Logistic regression-based queuing anomaly detection method and device

A logistic regression algorithm and logistic regression technology, applied in the direction of instruments, character and pattern recognition, computer parts, etc., can solve the problems of low recognition accuracy, affect the recognition accuracy, and block the head area, so as to improve the recognition accuracy. efficiency, improve work speed, and reduce workload

Active Publication Date: 2017-07-21
WEIHAI POWER SUPPLY COMPANY OF STATE GRID SHANDONG ELECTRIC POWER COMPANY +1
7 Cites 1 Cited by

AI-Extracted Technical Summary

Problems solved by technology

[0002] With the advent of the information age, there are more and more applications of intelligent video surveillance, and the research on video queuing detection is becoming more and more extensive. The traditional head detection based on contour or shape mainly extracts the head characteristics according to the characteristics of the head Contour, by calculating the contour to achieve the purpose of people counting, this method is relatively simple to implement, but the interference is large, and the recognition accuracy is not high
[0...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Abstract

The invention relates to a logistic regression-based queuing anomaly detection method and device. The device comprises a hard disk video recorder module and a format conversion module; the hard disk video recorder module is a device for storing monitoring videos; monitoring videos collected by a business hall are stored in the hard disk video recorder; when queuing detection is performed on different windows or counters in the business hall, the video information of corresponding windows or counters is required to be read from the hard disk video recorder; when the video information is read, user names and passwords are required to be provided; the format conversion module performs format conversion on the videos; the format of the videos stored in the hard disk video recorder is a YUV format; and since the queuing detection is realized based on opencv, the YUV format is required to be converted into a format which is recognizable for the opencv. The logistic regression-based queuing anomaly detection method and device of the invention are superior to traditional head recognition means and can assist in solving the problem of poor universality. According to the method and device of the invention, a large number of head samples are adopted to perform training; an obtained classifier is utilized to recognize heads; and therefore, recognition accuracy can be improved, and interference is little.

Application Domain

Technology Topic

Logistic regressionPassword +2

Image

  • Logistic regression-based queuing anomaly detection method and device
  • Logistic regression-based queuing anomaly detection method and device
  • Logistic regression-based queuing anomaly detection method and device

Examples

  • Experimental program(1)

