CN114882429A - Queue counting method and system based on fusion of multiple information characteristics - Google Patents

Queue counting method and system based on fusion of multiple information characteristics Download PDF

Info

Publication number
CN114882429A
CN114882429A CN202210424749.3A CN202210424749A CN114882429A CN 114882429 A CN114882429 A CN 114882429A CN 202210424749 A CN202210424749 A CN 202210424749A CN 114882429 A CN114882429 A CN 114882429A
Authority
CN
China
Prior art keywords
target
matrix
similarity matrix
targets
frame
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
CN202210424749.3A
Other languages
Chinese (zh)
Inventor
陆涛
曹颂
钟星
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.)
Suzhou Super Planet Venture Capital Co ltd
Original Assignee
Suzhou Super Planet Venture Capital 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 Suzhou Super Planet Venture Capital Co ltd filed Critical Suzhou Super Planet Venture Capital Co ltd
Priority to CN202210424749.3A priority Critical patent/CN114882429A/en
Publication of CN114882429A publication Critical patent/CN114882429A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • 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
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • 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
    • G06V10/761—Proximity, similarity or dissimilarity measures
    • 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
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • 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
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace 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
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Human Computer Interaction (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a queue counting method and a queue counting system based on fusion of various information characteristics, wherein the method comprises the following steps: performing monocular visual target detection and tracking; the method comprises the steps of using the position information of each tracked human head in each frame image as an extraction target, and obtaining high-level features of the extraction target, wherein the high-level features comprise track features, position features, time features and speed features of the extraction target; performing a computation of an adjacency matrix E of the decimated objects, comprising: calculating a trajectory similarity matrix E c Position similarity matrix E l Time similarity matrix E t And the velocity similarity matrix E s (ii) a A set of sets of actual queues for the N frame images is computed. The invention integrally solves the technical requirements that the clustering and distinguishing of the same group of targets in the passenger flow and queuing queue are not performed in the prior art, and a more complex and more accurate computer vision algorithm and system are not used, so that the internet platform can be helped to improve the operation efficiency and dynamically know the customer information.

