CN112307830A - Digital retina mass target retrieval and deployment control method - Google Patents

Digital retina mass target retrieval and deployment control method Download PDF

Info

Publication number
CN112307830A
CN112307830A CN201910701648.4A CN201910701648A CN112307830A CN 112307830 A CN112307830 A CN 112307830A CN 201910701648 A CN201910701648 A CN 201910701648A CN 112307830 A CN112307830 A CN 112307830A
Authority
CN
China
Prior art keywords
pedestrian
picture
cluster
data
retrieval
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201910701648.4A
Other languages
Chinese (zh)
Inventor
杨长水
齐峰
魏勇刚
贾惠柱
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.)
Beijing Boya Huishi Intelligent Technology Research Institute Co ltd
Original Assignee
Beijing Boya Huishi Intelligent Technology Research Institute 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 Beijing Boya Huishi Intelligent Technology Research Institute Co ltd filed Critical Beijing Boya Huishi Intelligent Technology Research Institute Co ltd
Priority to CN201910701648.4A priority Critical patent/CN112307830A/en
Publication of CN112307830A publication Critical patent/CN112307830A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/10File systems; File servers
    • G06F16/18File system types
    • G06F16/182Distributed file systems
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/27Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Library & Information Science (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Computing Systems (AREA)
  • Closed-Circuit Television Systems (AREA)

Abstract

A digital retina mass target retrieval and control method is characterized by comprising the following steps of 1, establishing a large data platform network, including a production cluster and a test cluster, and converging by adopting two core switches; for a production cluster, a tera optical interface of a service access switch and a gigabit interface of an out-of-band management access switch are hung below a convergence switch in pairs, wherein the tera is used for carrying service data, and the gigabit is used for carrying out-of-band management data and is selectable; for the test cluster, the aggregation switch is connected with a kilomega interface of a single service access switch; in order to prevent the exchange bandwidth between the nodes from becoming the bottleneck of the system performance, the service network adopts a 10GE network card; the cloud computing technology-based massive target retrieval and deployment and control system for searching people by pictures can replace manual monitoring, improve monitoring efficiency and reduce labor cost.