Example Embodiment

[0048] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0049] figure 2 It is a flowchart of a queuing anomaly detection method based on logistic regression, which mainly includes the following steps:
[0050]Step 101, extract a frame from the collected monitoring video when there is no customer as the background image, and the monitoring video image when the service is normally provided as the image to be judged, extract the region of interest that needs to detect the window, and specify the window to be detected Extract the region of interest from the background image and the image to be judged separately, and use the region function cvRect(int x, int y, int width, int height) to extract the region of interest image according to the position of the region of interest in the image, where x represents The x coordinate of the upper left corner of the region, y represents the y coordinate of the upper left corner of the region, width represents the width of the region, and height represents the height of the region. Output the background image after extracting the region of interest and the current image to be judged.
[0051] Step 102, respectively convert the background image after extracting the region of interest and the image to be judged from an RGB image to a GRAY image, make a difference between the background image after the grayscale conversion and the image to be judged, and use the frame difference method to obtain the image after the difference, Thus, the target to be detected is obtained.
[0052] Step 103, collecting 1000 positive samples and 1957 negative samples, extracting the HOG feature value of each sample, and then using the logistic regression algorithm to train the HOG features of the positive and negative samples. First construct the probability function for detecting human heads where z is the weighting function z=θ of the HOG feature vector of the image 0 +θ 1 x 1 +θ 2 x 2 +…+θ n x n. Then, the unknown parameter θ is obtained through the HOG feature vector of the collected sample data, and z is obtained, so as to obtain the head classifier. Finally, use the logistic regression algorithm to train the head classifier for the target to be detected, identify the head in the area, and use the probability function P(X) to predict the new data. If P(X)>0.5, it is considered that the detected head is a head, otherwise The detected head is not the head, and then count the number of heads, and finally output the counted number of heads.
[0053] Step 104, setting a queuing abnormality judgment threshold, comparing the head count outputted in 103 with the threshold, and judging as queuing abnormality if it exceeds the threshold. Record the abnormal queuing event and the number of queuing people in the local database, store the judgment image in the distributed system, and push the abnormal information to the system administrator for verification. Extract the next frame of image of the monitoring video, and return to step 102 to continue to judge the queuing exception.
[0054] image 3 It is the flow chart of HOG feature extraction, the main steps are as follows:
[0055] Step 201, use the Gamma space to standardize and correct each positive and negative sample image, first convert the sample image into a grayscale image, and then use I(x,y)=I(x,y) gamma Compress the image.
[0056] Step 202, calculate the gradient direction value of the sample image, first use the [-1,0,1] gradient operator to perform convolution operation on the original image, and obtain the gradient component G in the x direction x (x,y) 2 , then use [1,0,-1] T The gradient operator performs convolution operation on the original image to obtain the gradient component G in the y direction y (x,y) 2 , using the formula Calculate the size of the pixel, using the formula Compute the gradient direction.
[0057] Step 203, construct the gradient direction histogram of each cell unit, divide the image into several "cells", each cell is 6*6 pixels, and use the histogram of 9 bins to count the gradient information of each cell , that is, the gradient direction of the cell is divided into 9 direction blocks at 360 degrees.
[0058] Step 204, the normalization of the block gradient histogram, combining each cell into larger, spatially connected blocks (blocks), and the HOG feature of each block is obtained by concatenating the feature vectors of all cells in the block. The characteristic HOG features of all blocks are concatenated to form the HOG feature vector f=(x 1 ,x 2 ,...,x n ).
[0059] figure 1 It is a queuing anomaly detection device based on logistic regression, mainly including:
[0060] Hard disk video recorder module 001: mainly used to store surveillance video. The surveillance video collected in the business hall is stored in the hard disk video recorder. When performing queuing detection on different windows or counters in the business hall, it is necessary to read the information of the corresponding window or counter from the hard disk video recorder. For video information, user name and password need to be provided when reading the video, and then the video stream is input to the format conversion module 002.
[0061] Format replacement module 002: Convert the format of the read video. The video format stored in the hard disk video recorder is stream data in YUV format. The queuing detection is realized in combination with opencv, so it is necessary to convert the stream data in YUV format into one that can be recognized by opencv frame picture.
[0062] Region of interest extraction module 003: extract the window or counter area that needs to be analyzed and detected in the frame picture, mainly use the region function in opencv to extract the region of interest image, and hand over the extracted region of interest to the queuing statistics module 004 for queuing detection and analysis.
[0063] Queuing Statistics Module 004: Analyze and count the queuing situation in the area of ​​interest, collect positive samples and negative samples of the head, extract the HOG feature of the sample, use the logistic regression algorithm for head training, generate a head classifier, and use the head classifier to analyze and identify the area of ​​interest The number of people is counted, and the statistical results are output to the queuing exception monitoring module 005.
[0064] Queue abnormal monitoring module 005: analyze the statistical results output by the queuing statistical module 004, set the queuing abnormal threshold, if the statistical result exceeds the threshold, it is judged as queuing abnormal, record the abnormal event, and store the image in the data storage module, And push the exception information to the system administrator for verification.
[0065] Data storage module 006: a distributed data storage system, mainly used to store the images determined by the abnormal queuing monitoring module 005 to be abnormal in queuing.
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

no PUM

Description & Claims & Application Information

We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Similar technology patents

Video face identification algorithm on basis of multi-instance learning

ActiveCN104778457AImprove recognition accuracyAvoid difficult choicesCharacter and pattern recognitionInstance-based learningCharacteristic space
Owner:JILIN UNIV

Classification and recommendation of technical efficacy words

  • Improve recognition accuracy
  • Less intrusive

Distributed service tracking method, system, storage medium and electronic equipment

InactiveCN108038145AGuaranteed stability and high availabilityLess intrusiveData switching networksSpecial data processing applicationsService processElectronic equipment
Owner:CTRIP TRAVEL NETWORK TECH SHANGHAI0

Recessed ceiling mounted passive infrared detector

ActiveUS20060237651A1Facilitate typeLess intrusiveRadiation pyrometryMaterial analysis by optical meansCeiling mountEngineering
Owner:ADEMCO INC

Electric line fitting wire cross arm

ActiveCN110778203ALess intrusiveImprove stabilityTowersSpatial arrangements/dispositions of cablesUtility poleEngineering
Owner:湖南恒力通电气设备科技有限公司

High-purity aluminium single-layer crystal solidification purifying method

InactiveCN1438339ALess intrusiveImprove removal efficiencyLiquid interfaceAnalysis method
Owner:SHANGHAI JIAO TONG UNIV

Scanner and coaxial and non-coaxial radar systems using same

ActiveCN111856481ALess intrusiveRealize 3D ImagingElectromagnetic wave reradiationOptical switchOptical amplifier
Owner:HANGZHOU XIGHT SEMICON CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products