Description

Queue counting method and system based on fusion of multiple information characteristics
Technical Field
The invention relates to a queue counting method and a queue counting system based on fusion of various information characteristics, relates to flow statistics, and belongs to the technical field of artificial intelligence.
Background
With the rise and development of the service industry, various consuming activities are in the layer. In a plurality of gold time periods or peak time periods, a plurality of business places such as hospitals, supermarkets, banks, fast food restaurants and the like have the problems that the dynamic information of customers is difficult to accurately know, the queuing queue of windows is too long, the windows are not opened timely, the number of people in the queue actually cannot be estimated, and the like. The existing monitoring technology plays a great role in helping shops operate, for example, a camera is counted by a person, the counting precision of the camera is very high, but the installation requirement is very high, and the camera is usually only limited to a hall entrance; the AI intelligent camera for face recognition and analysis is also provided, which has a risk of invading the privacy of the client, and the AI intelligent camera can only analyze a single target and cannot distinguish and analyze a group of objects with the same target. If a user wants to check out in a checkout queue of a supermarket, the method can not meet the complex counting requirement because many people in the queue belong to group shopping, such as couples, parents, children, friends and the like.
Therefore, the computer vision algorithm and the system which are more complex and accurate can help the Internet platform to improve the operation efficiency and dynamically know the customer information, and play a great role in the development of the industry.
Patent 1 discloses a passenger flow statistics device, a passenger flow statistics method and a storage medium, wherein the patent publication number is CN109460811A, and specifically discloses that a distance measuring sensor is installed, an infrared sensor is used for counting the number of passengers entering and leaving, the passenger flow statistics device is limited by the sensor, needs to be installed in a specific area, and installation conditions are harsh, and the device is not as strong in universality as a common camera. Patent 2 discloses a passenger flow statistics device based on binocular vision, with publication number CN202058221U, which discloses carrying out passenger flow statistics through a binocular vision system and technology, and is mainly used for passenger flow monitoring of a public transport system; the precision of the binocular matching algorithm will restrict the precision of passenger flow statistics, and in addition, the efficiency of operation and use is difficult to guarantee. Patent 3 discloses a crowd tracking and pedestrian volume counting method and device, with publication number CN104751491A, which uses RGB-D camera to track pedestrians and count passenger flow; the RGB-D can provide distance information of the target, but the detection accuracy of the depth camera is greatly reduced for targets with a distance of more than 4 m. Patent 4 discloses a method for segmenting and tracking the conglutinated crowd based on superpixels and a graph model, wherein the publication number is CN103164858B, the human body form is segmented by using an image segmentation method and the prior knowledge of the human head form, and each region of the human body to be analyzed is converted into an undirected connected graph. However, the super-resolution method is not robust to the scene with changed illumination, and is easy to segment errors.
None of the prior art, including the above patents, clusters and differentiates the same group of objects in the passenger flow, queue.
Disclosure of Invention
The invention aims to overcome the technical defects in the prior art, solve the technical problems and provide a queue counting method and system based on fusion of various information characteristics. A group object refers to a group of people who belong together in a queue.
The invention specifically adopts the following technical scheme: a queue counting method based on fusion of multiple information characteristics comprises the following steps:
step SS 1: performing monocular visual target detection and tracking, comprising: acquiring continuous N frames of images, detecting the head and shoulder positions of each frame of image by using a convolutional neural network, tracking multiple targets by using a Deepsort tracking algorithm, acquiring the position information of each head in each frame of image as an extraction target, and generating a target sequence T;
step SS 2: performing behavior analysis on each sequence in the target sequence T to obtain high-level features of the extracted target, wherein the high-level features comprise track features, position features, time features and speed features of the extracted target;
step SS 3: performing a computation of an adjacency matrix E of the decimated objects, comprising: calculating a trajectory similarity matrix E c Position similarity matrix E l Time similarity matrix E t And a velocity similarity matrix E s And E is calculated in the following manner,
Figure BDA0003609220180000031
wherein E ═ { E ═ E ij Is a normalized n × n symmetric matrix representing the degree of association between the various extracted targets, i.e. the probability of belonging to the same group of targets, e ij The larger the probability that object i and object j belong to the same group of objects, n the number of detected objects, w c Weights representing a trajectory similarity matrix; w is a l Weights representing the location similarity matrix; w is a t Weights representing a temporal similarity matrix; w is a s Weights representing a velocity similarity matrix;
step SS 4: performing a set of sets of actual queues for calculating N frames of images, comprising: and calculating the Laplacian matrix L of the adjacent matrix E of the extracted target, obtaining the optimal clustering decomposition by using a graph cut algorithm, and finally outputting the actual queued group number set.
As a preferred embodiment, step SS1 specifically includes:
the detecting the human head and shoulder position in the picture by using the method based on the convolutional neural network specifically comprises the following steps: using a 19-layer convolutional neural network in each frame image for target detection, detecting the part from the shoulder to the top of the head of the human head, and outputting target frame information l (u) p ,v p ,w p ,h p ) Wherein (u) p ,v p ) Is the coordinate of the center point, w p And h p Respectively representing the width and height of the target frame;
the tracking the detected multiple targets by using the deep sort tracking algorithm specifically comprises the following steps: finding a matching frame of the next frame aiming at the target detected by each frame of image, if the matching frame cannot be found, taking the current frame as a starting point to be counted into the track, and finally outputting a target sequence detected in the N frames of images as follows:
T＝{t 1 ,t 2 ,t i ,...,t n }，
t i ＝{l s ,l s+1 ,...,l e },1≤s≤e≤N；
where N represents the number of detected targets and N represents the number of frames.
As a preferred embodiment, step SS2 specifically includes: for the ith target T in the target sequence T detected in the N frames of images i The following high-level features are calculated:
trajectory feature A for characterizing the motion of an object over time c ，A c ＝{p i }，
Figure BDA0003609220180000041
Position feature B characterizing the position of the target c ，B c Representing the track characteristics A c The first coordinate of (1);
temporal characteristics D c ，D c Is a number between 1 and N, indicating that the object appears in the picture starting at the frame number;
speed characteristic S c According to the track characteristics A c Calculating a number, see formula (2):
Figure BDA0003609220180000042
wherein, omega represents the Euclidean distance,
Figure BDA0003609220180000043
is the coordinates of the ith object in the object sequence T.
As a preferred embodiment, step SS3 specifically includes: adjacency matrix E is formed by trajectory similarity matrix E c A position similarity matrix E l Time similarity matrix E t And velocity similarity matrix E s The weighting results, wherein:
trajectory similarity matrix E c ＝{1-a ij }，a ij The sum of the squared differences of the track characteristic of any target i and the track characteristic of a target j is normalized to 0 to 1, and the probability of belonging to the same group is higher when the track difference between the two targets is smaller;