Description

Digital retina mass target retrieval and deployment control method
Technical Field
The invention discloses a digital retina mass target retrieval and control method, relates to the field of security monitoring and artificial intelligence, and more particularly relates to a digital retina intelligent video monitoring mass target retrieval and control method.
Background
The video monitoring system deployed at present adopts the technical standard H.264 more than ten years ago, has low data compression efficiency, high construction cost and poor application effect, and is mainly expressed as follows:
1) early standards compressed inefficient. Under the condition of ensuring the video quality, the estimated cost of hundreds of millions of cameras deployed in China needs storage cost, and under the condition that the storage space is insufficient in each place, the videos are often over-compressed, so that the quality of a large number of video images is seriously degraded, and key people and vehicles cannot be clearly seen when a case or a safety accident occurs;
2) and the monitoring video is difficult to network. Cameras deployed in many provinces and cities exceed millions, but the cameras adopt old standard codes, so that hundreds of videos can be transmitted in real time under the existing communication bandwidth condition, and most monitoring videos cannot be effectively utilized;
3) highly dense cameras cannot cover the full scene. Although the cameras in partial areas are distributed at high density, the full scene coverage still cannot be carried out, the information shot by the ground cameras in the area covered by the cameras is limited, and meanwhile, the redundancy of video data acquired all weather is high, the global valuable information is difficult to extract, so that huge information waste is caused;
4) and massive videos are difficult to retrieve. The traditional video monitoring system realizes the playback and evidence collection of an event by monitoring personnel looking up and reading a historical video, the manual playback and evidence collection mode of the video has low efficiency, and although the image retrieval technology is rapidly developed, the traditional video monitoring system is applied in the industrial field, particularly the large-scale application in the security field is still in need of solution;
5) video precision analysis is lacking. In the actual combat application of departments such as public security and the like, the video monitoring technology has the problems of slow video retrieval and difficult analysis.
Disclosure of Invention
The invention aims to provide a method for establishing an efficient pedestrian multi-modal feature vector data index table, which ensures real, timely and reliable spatial index of a pedestrian multi-modal feature search engine by utilizing a big data cloud computing platform.
A digital retina mass target retrieval and control method is characterized by comprising the following steps,
step 1, establishing a big data platform network, including a production cluster and a test cluster, and converging by adopting two core switches;
for a production cluster, a tera optical interface of a service access switch and a gigabit interface of an out-of-band management access switch are hung below a convergence switch in pairs, wherein the tera is used for carrying service data, and the gigabit is used for carrying out-of-band management data and is selectable; for the test cluster, the aggregation switch is connected with a kilomega interface of a single service access switch; in order to prevent the exchange bandwidth between the nodes from becoming the bottleneck of the system performance, the service network adopts a 10GE network card;
step 2, after the cluster and each assembly are built, firstly formatting a system storage space where an HDFS file system in the whole cluster is located, and then starting the cluster;
step 4, after starting the Hadoop, starting a ZooKeeper assembly, and finally starting an HBase assembly and a Flink assembly;
step 5, data acquisition: the method is mainly used for traffic main roads, and is particularly suitable for public places with high difficulty in identification and dense personnel; the source approach of the image for identifying and comparing the pedestrian is a pedestrian image library; the pedestrian image library comprises real-time traffic sequence pedestrian images shot by a plurality of high-speed cameras in a traffic network; the traffic sequence pedestrian image comprises shooting time and spatial position information of a camera; the picture to be retrieved can be a picture of a certain pedestrian to be identified, which is extracted from a pedestrian image library and shot by a camera in a public place, or can be a picture of the pedestrian to be identified uploaded through other ways;
step 6, cleaning data: preprocessing the pedestrian picture to be retrieved to remove noise interference, wherein the preprocessing comprises size conversion, cutting, pixel value dithering and normalization according to a preset rule; obtaining a picture meeting preset requirements through preprocessing;
step 7, data distributed cloud storage: storing unstructured data into an HDFS (Hadoop distributed file system), and storing structured data into an Hbase;
step 8, multi-modal feature combined accurate retrieval: the similarity calculation between the pedestrian images is mainly used for grading the pedestrian images according to the similarity of the attributes or the characteristics of the pedestrian images, and judging the approximation degree of the whole content of the pedestrian images according to the grades;
step 9, screening search results: if the picture is not the needed picture of the pedestrian, performing secondary retrieval, tertiary retrieval or more times according to the semantics or the searched picture until the picture needed by control is found;
step 10, controlling the target pedestrian, and issuing a control task to the searched pedestrian picture or feature semantic;
step 11, warning the target pedestrian: if the target pedestrian appears under a certain monitoring point, the system can give an alarm by sound or short messages;
and 12, replaying the target pedestrian track.
The invention has the following beneficial effects:
the invention is designed based on a cloud architecture, makes full use of the super-strong computing capability of a cloud computing platform, deploys various algorithms, realizes the mixing of the algorithms, simultaneously absorbs the advantages of the algorithms, improves the identification and comparison performance of face and pedestrian images under large database capacity, and simultaneously supports semantic retrieval of attributes such as pedestrian age, handbag, lower body clothes color, lower body clothes, hairstyle, hat, backpack, direction, gender, umbrella, upper body color, child holding, upper body clothes and the like. The cloud computing technology-based massive target retrieval and deployment and control system for searching people by pictures can replace manual monitoring, improve monitoring efficiency and reduce labor cost.
Drawings
FIG. 1 is a system framework diagram of the present invention;
FIG. 2 is a flow chart of the present invention;
FIG. 3 is a diagram of a quasi-search concept of the present invention;
FIG. 4 is a vehicle trajectory playback diagram of the present invention;
fig. 5 is a vehicle position playback diagram of the present invention.
Detailed Description
A digital retina mass target retrieval and control method is characterized by comprising the following steps,
step 1, establishing a big data platform network, including a production cluster and a test cluster, and converging by adopting two core switches;
for a production cluster, a tera optical interface of a service access switch and a gigabit interface of an out-of-band management access switch are hung below a convergence switch in pairs, wherein the tera is used for carrying service data, and the gigabit is used for carrying out-of-band management data and is selectable; for the test cluster, the aggregation switch is connected with a kilomega interface of a single service access switch; in order to prevent the exchange bandwidth between the nodes from becoming the bottleneck of the system performance, the service network adopts a 10GE network card;
step 2, after the cluster and each assembly are built, firstly formatting a system storage space where an HDFS file system in the whole cluster is located, and then starting the cluster;
step 4, after starting the Hadoop, starting a ZooKeeper assembly, and finally starting an HBase assembly and a Flink assembly;
step 5, data acquisition: the method is mainly used for traffic main roads, and is particularly suitable for public places with high difficulty in identification and dense personnel; the source approach of the image for identifying and comparing the pedestrian is a pedestrian image library; the pedestrian image library comprises real-time traffic sequence pedestrian images shot by a plurality of high-speed cameras in a traffic network; the traffic sequence pedestrian image comprises shooting time and spatial position information of a camera; the picture to be retrieved can be a picture of a certain pedestrian to be identified, which is extracted from a pedestrian image library and shot by a camera in a public place, or can be a picture of the pedestrian to be identified uploaded through other ways;
step 6, cleaning data: preprocessing the pedestrian picture to be retrieved to remove noise interference, wherein the preprocessing comprises size conversion, cutting, pixel value dithering and normalization according to a preset rule; obtaining a picture meeting preset requirements through preprocessing;
step 7, data distributed cloud storage: storing unstructured data into an HDFS (Hadoop distributed file system), and storing structured data into an Hbase;
step 8, multi-modal feature combined accurate retrieval: the similarity calculation between the pedestrian images is mainly used for grading the pedestrian images according to the similarity of the attributes or the characteristics of the pedestrian images, and judging the approximation degree of the whole content of the pedestrian images according to the grades;
step 9, screening search results: if the picture is not the needed picture of the pedestrian, performing secondary retrieval, tertiary retrieval or more times according to the semantics or the searched picture until the picture needed by control is found;
step 10, controlling the target pedestrian, and issuing a control task to the searched pedestrian picture or feature semantic;
step 11, warning the target pedestrian: if the target pedestrian appears under a certain monitoring point, the system can give an alarm by sound or short messages;
and 12, replaying the target pedestrian track.
In the process of rapid development, the video monitoring industry is continuously developing towards networking, high-definition, intellectualization and diversification. With the deep application of artificial intelligence, cloud computing, big data and unmanned aerial vehicle technology, the diversification of video intelligent analysis becomes the most distinctive feature of a new generation of video monitoring system. An intelligent video monitoring platform (platform for short) which is developed by a digital retina end-to-end system in Beijing university digital video coding and decoding national engineering laboratory and simultaneously supports functions of concentrating transcoding, image analysis, feature retrieval, application display and the like of monitoring videos, the platform integrates a plurality of leading-edge technologies such as visual content analysis, visual feature retrieval, big data analysis, cloud storage, deep learning and the like, develops a plurality of technologies such as parallel multi-channel video concentrating transcoding, human and vehicle visual feature extraction, massive visual big data quick retrieval, double-current remote communication, a software defined camera network, a real-time service application middleware, an online camera and an offline video file, is suitable for efficient storage, quick retrieval and intelligent application of city-level large-scale monitoring videos, and can provide an integral large-scale monitoring video intelligent application solution for users, the method can be widely applied to intelligent video processing of various bayonet and micro-bayonet public security surveillance video scenes accessed by public security organs. Provides an effective technical means for comprehensive management of cities and detection of cases by public security institutions.
