disk, a hard disk, or a magnetic tape), an optical medium (for example, a DVD), a semiconductor medium (for example, a solid-state drive (SSD)), or the like.
  • the program may be stored in a computer-readable storage medium.
  • the storage medium may include: a ROM, a RAM, a magnetic disk, or a compact disc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Library & Information Science (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Image Analysis (AREA)
US17/350,205 2018-12-18 2021-06-17 Distributed image analysis method and system, and storage medium Pending US20220138893A9 (en)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
CN201811554821.4A CN111343416B (zh) 2018-12-18 2018-12-18 一种分布式图像分析方法、系统及存储介质
CN201811554821.4 2018-12-18
PCT/CN2019/124967 WO2020125544A1 (zh) 2018-12-18 2019-12-13 一种分布式图像分析方法、系统及存储介质

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/124967 Continuation WO2020125544A1 (zh) 2018-12-18 2019-12-13 一种分布式图像分析方法、系统及存储介质

Publications (2)

Publication Number Publication Date
US20210312587A1 US20210312587A1 (en) 2021-10-07
US20220138893A9 true US20220138893A9 (en) 2022-05-05

Family

ID=71101059

Family Applications (1)

Application Number Title Priority Date Filing Date
US17/350,205 Pending US20220138893A9 (en) 2018-12-18 2021-06-17 Distributed image analysis method and system, and storage medium

Country Status (4)

Country Link
US (1) US20220138893A9 (zh)
EP (1) EP3890312B1 (zh)
CN (2) CN111343416B (zh)
WO (1) WO2020125544A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110442444B (zh) * 2019-06-18 2021-12-10 中国科学院计算机网络信息中心 一种面向海量遥感影像的并行数据访问方法与系统
CN112131424A (zh) * 2020-09-22 2020-12-25 深圳市天维大数据技术有限公司 一种分布式图像分析方法及系统
CN113420183B (zh) * 2021-06-21 2022-12-02 成都神州数码索贝科技有限公司 一种在视频侦查系统中跨视频目标检索及碰撞方法及系统

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7406216B2 (en) * 2004-12-15 2008-07-29 Micron Technology, Inc. Method and apparatus for distributed analyses of images
CN101159603B (zh) * 2007-10-30 2010-06-02 中兴通讯股份有限公司 一种无线网络海量数据存储方法
CN101777064A (zh) * 2009-01-12 2010-07-14 鸿富锦精密工业（深圳）有限公司 图片搜索系统及方法
CN101505412A (zh) * 2009-03-11 2009-08-12 上海伯里春贸易有限公司 宾馆、写字楼和娱乐场所人像采集识别监控系统及方法
CN102638456B (zh) * 2012-03-19 2015-09-23 杭州海康威视数字技术股份有限公司 基于云计算的海量实时视频码流智能分析方法及其系统
US9589181B2 (en) * 2013-02-28 2017-03-07 Hitachi Kokusai Electric Inc. Person search method and device for searching person staying on platform
KR101607934B1 (ko) * 2014-05-14 2016-03-31 숭실대학교산학협력단 전염병 모니터링 방법 및 이를 이용한 시스템, 이를 수행하기 위한 기록매체
AU2014240213B2 (en) * 2014-09-30 2016-12-08 Canon Kabushiki Kaisha System and Method for object re-identification
CN105183917B (zh) * 2015-10-15 2018-06-05 国家电网公司 一种用于多级存储数据的多维分析方法
CN105844634B (zh) * 2016-03-18 2019-04-05 阜阳师范学院 一种多运动目标追踪监测方法
KR20180058019A (ko) * 2016-11-23 2018-05-31 한화에어로스페이스 주식회사 영상 검색 장치, 데이터 저장 방법 및 데이터 저장 장치
CN106971142B (zh) * 2017-02-07 2018-07-17 深圳云天励飞技术有限公司 一种图像处理方法及装置
CN107316016B (zh) * 2017-06-19 2020-06-23 桂林电子科技大学 一种基于Hadoop和监控视频流的车辆轨迹统计方法
CN107862242A (zh) * 2017-09-19 2018-03-30 汉柏科技有限公司 一种基于Map‑Reduce的海量人脸特征比对方法及系统
CN108491543A (zh) * 2018-03-30 2018-09-04 广州汇智通信技术有限公司 图像检索方法、图像存储方法及图像检索系统

Also Published As

Publication number Publication date
CN113505250A (zh) 2021-10-15
US20210312587A1 (en) 2021-10-07
CN111343416A (zh) 2020-06-26
EP3890312A1 (en) 2021-10-06
WO2020125544A1 (zh) 2020-06-25
EP3890312B1 (en) 2024-03-20
CN111343416B (zh) 2021-06-01
EP3890312A4 (en) 2022-02-16

Similar Documents

Publication Publication Date Title
US20220138893A9 (en) Distributed image analysis method and system, and storage medium
WO2020093694A1 (zh) 生成视频分析模型的方法及视频分析系统
US10776665B2 (en) Systems and methods for object detection
CN116188821B (zh) 版权检测方法、系统、电子设备和存储介质
CN108683877B (zh) 基于Spark的分布式海量视频解析系统
US11250250B2 (en) Pedestrian retrieval method and apparatus
US10997459B2 (en) Video content indexing and searching
CN110853033B (zh) 基于帧间相似度的视频检测方法和装置
CN109104620A (zh) 一种短视频推荐方法、装置和可读介质
WO2022001918A1 (zh) 构建预测模型的方法、装置、计算设备和存储介质
US11151386B1 (en) Automated identification and tagging of video content
US20160005171A1 (en) Image Analysis Device, Image Analysis System, and Image Analysis Method
WO2023221713A1 (zh) 图像编码器的训练方法、装置、设备及介质
US20150058579A1 (en) Systems and methods for memory utilization for object detection
US11907291B2 (en) System for integral analysis and management of video data
CN113569657A (zh) 一种行人重识别方法、装置、设备及存储介质
CN116503596A (zh) 图片分割方法、装置、介质和电子设备
CN115273191A (zh) 一种人脸聚档方法、人脸识别方法、装置、设备及介质
CN111191591A (zh) 一种水印检测、视频处理方法和相关设备
Zuo et al. Bandwidth and energy efficient image sharing for situation awareness in disasters
CN114078269A (zh) 一种人脸图像聚类方法、装置、服务器和存储介质
KR20210006429A (ko) 오브젝트 검출 방법, 장치, 장비 및 컴퓨터 판독 가능 저장 매체
EP4340374A1 (en) Picture quality adjustment method and apparatus, and device and medium
JP2016045538A (ja) 情報処理装置、画像判定方法、及びプログラム
CN114155578A (zh) 人像聚类方法、装置、电子设备和存储介质

Legal Events

Date Code Title Description
STPP Information on status: patent application and granting procedure in general

Free format text: DOCKETED NEW CASE - READY FOR EXAMINATION

STPP Information on status: patent application and granting procedure in general

Free format text: NON FINAL ACTION MAILED

STPP Information on status: patent application and granting procedure in general

Free format text: RESPONSE TO NON-FINAL OFFICE ACTION ENTERED AND FORWARDED TO EXAMINER

STPP Information on status: patent application and granting procedure in general

Free format text: FINAL REJECTION MAILED

STPP Information on status: patent application and granting procedure in general

Free format text: ADVISORY ACTION MAILED

STPP Information on status: patent application and granting procedure in general

Free format text: NON FINAL ACTION MAILED