position similarity matrix E l ＝{1-l ij }，l ij The Euclidean distance is calculated by the position characteristic value of any target i and the position characteristic value of a target j, the normalization is from 0 to 1, and the closer the positions of the two targets are, the higher the probability of belonging to the same group is;
time similarity matrix E t ＝{1-t ij }，t ij Is the time characteristic value of an arbitrary object i and the time characteristic value of an object j 1 The closer the time of occurrence of the two targets is, the greater the probability of belonging to the same group is;
velocity similarity matrix E s ＝{1-s ij }，s ij L being the speed of an arbitrary target i and the speed of a target j 1 Distance, normalized to 0 to 1, two target speedsThe closer together, the greater the probability of belonging to the same group.
As a preferred embodiment, step SS4 specifically includes:
step SS 41: taking the adjacent matrix E as a relation matrix of the relation between the standard targets to obtain a degree information matrix D, which is shown in a formula (3); the degree information matrix D is combined with the adjacent matrix E to obtain a Laplacian matrix L, which is a symmetric matrix, according to a formula (4);
Figure BDA0003609220180000044
L＝D -1/2 (D-E)D -1/2 formula (4)
Step SS 42: carrying out SVD on the Laplacian matrix L to obtain a characteristic value lambda and a characteristic vector v; calculating the eigenvector z ═ D 1/2 v, classifying the samples i corresponding to the components smaller than 0 in the converted eigenvector z into one class, recalculating the energy matrix E 'between the samples i' and the Laplacian matrix L 'and the eigenvector z' for the samples i 'corresponding to the components larger than 0, and classifying the corresponding samples with the components smaller than 0 in the eigenvector z' into a new class, so that the recursion is ended until the components larger than 0 do not exist in the eigenvector z 'or the dimension of the eigenvector z' is 1 according to the recursion mode, and forming a new target group C by the classified targets each time (C) 1 ,c 2 ,...,c k ) Finally, the population number k of the whole queue is obtained.
The invention also provides a queue counting system based on the fusion of various information characteristics, which comprises the following components:
the target detection and tracking module specifically executes: performing monocular visual target detection and tracking, comprising: acquiring continuous N frames of images, detecting the positions of the heads and shoulders of the people in the images of each frame of image by using a convolutional neural network, tracking multiple targets by using a Deepsort tracking algorithm, acquiring position information of each head in each frame of image as an extraction target, and generating a target sequence T;
and the target feature extraction module specifically executes: performing behavior analysis on each sequence in the target sequence T to obtain high-level features of the extracted target, wherein the high-level features comprise track features, position features, time features and speed features of the extracted target;
the module for calculating the adjacency matrix specifically executes: performing a computation of an adjacency matrix E of the decimated objects, comprising: calculating a trajectory similarity matrix E c Position similarity matrix E l Time similarity matrix E t And the velocity similarity matrix E s And E is calculated in the following manner,
Figure BDA0003609220180000051
wherein E ═ E { E ═ E ij Is a normalized n × n symmetric matrix representing the degree of association between the various extracted targets, i.e. the probability of belonging to the same group of targets, e ij The larger the probability that object i and object j belong to the same group of objects, n the number of detected objects, w c Weights representing a trajectory similarity matrix; w is a l Weights representing the location similarity matrix; w is a t Weights representing a temporal similarity matrix; w is a s Weights representing a velocity similarity matrix;
and a module for calculating actual group number: performing a set of sets of actual queues for calculating N frames of images, comprising: and calculating the Laplacian matrix L of the adjacent matrix E of the extracted target, obtaining the optimal clustering decomposition by using a graph cut algorithm, and finally outputting the actual queued group number set.
As a preferred embodiment, the target detecting and tracking module specifically includes:
the detecting the human head and shoulder position in the picture by using the method based on the convolutional neural network specifically comprises the following steps: using a 19-layer convolutional neural network in each frame image for target detection, detecting the part from the shoulder to the top of the head of the human head, and outputting target frame information l (u) p ,v p ,w p ,h p ) Wherein (u) p ,v p ) Is the coordinate of the center point, w p And h p Respectively representing the width and height of the target frame;
the tracking the detected multiple targets by using the deep sort tracking algorithm specifically comprises the following steps: finding a matching frame of the next frame aiming at the target detected by each frame of image, if the matching frame cannot be found, taking the current frame as a starting point to be counted into the track, and finally outputting a target sequence detected in the N frames of images as follows:
T＝{t 1 ,t 2 ,t i ,...,t n }，
t i ＝{l s ,l s+1 ,...,l e },1≤s≤e≤N；
where N represents the number of detected targets and N represents the number of frames.
As a preferred embodiment, the target feature extraction module specifically includes: for the ith target T in the target sequence T detected in the N frames of images i The following high-level features are calculated:
trajectory feature A for characterizing the motion of an object over time c ，A c ＝{p i }，
Figure BDA0003609220180000061
Position feature B characterizing the position of the target c ，B c Representing the track characteristics A c The first coordinate of (1);
temporal characteristics D c ，D c Is a number between 1 and N, indicating that the object appears in the picture starting at the frame number;
speed characteristic S c According to the track characteristics A c Calculating a number, see formula (2):
Figure BDA0003609220180000062
wherein, omega represents the Euclidean distance,
Figure BDA0003609220180000063
is the coordinates of the ith object in the object sequence T.
As a preferred embodiment, the calculate adjacency matrix moduleThe method specifically comprises the following steps: adjacency matrix E is formed by trajectory similarity matrix E c A position similarity matrix E l Time similarity matrix E t And velocity similarity matrix E s The weighting results, wherein:
trajectory similarity matrix E c ＝{1-a ij }，a ij The sum of the squared differences of the track characteristic of any target i and the track characteristic of a target j is normalized to 0 to 1, and the probability of belonging to the same group is higher as the track difference between the two targets is smaller;
position similarity matrix E l ＝{1-l ij }，l ij The Euclidean distance is calculated by the position characteristic value of any target i and the position characteristic value of a target j, the normalization is from 0 to 1, and the probability of belonging to the same group is higher based on the closer the two occurring positions;
time similarity matrix E t ＝{1-t ij }，t ij L being the time characteristic value of an arbitrary object i and the time characteristic value of an object j 1 Distance, based on the closer the time of occurrence of two targets, the greater the probability of belonging to the same group;
velocity similarity matrix E s ＝{1-s ij }，s ij L being the speed of an arbitrary target i and the speed of a target j 1 The distance, normalized to 0 to 1, is based on the closer the two target velocities are, the greater the probability of belonging to the same group.
As a preferred embodiment, the module for calculating the actual number of groups specifically includes: taking the adjacent matrix E as a relation matrix of the relation between the standard targets to obtain a degree information matrix D, which is shown in a formula (3); the degree information matrix D is combined with the adjacent matrix E to obtain a Laplacian matrix L, which is a symmetric matrix, according to a formula (4);
Figure BDA0003609220180000071
L＝D -1/2 (D-E)D -1/2 formula (4)