With the continuous deepening of the construction of 'skynet projects' and 'safe cities' in various regions, the construction of the video monitoring three-dimensional prevention and control system is continuously improved, and the physical space covered by the video monitoring system is wider and wider. With the rapid and normal scale of front-end equipment, video data generated in each city increases in surge mode every moment. How to process the mass data provides a great challenge for the traditional system to use the technical architecture of the relational database (Oracle, Mysql, SQLServer), and the user requirements cannot be met from the data storage level alone. If business analysis and associated data cross operation are carried out based on the data, collision is carried out by utilizing a technical and tactical method, which is more impossible to be completed for the traditional technical architecture. From the practical perspective, the technical architecture needs to be upgraded and updated so as to meet the challenges brought by big data processing.
We need to think in focus: how to quickly and effectively lock suspected pedestrians and improve the actual combat efficiency in the face of massive pedestrian data? How to do targeted pedestrian screening on suspected pedestrians to optimize police deployment? How can suspected pedestrians be effectively checked and controlled?
Based on the factors, a digital retina mass target retrieval and deployment control method is provided. The system combines three core technologies of video structuralization, pedestrian recognition and big data processing to perform feature extraction and label analysis processing on 'pedestrian' actual combat elements related to urban videos, and combines the structured mass data and the big data processing system together to provide a comprehensive solution for fitting actual combat and social management function targets for intelligent public security.
A digital retina mass target retrieval and deployment control method comprises the following steps:
step 1: acquiring pedestrian data:
step 2: and (4) pedestrian data cloud storage, wherein unstructured data are stored in the HDFS, and structured data are stored in the Hbase.
And step 3: performing combined accurate retrieval on the multi-modal features;
and 4, step 4: the target pedestrian performs deployment control;
and 5: the cloud computing platform carries out real-time computing:
(1) detecting target pedestrian features and matching feature points in the pedestrian images and the checkpoint pedestrian images through an SIFT feature matching algorithm, and screening feature points with high matching degree;
(2) determining a related feature point group of the matched feature points, and determining a candidate matched pedestrian image area in the checkpoint pedestrian image according to the related feature point group;
(3) the Flink cloud computing platform computes the similarity R1 of the target characteristic pedestrian image and the candidate matching pedestrian image region according to the number of the characteristic points in the related characteristic group;
(4) the method comprises the steps that a Flink cloud computing platform calculates the similarity R2 of a target characteristic pedestrian image and a candidate matching pedestrian image area by using a perceptual hash algorithm;
(5) the Flink cloud computing platform calculates gray level color histograms of the target characteristic pedestrian image and the candidate matching pedestrian image region, and calculates similarity R3 of the target characteristic pedestrian image and the candidate matching pedestrian image region according to the gray level color histograms;
(6) weighting the similarity R1, R2 and R3, and carrying out cloud computing on the final similarity R of the target characteristic pedestrian image and the candidate matching pedestrian image region by a Flink cloud computing platform;
(7) and sequentially retrieving the images of the pedestrians at the gate according to the steps, screening the images of the pedestrians at the gate where the candidate matching pedestrian image area with the similarity value larger than the threshold value of 0.6 is located, and sequentially arranging and displaying the images according to the sequence of the similarity values from large to small.
Step 6: alarming according to the control result of the target pedestrian;
and 7: target pedestrian control track playback;
in order to achieve the purpose, the invention provides a digital retina mass target retrieval and deployment control method.
The invention discloses a digital retina mass target retrieval and control method, which comprises the following steps: the method comprises the steps that characteristic information for filtering pedestrians is obtained in advance, wherein the characteristic information comprises the age of the pedestrians, whether a bag is carried or not, the color of lower clothes, the hairstyle, a hat, a backpack, the direction, the gender, an umbrella, the color of upper clothes, whether a child is held or not, the upper clothes and the like; narrowing the identification range in the picture to be retrieved according to the macroscopic characteristic information of the pedestrian, and analyzing the picture to be retrieved to obtain multidimensional characteristic information; carrying out depth feature fusion on the multi-dimensional feature information to obtain global features; and obtaining similar pedestrian images in a pedestrian image library through feature indexes according to the global features, and sequencing according to the similarity of the pedestrian images to obtain a retrieval result. The method and the system use a plurality of feature fusion and deep neural network methods, thereby effectively improving the accuracy of pedestrian search; through pedestrian retrieval based on the cloud computing technology, the search performance of mass data is effectively improved.
The invention discloses a digital retina mass target retrieval and control method, and the organization structure comprises
(1) Hardware layer: the layer mainly provides support for basic environments such as bottom layer server hardware and an operating system of a distributed storage and retrieval system of massive videos based on Hadoop.
(2) Distributed file system layer: the layer mainly adopts an HDFS distributed file system to provide distributed storage and reading functions for upper-layer application on the basis of a hardware layer. The distributed storage, copy storage, load balancing and other strategies adopted by the HDFS ensure high reliability, expandability and high availability of the whole storage platform. The invention adopts HDFS distributed storage video and other big data files, and provides high-throughput streaming reading and writing for large-scale data.