Carrying out SVD on the Laplacian matrix L to obtain a characteristic value lambda and a characteristic vector v; calculating a feature vector z ═D 1/ 2 v, classifying the samples i corresponding to the components smaller than 0 in the converted eigenvector z into one class, recalculating the energy matrix E 'between the samples i' and the Laplacian matrix L 'and the eigenvector z' for the samples i 'corresponding to the components larger than 0, and classifying the corresponding samples with the components smaller than 0 in the eigenvector z' into a new class, so that the recursion is ended until the components larger than 0 do not exist in the eigenvector z 'or the dimension of the eigenvector z' is 1 according to the recursion mode, and forming a new target group C by the classified targets each time (C) 1 ,c 2 ,...,c k ) Finally, the population number k of the whole queue is obtained.
The invention achieves the following beneficial effects: aiming at solving the technical requirement of solving the problem that the prior art does not perform clustering and distinguishing on the same group of targets in passenger flow and queuing queues, the invention uses a more complex and more accurate computer vision algorithm and system to help an internet platform to improve the operation efficiency, dynamically know the customer information and execute the target detection and tracking of monocular vision, and comprises the following steps: acquiring continuous N frames of images, detecting the positions of the head and the shoulders of the people in the frame of each frame of image by using a convolutional neural network-based method, and tracking the detected multiple targets by using a Deepsort tracking algorithm to obtain a target sequence T; extracting high-level features of each target in the T, wherein the high-level features comprise track features, position features, time features and speed features of the extracted targets; performing a computation of an adjacency matrix E of the decimated objects, comprising: calculating a trajectory similarity matrix E c Position similarity matrix E l Time similarity matrix E t And the velocity similarity matrix E s Performing a set of sets of actual queues for calculating the N frames of images, comprising: calculating a Laplacian matrix L of the adjacent matrix E of the extracted target, obtaining optimal clustering decomposition by using a graph cut algorithm, finally outputting a group number set which is actually queued, carrying out passenger flow statistics and counting by using a computer vision technology, and having great gain for the video monitoring industry; secondly, the invention firstly proposes that the method of feature extraction, target transformation and graph theory is used for decomposing and clustering the targets in the picture, the actual queue group number is calculated, and the method plays an important role in a plurality of actual scenesThe method has the advantages of solving the technical blind spot that the clustering and distinguishing of the same group of targets in the passenger flow and queuing queue are not performed in the prior art on the whole.
Drawings
FIG. 1 is a flow chart of a queue counting method based on fusion of multiple information characteristics according to the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
Example 1: as shown in fig. 1, the present invention provides a queue counting method based on multiple information feature fusion, which includes the following steps:
step SS 1: performing monocular visual target detection and tracking, comprising: designing a buffer queue T to store N frames of continuous images for subsequent detection, tracking, feature extraction and adjacency matrix calculation, detecting the head and shoulder positions of each frame of image by using a convolutional neural network-based method, tracking multiple targets by using a Deepsort tracking algorithm, and acquiring the position information of each head in each frame of image as a target sequence T detected in the N frames of image; the size of N is set according to different scenes, and is generally set between 150 and 250;
step SS 2: extracting high-level features of the target by utilizing the position information of each tracked human head in each frame image, wherein the high-level features comprise track features, position features, time features and speed features of the extracted target;
step SS 3: performing a computation of an adjacency matrix E of the decimated objects, comprising: calculating a trajectory similarity matrix E c Position similarity matrix E l Time similarity matrix E t And the velocity similarity matrix E s And E is calculated in the following manner,
Figure BDA0003609220180000091
wherein E ═ { E ═ E ij Is oneA normalized n × n symmetric matrix characterizing the degree of association between the various extracted targets, i.e. the probability of belonging to the same group of targets, e ij The larger the probability that object i and object j belong to the same group of objects, n the number of detected objects, w c Weights representing a trajectory similarity matrix; w is a l Weights representing the location similarity matrix; w is a t Weights representing a temporal similarity matrix; w is a s Weights representing a velocity similarity matrix;
step SS 4: performing a set of sets of actual queues for calculating N frames of images, comprising: and calculating the Laplacian matrix L of the adjacent matrix E of the extracted target, obtaining the optimal clustering decomposition by using a graph cut algorithm, and finally outputting the actual queued group number set.
As a preferred embodiment, step SS1 specifically includes:
the detecting the human head and shoulder position in the picture by using the method based on the convolutional neural network specifically comprises the following steps: in each frame of image, a 19-layer convolutional neural network is used for target detection, only the part from the shoulder to the top of the head of a person is detected, and target frame information l (u) is output p ,v p ,w p ,h p ) Wherein (u) p ,v p ) Is the coordinate of the center point, w p And h p Respectively representing the width and height of the target frame;
the tracking the detected multiple targets by using the deep sort tracking algorithm specifically comprises the following steps: finding a matching frame of the next frame aiming at the target detected by each frame of image, if the matching frame cannot be found, taking the current frame as a starting point to be counted into the track, and finally outputting a target sequence detected in the N frames of images as follows:
T＝{t 1 ,t 2 ,t i ,...,t n }，
t i ＝{l s ,l s+1 ,...,l e },1≤s≤e≤N；
where N represents the number of detected targets and N represents the number of frames.
As a preferred embodiment, step SS2 specifically includes: for the ith target T in the target sequence T detected in the N frames of images i The following high-level features are calculated:
trajectory feature A for characterizing the motion of an object over time c ，A c ＝{p i }，
Figure BDA0003609220180000092
Any detected object must leave a track in each frame of image;
position feature B characterizing the position of the target c ，B c Representing the track characteristics A c The first coordinate of (1);
temporal characteristics D c ，D c Is a number between 1 and N, indicating that the object appears in the picture starting at the frame number;
speed characteristic S c According to the track characteristics A c Calculating a number, see formula (2):
Figure BDA0003609220180000101
wherein, omega represents the Euclidean distance,
Figure BDA0003609220180000102
is the coordinates of the ith object in the object sequence T.
As a preferred embodiment, step SS3 specifically includes: adjacency matrix E is formed by trajectory similarity matrix E c A position similarity matrix E l Time similarity matrix E t And velocity similarity matrix E s The weighting results, wherein:
trajectory similarity matrix E c ＝{1-a ij }，a ij The sum of the squared differences of the track characteristic of any target i and the track characteristic of a target j is normalized to be 0 to 1, because the smaller the track difference between the two targets is, the higher the probability of belonging to the same group is;
position similarity matrix E l ＝{1-l ij }，l ij Is the Euclidean distance between the position characteristic value of any object i and the position characteristic value of object j, normalized to 0 to 1, because of the two occurring positionsThe closer the more closely, the greater the probability of belonging to the same group;
time similarity matrix E t ＝{1-t ij }，t ij Is the time characteristic value of an arbitrary object i and the time characteristic value of an object j 1 Distance, because the closer the time two objects appear, the greater the probability of belonging to the same group;
velocity similarity matrix E s ＝{1-s ij }，s ij L being the speed of an arbitrary target i and the speed of a target j 1 The distance, normalized to 0 to 1, is because the closer the two target velocities are, the greater the probability of belonging to the same group.
As a preferred embodiment, step SS4 specifically includes:
step SS 41: taking the adjacent matrix E as a relation matrix of the relation between the standard targets to obtain a degree information matrix D, which is shown in a formula (3); the degree information matrix D is combined with the adjacent matrix E to obtain a Laplacian matrix L, which is a symmetric matrix, according to a formula (4);
Figure BDA0003609220180000103
L＝D -1/2 (D-E)D -1/2 formula (4)
Step SS 42: carrying out SVD on the Laplacian matrix L to obtain a characteristic value lambda and a characteristic vector v; calculating the eigenvector z ═ D 1/2 v, classifying the samples i corresponding to the components smaller than 0 in the converted eigenvector z into one class, recalculating the energy matrix E 'between the samples i' and the Laplacian matrix L 'and the eigenvector z' for the samples i 'corresponding to the components larger than 0, and classifying the corresponding samples with the components smaller than 0 in the eigenvector z' into a new class, so that the recursion is ended until the components larger than 0 do not exist in the eigenvector z 'or the dimension of the eigenvector z' is 1 according to the recursion mode, and forming a new target group C by the classified targets each time (C) 1 ,c 2 ,...,c k ) Finally, the population number k of the whole queue is obtained.
The invention also provides a queue counting system based on fusion of various information characteristics, which comprises the following steps:
the target detection and tracking module specifically executes: performing monocular visual target detection and tracking, comprising: acquiring N frames of images, detecting the positions of the heads and shoulders of the people in the images of each frame of image by using a convolutional neural network-based method, tracking the detected multiple targets by using a Deepsort tracking algorithm, and acquiring the position information of each head in each frame of image as a target sequence T detected in the N frames of images;
and the target feature extraction module specifically executes: the method comprises the steps of using the position information of each tracked human head in each frame image as an extraction target, and obtaining high-level features of the extraction target, wherein the high-level features comprise track features, position features, time features and speed features of the extraction target;
the module for calculating the adjacency matrix specifically executes: performing a computation of an adjacency matrix E of the decimated objects, comprising: calculating a trajectory similarity matrix E c Position similarity matrix E l Time similarity matrix E t And the velocity similarity matrix E s And E is calculated in the following manner,
Figure BDA0003609220180000111
wherein E ═ { E ═ E ij Is a normalized n × n symmetric matrix representing the degree of association between the various extracted targets, i.e. the probability of belonging to the same group of targets, e ij The larger the probability that object i and object j belong to the same group of objects, n the number of detected objects, w c Weights representing a trajectory similarity matrix; w is a l Weights representing the location similarity matrix; w is a t Weights representing a temporal similarity matrix; w is a s Weights representing a velocity similarity matrix;
and a module for calculating actual group number: performing a set of sets of actual queues for calculating N frames of images, comprising: and calculating the Laplacian matrix L of the adjacent matrix E of the extracted target, obtaining the optimal clustering decomposition by using a graph cut algorithm, and finally outputting the actual queued group number set.
As a preferred embodiment, the target detecting and tracking module specifically includes:
the detecting the human head and shoulder position in the picture by using the method based on the convolutional neural network specifically comprises the following steps: using a 19-layer convolutional neural network in each frame image for target detection, detecting the part from the shoulder to the top of the head of the human head, and outputting target frame information l (u) p ,v p ,w p ,h p ) Wherein (u) p ,v p ) Is the coordinate of the center point, w p And h p Respectively representing the width and height of the target frame;
the tracking the detected multiple targets by using the deep sort tracking algorithm specifically comprises the following steps: finding a matching frame of the next frame aiming at the target detected by each frame of image, if the matching frame cannot be found, taking the current frame as a starting point to be counted into the track, and finally outputting a target sequence detected in the N frames of images as follows:
T＝{t 1 ,t 2 ,t i ,...,t n }，
t i ＝{l s ,l s+1 ,...,l e },1≤s≤e≤N；
where N represents the number of detected targets and N represents the number of frames.
As a preferred embodiment, the target feature extraction module specifically includes: for the ith target T in the target sequence T detected in the N frames of images i The following high-level features are calculated:
trajectory feature A for characterizing the motion of an object over time c ，A c ＝{p i }，
Figure BDA0003609220180000121
Position feature B characterizing the position of the target c ，B c Representing the track characteristics A c The first coordinate of (1);
temporal characteristics D c ，D c Is a number between 1 and N, indicating that the object appears in the picture starting at the frame number;
speed characteristic S c According to the track characteristics A c Calculating a number, see formula (2):
Figure BDA0003609220180000122
wherein, omega represents the Euclidean distance,
Figure BDA0003609220180000123
is the coordinates of the ith object in the object sequence T.
As a preferred embodiment, the module for calculating an adjacency matrix specifically includes: adjacency matrix E is formed by trajectory similarity matrix E c A position similarity matrix E l Time similarity matrix E t And velocity similarity matrix E s The weighting results, wherein:
trajectory similarity matrix E c ＝{1-a ij }，a ij The sum of the squared differences of the track characteristic of any target i and the track characteristic of a target j is normalized to 0 to 1, and the probability of belonging to the same group is higher when the track difference between the two targets is smaller;
position similarity matrix E l ＝{1-l ij }，l ij The Euclidean distance is calculated by the position characteristic value of any target i and the position characteristic value of a target j, the normalization is from 0 to 1, and the closer the positions of the two targets are, the higher the probability of belonging to the same group is;
time similarity matrix E t ＝{1-t ij }，t ij Is the time characteristic value of an arbitrary object i and the time characteristic value of an object j 1 The closer the time of occurrence of the two targets is, the greater the probability of belonging to the same group is;
velocity similarity matrix E s ＝{1-s ij }，s ij L being the speed of an arbitrary target i and the speed of a target j 1 The closer the two target velocities are, the greater the probability of belonging to the same group, normalized to 0 to 1.
As a preferred embodiment, the module for calculating the actual number of groups specifically includes: taking the adjacent matrix E as a relation matrix of the relation between the standard targets to obtain a degree information matrix D, which is shown in a formula (3); the degree information matrix D is combined with the adjacent matrix E to obtain a Laplacian matrix L, which is a symmetric matrix, according to a formula (4);
Figure BDA0003609220180000131
L＝D -1/2 (D-E)D -1/2 formula (4)
Carrying out SVD on the Laplacian matrix L to obtain a characteristic value lambda and a characteristic vector v; calculating the eigenvector z ═ D 1/ 2 v, classifying the samples i corresponding to the components smaller than 0 in the converted eigenvector z into one class, recalculating the energy matrix E 'between the samples i' and the Laplacian matrix L 'and the eigenvector z' for the samples i 'corresponding to the components larger than 0, and classifying the corresponding samples with the components smaller than 0 in the eigenvector z' into a new class, so that the recursion is ended until the components larger than 0 do not exist in the eigenvector z 'or the dimension of the eigenvector z' is 1 according to the recursion mode, and forming a new target group C by the classified targets each time (C) 1 ,c 2 ,...,c k ) Finally, the population number k of the whole queue is obtained.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.