(3) Database layer: the layer is mainly based on the HDFS as a bottom storage mechanism, and adopts HBase database storage facing column storage. HBase provides a real-time and quick access mechanism for mass data. The invention adopts the HBase database to store the video attribute information, the video abstract attribute information, the target image and other massive small files which need to be accessed in real time in a massive video retrieval system, and designs the HBase database according to the requirements of upper application so as to meet different storage and reading requirements.
(4) Distributed computing layer
The layer mainly solves the distributed computing problem of feature extraction and retrieval in massive video retrieval. Flink is an open-source distributed computing framework, and provides a quick computing scheme for data stored in Hadoop. The invention applies a Flink computing framework to perform distributed extraction and retrieval of the characteristics of the target image stored in HBase, thereby greatly improving the processing speed.
(5) Business logic layer
The layer has the main functions of realizing the operation of each service logic in the system, including the operations related to video preprocessing, feature extraction, video retrieval and the like, and realizing the read-write operation with the bottom database. The video preprocessing comprises the steps of performing storage operation, video abstract processing and storage operation, and target image extraction and storage operation on an original video; the feature extraction operation comprises the steps of carrying out Flink-based distributed feature extraction on a target image, and then storing features into a bottom HBase database; the video retrieval comprises the steps of carrying out feature extraction on an image to be retrieved, then adopting Flink distributed matching, and returning a processing result to an upper user interface layer.
(6) User interface layer
The layer mainly has the functions of providing the user to request information for the control conditions of people, vehicles and the like, and displaying the result data of control in a track mode.
The technical scheme of the invention is as follows:
in order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described below with reference to the accompanying drawings by referring to specific examples.
The hardware environment for implementation is: the hardware environment for implementation is: the system server side runs in a Hadoop cluster, and the cluster comprises four servers. The cluster adopts a Master-Slave server architecture, one server is used as a Master node Master, the other three servers are used as Slave nodes Slave, the three servers run in the same local area network, and the hardware environment is as shown in table 1.
TABLE 1
Figure BDA0002150984310000071
The invention uses a FastDFS cluster file system to store picture and video information, an HBase cluster to store structural characteristic information, a flash cluster to search people and vehicles, and a Zookeeper to provide service reliability, and the role distribution is as follows 2:
TABLE 2
Figure BDA0002150984310000072
Figure BDA0002150984310000081
The invention is implemented as follows:
step 1, networking: the project big data platform comprises a production cluster and a test cluster, and two core switches (convergence) are adopted.
For a production cluster, a service access switch (a tera optical interface) and an out-of-band management access switch (a gigabit interface) are hooked under a convergence switch in pairs, wherein tera is used for running service data, and gigabit is used for running out-of-band management data (optional). For the test cluster, a single service access switch (gigabit interface) is connected below the aggregation switch. In order to prevent the exchange bandwidth between the nodes from becoming the bottleneck of the system performance, the service network adopts 10GE network cards.
And 2, after the cluster and each assembly are built, firstly formatting a system storage space where the HDFS file system in the whole cluster is located, and then starting the cluster.
And 4, after starting the Hadoop, starting the ZooKeeper assembly, and finally starting the HBase assembly and starting the Flink assembly.
Step 5, data acquisition: the method is mainly used for traffic main roads, and is particularly suitable for public places with high difficulty in identification and dense personnel; the source approach of the image for identifying and comparing the pedestrian is a pedestrian image library; the pedestrian image library comprises real-time traffic sequence pedestrian images shot by a plurality of high-speed cameras in a traffic network; the traffic sequence pedestrian image comprises shooting time and spatial position information of the camera. The picture to be retrieved can be a picture of a certain pedestrian to be identified, which is shot by a camera in a public place and extracted from a pedestrian image library, or can be a picture of the pedestrian to be identified, which is uploaded through other ways.
Step 6, cleaning data: preprocessing the pedestrian picture to be retrieved to remove noise interference, wherein the preprocessing comprises size conversion, cutting, pixel value dithering and normalization according to a preset rule; and obtaining the picture meeting the preset requirement through preprocessing.
Step 7, data distributed cloud storage: unstructured data are stored in HDFS and structured data are stored in Hbase.
Step 8, multi-modal feature combined accurate retrieval: the similarity calculation between the pedestrian images is mainly used for grading the pedestrian images according to the similarity of the attributes or the characteristics of the pedestrian images, and judging the similarity of the whole content of the pedestrian images according to the grades.
Step 9, screening search results: if the picture is not the needed picture of the pedestrian, the picture can be searched for the second time, the third time or more times according to the semantic meaning or the searched picture until the picture needed by the control is found.
And step 10, controlling the target pedestrian, and issuing a control task to the searched pedestrian picture or feature semantic.
Step 11, warning the target pedestrian: if the target pedestrian appears under a certain monitoring point, the system can generate sound or short message alarm.
And 12, replaying the target pedestrian track.

Claims (1)

1. A digital retina mass target retrieval and control method is characterized by comprising the following steps,
step 1, establishing a big data platform network, including a production cluster and a test cluster, and converging by adopting two core switches;
for a production cluster, a tera optical interface of a service access switch and a gigabit interface of an out-of-band management access switch are hung below a convergence switch in pairs, wherein the tera is used for carrying service data, and the gigabit is used for carrying out-of-band management data and is selectable; for the test cluster, the aggregation switch is connected with a kilomega interface of a single service access switch; in order to prevent the exchange bandwidth between the nodes from becoming the bottleneck of the system performance, the service network adopts a 10GE network card;
step 2, after the cluster and each assembly are built, firstly formatting a system storage space where an HDFS file system in the whole cluster is located, and then starting the cluster;
step 4, after starting the Hadoop, starting a ZooKeeper assembly, and finally starting an HBase assembly and a Flink assembly;
step 5, data acquisition: the method is mainly used for traffic main roads, and is particularly suitable for public places with high difficulty in identification and dense personnel; the source approach of the image for identifying and comparing the pedestrian is a pedestrian image library; the pedestrian image library comprises real-time traffic sequence pedestrian images shot by a plurality of high-speed cameras in a traffic network; the traffic sequence pedestrian image comprises shooting time and spatial position information of a camera; the picture to be retrieved can be a picture of a certain pedestrian to be identified, which is extracted from a pedestrian image library and shot by a camera in a public place, or can be a picture of the pedestrian to be identified uploaded through other ways;
step 6, cleaning data: preprocessing the pedestrian picture to be retrieved to remove noise interference, wherein the preprocessing comprises size conversion, cutting, pixel value dithering and normalization according to a preset rule; obtaining a picture meeting preset requirements through preprocessing;
step 7, data distributed cloud storage: storing unstructured data into an HDFS (Hadoop distributed file system), and storing structured data into an Hbase;
step 8, multi-modal feature combined accurate retrieval: the similarity calculation between the pedestrian images is mainly used for grading the pedestrian images according to the similarity of the attributes or the characteristics of the pedestrian images, and judging the approximation degree of the whole content of the pedestrian images according to the grades;
step 9, screening search results: if the picture is not the needed picture of the pedestrian, performing secondary retrieval, tertiary retrieval or more times according to the semantics or the searched picture until the picture needed by control is found;
step 10, controlling the target pedestrian, and issuing a control task to the searched pedestrian picture or feature semantic;
step 11, warning the target pedestrian: if the target pedestrian appears under a certain monitoring point, the system can give an alarm by sound or short messages;
and 12, replaying the target pedestrian track.
CN201910701648.4A 2019-07-31 2019-07-31 Digital retina mass target retrieval and deployment control method Pending CN112307830A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910701648.4A CN112307830A (en) 2019-07-31 2019-07-31 Digital retina mass target retrieval and deployment control method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910701648.4A CN112307830A (en) 2019-07-31 2019-07-31 Digital retina mass target retrieval and deployment control method