Claims (10)

1. A queue counting method based on fusion of multiple information characteristics is characterized by comprising the following steps:
step SS 1: performing monocular visual target detection and tracking, comprising: acquiring continuous N frames of images, detecting the positions of the heads and shoulders of the people in the images of each frame of image by using a convolutional neural network, tracking multiple targets by using a Deepsort tracking algorithm, acquiring position information of each head in each frame of image as an extraction target, and generating a target sequence T;
step SS 2: performing behavior analysis on each sequence in the target sequence T to obtain high-level features of the extracted target, wherein the high-level features comprise track features, position features, time features and speed features of the extracted target;
step SS 3: performing a computation of an adjacency matrix E of the decimated objects, comprising: calculating a trajectory similarity matrix E c Position similarity matrix E l Time similarity matrix E t And the velocity similarity matrix E s And E is calculated in the following manner,
Figure FDA0003609220170000011
wherein E ═ { E ═ E ij Is a normalized n × n symmetric matrix representing the degree of association between the various extracted targets, i.e. the probability of belonging to the same group of targets, e ij The larger the probability that object i and object j belong to the same group of objects, n the number of detected objects, w c Weights representing a trajectory similarity matrix; w is a l Weights representing the location similarity matrix; w is a t Weights representing a temporal similarity matrix; w is a s Weights representing a velocity similarity matrix;
step SS 4: performing a set of sets of actual queues for calculating N frames of images, comprising: and calculating the Laplacian matrix L of the adjacent matrix E of the extracted target, obtaining the optimal clustering decomposition by using a graph cut algorithm, and finally outputting the actual queued group number set.
2. The method according to claim 1, wherein step SS1 specifically comprises:
the detecting the human head and shoulder position in the picture by using the method based on the convolutional neural network specifically comprises the following steps: using a 19-layer convolutional neural network in each frame image for target detection, and detecting the position from the shoulder to the top of the headA section for outputting the target frame information l (u) p ,v p ,w p ,h p ) Wherein (u) p ,v p ) Is the coordinate of the center point, w p And h p Respectively representing the width and height of the target frame;
the tracking the detected multiple targets by using the deep sort tracking algorithm specifically comprises the following steps: finding a matching frame of the next frame aiming at the target detected by each frame of image, if the matching frame cannot be found, taking the current frame as a starting point to be counted into the track, and finally outputting a target sequence detected in the N frames of images as follows:
T＝{t 1 ,t 2 ,t i ,...,t n }，
t i ＝{l s ,l s+1 ,...,l e },1≤s≤e≤N；
where N represents the number of detected targets and N represents the number of frames.
3. The method according to claim 1, wherein step SS2 specifically comprises: for the ith target T in the target sequence T detected in the N frames of images i The following high-level features are calculated:
trajectory feature A for characterizing the motion of an object over time c ，A c ＝{p i }，
Figure FDA0003609220170000023
Position feature B characterizing the position of the target c ，B c Representing the track characteristics A c The first coordinate of (1);
temporal characteristics D c ，D c Is a number between 1 and N, indicating that the object appears in the picture starting at the frame number;
speed characteristic S c According to the track characteristics A c Calculating a number, see formula (2):
Figure FDA0003609220170000021
wherein, omega represents the Euclidean distance,
Figure FDA0003609220170000022
is the coordinates of the ith object in the object sequence T.
4. The method according to claim 1, wherein step SS3 specifically comprises: adjacency matrix E is formed by trajectory similarity matrix E c A position similarity matrix E l Time similarity matrix E t And velocity similarity matrix E s The weighting results, wherein:
trajectory similarity matrix E c ＝{1-a ij }，a ij The sum of the squared differences of the track characteristic of any target i and the track characteristic of a target j is normalized to 0 to 1, and the probability of belonging to the same group is higher when the track difference between the two targets is smaller;
position similarity matrix E l ＝{1-l ij }，l ij The Euclidean distance is calculated by the position characteristic value of any target i and the position characteristic value of a target j, the normalization is from 0 to 1, and the closer the positions of the two targets are, the higher the probability of belonging to the same group is;
time similarity matrix E t ＝{1-t ij }，t ij Is the time characteristic value of an arbitrary object i and the time characteristic value of an object j 1 The closer the time of occurrence of the two targets is, the greater the probability of belonging to the same group is;
velocity similarity matrix E s ＝{1-s ij }，s ij L being the speed of an arbitrary target i and the speed of a target j 1 The closer the two target velocities are, the greater the probability of belonging to the same group, normalized to 0 to 1.
5. The method according to claim 1, wherein step SS4 specifically comprises:
step SS 41: taking the adjacent matrix E as a relation matrix of the relation between the standard targets to obtain a degree information matrix D, which is shown in a formula (3); the degree information matrix D is combined with the adjacent matrix E to obtain a Laplacian matrix L, which is a symmetric matrix, according to a formula (4);
Figure FDA0003609220170000031
L＝D -1/2 (D-E)D -1/2 formula (4)
Step SS 42: performing SVD on the Laplacian matrix L to obtain a characteristic value lambda and a characteristic vector v; calculating eigenvector z ═ D 1/2 v, classifying the samples i corresponding to the components smaller than 0 in the converted eigenvector z into one class, recalculating the energy matrix E 'between the samples i' and the Laplacian matrix L 'and the eigenvector z' for the samples i 'corresponding to the components larger than 0, and classifying the corresponding samples with the components smaller than 0 in the eigenvector z' into a new class, so that the recursion is ended until the components larger than 0 do not exist in the eigenvector z 'or the dimension of the eigenvector z' is 1 according to the recursion mode, and forming a new target group C by the classified targets each time (C) 1 ,c 2 ,...,c k ) Finally, the population number k of the whole queue is obtained.
6. A queue counting system based on fusion of multiple information features, comprising:
the target detection and tracking module specifically executes: performing monocular visual target detection and tracking, comprising: acquiring continuous N frames of images, detecting the positions of the heads and shoulders of the people in the images of each frame of image by using a convolutional neural network, tracking multiple targets by using a Deepsort tracking algorithm, acquiring position information of each head in each frame of image as an extraction target, and generating a target sequence T;
and the target feature extraction module specifically executes: performing behavior analysis on each sequence in the target sequence T to obtain high-level features of the extracted target, wherein the high-level features comprise track features, position features, time features and speed features of the extracted target;
the module for calculating the adjacency matrix specifically executes: performing a computation of an adjacency matrix E of the decimated objects, comprising: calculating a trajectory similarity matrix E c Position similarity matrix E l Time similarity matrix E t And the velocity similarity matrix E s And E is calculated in the following manner,
Figure FDA0003609220170000041
wherein E ═ { E ═ E ij Is a normalized n × n symmetric matrix representing the degree of association between the various extracted targets, i.e. the probability of belonging to the same group of targets, e ij The larger the probability that object i and object j belong to the same group of objects, n the number of detected objects, w c Weights representing a trajectory similarity matrix; w is a l Weights representing the location similarity matrix; w is a t Weights representing a temporal similarity matrix; w is a s Weights representing a velocity similarity matrix;
and a module for calculating actual group number: performing a set of sets of actual queues for calculating N frames of images, comprising: and calculating the Laplacian matrix L of the adjacent matrix E of the extracted target, obtaining the optimal clustering decomposition by using a graph cut algorithm, and finally outputting the actual queued group number set.
7. The queue counting system based on fusion of multiple information features according to claim 6, wherein the target detection and tracking module specifically comprises:
the detecting the human head and shoulder position in the picture by using the method based on the convolutional neural network specifically comprises the following steps: using a 19-layer convolutional neural network in each frame image for target detection, detecting the part from the shoulder to the top of the head of the human head, and outputting target frame information l (u) p ,v p ,w p ,h p ) Wherein (u) p ,v p ) Is the coordinate of the center point, w p And h p Respectively representing the width and height of the target frame;
the tracking the detected multiple targets by using the deep sort tracking algorithm specifically comprises the following steps: finding a matching frame of the next frame aiming at the target detected by each frame of image, if the matching frame cannot be found, taking the current frame as a starting point to be counted into the track, and finally outputting a target sequence detected in the N frames of images as follows:
T＝{t 1 ,t 2 ,t i ,...,t n }，
t i ＝{l s ,l s+1 ,...,l e },1≤s≤e≤N；
where N represents the number of detected targets and N represents the number of frames.
8. The queue counting system based on fusion of multiple information features according to claim 6, wherein the target feature extraction module specifically comprises: for the ith target T in the target sequence T detected in the N frames of images i The following high-level features are calculated:
trajectory feature A for characterizing the motion of an object over time c ，
Figure FDA0003609220170000051
Position feature B characterizing the position of the target c ，B c Representing the track characteristics A c The first coordinate of (1);
temporal characteristics D c ，D c Is a number between 1 and N, indicating that the object appears in the picture starting at the frame number;
speed characteristic S c According to the track characteristics A c Calculating a number, see formula (2):
Figure FDA0003609220170000052
wherein, omega represents the Euclidean distance,
Figure FDA0003609220170000053
is the order of eyesThe coordinates of the ith target in the target sequence T.
9. The queue counting system based on fusion of multiple information features according to claim 6, wherein the module for calculating the adjacency matrix specifically comprises: adjacency matrix E is formed by trajectory similarity matrix E c A position similarity matrix E l Time similarity matrix E t And velocity similarity matrix E s The weighting results, wherein:
trajectory similarity matrix E c ＝{1-a ij }，a ij The sum of the squared differences of the track characteristic of any target i and the track characteristic of a target j is normalized to 0 to 1, and the probability of belonging to the same group is higher when the track difference between the two targets is smaller;
position similarity matrix E l ＝{1-l ij }，l ij The Euclidean distance is calculated by the position characteristic value of any target i and the position characteristic value of a target j, the normalization is from 0 to 1, and the closer the positions of the two targets are, the higher the probability of belonging to the same group is;
time similarity matrix E t ＝{1-t ij }，t ij Is the time characteristic value of an arbitrary object i and the time characteristic value of an object j 1 The closer the time of occurrence of the two targets is, the greater the probability of belonging to the same group is;
velocity similarity matrix E s ＝{1-s ij }，s ij L being the speed of an arbitrary target i and the speed of a target j 1 The closer the two target velocities are, the greater the probability of belonging to the same group, normalized to 0 to 1.
10. The queue counting system based on fusion of multiple information features according to claim 6, wherein the module for calculating the actual group number specifically comprises: taking the adjacent matrix E as a relation matrix of the relation between the standard targets to obtain a degree information matrix D, which is shown in a formula (3); the degree information matrix D is combined with the adjacent matrix E to obtain a Laplacian matrix L, which is a symmetric matrix, according to a formula (4);
Figure FDA0003609220170000054
L＝D -1/2 (D-E)D -1/2 formula (4)
Carrying out SVD on the Laplacian matrix L to obtain a characteristic value lambda and a characteristic vector v; calculating eigenvector z ═ D 1/2 v, classifying the samples i corresponding to the components smaller than 0 in the converted eigenvector z into one class, recalculating the energy matrix E 'between the samples i' and the Laplacian matrix L 'and the eigenvector z' for the samples i 'corresponding to the components larger than 0, and classifying the corresponding samples with the components smaller than 0 in the eigenvector z' into a new class, so that the recursion is ended until the components larger than 0 do not exist in the eigenvector z 'or the dimension of the eigenvector z' is 1 according to the recursion mode, and forming a new target group C by the classified targets each time (C) 1 ,c 2 ,...,c k ) Finally, the population number k of the whole queue is obtained.
CN202210424749.3A 2022-04-22 2022-04-22 Queue counting method and system based on fusion of multiple information characteristics Pending CN114882429A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210424749.3A CN114882429A (en) 2022-04-22 2022-04-22 Queue counting method and system based on fusion of multiple information characteristics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210424749.3A CN114882429A (en) 2022-04-22 2022-04-22 Queue counting method and system based on fusion of multiple information characteristics