Publications (1)

Publication Number Publication Date
CN112307830A true CN112307830A (en) 2021-02-02

Family

ID=74486297

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910701648.4A Pending CN112307830A (en) 2019-07-31 2019-07-31 Digital retina mass target retrieval and deployment control method

Country Status (1)

Country Link
CN (1) CN112307830A (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102646334A (en) * 2012-04-25 2012-08-22 杭州海康威视数字技术股份有限公司 Method for automatically obtaining evidences of highway traffic incidents and system adopting method
CN102663431A (en) * 2012-04-17 2012-09-12 北京博研新创数码科技有限公司 Image matching calculation method on basis of region weighting
US20180007429A1 (en) * 2015-01-26 2018-01-04 Hangzhou Hikvision Digital Technology Co., Ltd. Intelligent processing method and system for video data
CN107566785A (en) * 2017-08-02 2018-01-09 深圳微品时代网络技术有限公司 A kind of video monitoring system and method towards big data
CN108009489A (en) * 2017-11-29 2018-05-08 合肥寰景信息技术有限公司 Face for mass data is deployed to ensure effective monitoring and control of illegal activities analysis system
CN108875797A (en) * 2018-05-29 2018-11-23 腾讯科技(深圳)有限公司 A kind of method of determining image similarity, photograph album management method and relevant device
CN109299642A (en) * 2018-06-08 2019-02-01 嘉兴弘视智能科技有限公司 Logic based on Identification of Images is deployed to ensure effective monitoring and control of illegal activities early warning system and method

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102663431A (en) * 2012-04-17 2012-09-12 北京博研新创数码科技有限公司 Image matching calculation method on basis of region weighting
CN102646334A (en) * 2012-04-25 2012-08-22 杭州海康威视数字技术股份有限公司 Method for automatically obtaining evidences of highway traffic incidents and system adopting method
US20180007429A1 (en) * 2015-01-26 2018-01-04 Hangzhou Hikvision Digital Technology Co., Ltd. Intelligent processing method and system for video data
CN107566785A (en) * 2017-08-02 2018-01-09 深圳微品时代网络技术有限公司 A kind of video monitoring system and method towards big data
CN108009489A (en) * 2017-11-29 2018-05-08 合肥寰景信息技术有限公司 Face for mass data is deployed to ensure effective monitoring and control of illegal activities analysis system
CN108875797A (en) * 2018-05-29 2018-11-23 腾讯科技(深圳)有限公司 A kind of method of determining image similarity, photograph album management method and relevant device
CN109299642A (en) * 2018-06-08 2019-02-01 嘉兴弘视智能科技有限公司 Logic based on Identification of Images is deployed to ensure effective monitoring and control of illegal activities early warning system and method

Similar Documents

Publication Publication Date Title
Xu et al. Video structured description technology based intelligence analysis of surveillance videos for public security applications
US9251425B2 (en) Object retrieval in video data using complementary detectors
CN105320705B (en) The search method and device of similar vehicle
CN106127114A (en) Intelligent video analysis method
CN107169106B (en) Video retrieval method, device, storage medium and processor
CN105574506A (en) Intelligent face tracking system and method based on depth learning and large-scale clustering
CN103347167A (en) Surveillance video content description method based on fragments
CN102880692A (en) Retrieval-oriented monitoring video semantic description and inspection modeling method
CN105138525B (en) Traffic video processing unit and method and retrieval device and method
CN108182218B (en) Video character recognition method and system based on geographic information system and electronic equipment
CN108363771B (en) Image retrieval method for public security investigation application
KR102028930B1 (en) method of providing categorized video processing for moving objects based on AI learning using moving information of objects
CN102902965B (en) Realize the method that the vedio data structural description of multiple target tracking processes
Hu et al. Parallel spatial-temporal convolutional neural networks for anomaly detection and location in crowded scenes
CN111405249A (en) Monitoring method, monitoring device, server and computer-readable storage medium
Lv et al. Digital twins based VR simulation for accident prevention of intelligent vehicle
CN112766119A (en) Method for accurately identifying strangers and constructing community security based on multi-dimensional face analysis
Pandey et al. KUMBH MELA: a case study for dense crowd counting and modeling
CN109002746A (en) 3D solid fire identification method and system
Tao et al. Smoke vehicle detection based on robust codebook model and robust volume local binary count patterns
Song et al. Crowd counting and abnormal behavior detection via multiscale GAN network combined with deep optical flow
Harakannanavar et al. Robust video summarization algorithm using supervised machine learning
Castellano et al. Crowd counting from unmanned aerial vehicles with fully-convolutional neural networks
CN112306985A (en) Digital retina multi-modal feature combined accurate retrieval method
CN113222109A (en) Internet of things edge algorithm based on multi-source heterogeneous data aggregation technology

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