Publications (1)

Publication Number Publication Date
CN114882429A true CN114882429A (en) 2022-08-09

Family

ID=82671470

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210424749.3A Pending CN114882429A (en) 2022-04-22 2022-04-22 Queue counting method and system based on fusion of multiple information characteristics

Country Status (1)

Country Link
CN (1) CN114882429A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116561372A (en) * 2023-07-03 2023-08-08 北京瑞莱智慧科技有限公司 Personnel gear gathering method and device based on multiple algorithm engines and readable storage medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116561372A (en) * 2023-07-03 2023-08-08 北京瑞莱智慧科技有限公司 Personnel gear gathering method and device based on multiple algorithm engines and readable storage medium
CN116561372B (en) * 2023-07-03 2023-09-29 北京瑞莱智慧科技有限公司 Personnel gear gathering method and device based on multiple algorithm engines and readable storage medium

Similar Documents

Publication Publication Date Title
Sun et al. Benchmark data and method for real-time people counting in cluttered scenes using depth sensors
CN107967451B (en) Method for counting crowd of still image
Kuo et al. How does person identity recognition help multi-person tracking?
WO2017122258A1 (en) Congestion-state-monitoring system
CN111191667B (en) Crowd counting method based on multiscale generation countermeasure network
US20170169297A1 (en) Computer-vision-based group identification
Seow et al. Neural network based skin color model for face detection
Mukherjee et al. Anovel framework for automatic passenger counting
CN109325404A (en) A kind of demographic method under public transport scene
CN113011367A (en) Abnormal behavior analysis method based on target track
CN111709300B (en) Crowd counting method based on video image
Guo et al. Improved hand tracking system
CN112232199A (en) Wearing mask detection method based on deep learning
CN111353338B (en) Energy efficiency improvement method based on business hall video monitoring
de Almeida et al. Change detection in human crowds
CN114882429A (en) Queue counting method and system based on fusion of multiple information characteristics
Farhood et al. Counting people based on linear, weighted, and local random forests
Ilao et al. Crowd estimation using region-specific HOG With SVM
US20230154192A1 (en) Person flow analysis method and system, electronic device, and readable storage medium
CN106845361B (en) Pedestrian head identification method and system
Padmashini et al. Vision based algorithm for people counting using deep learning
Pane et al. A people counting system for business analytics
Arivazhagan Versatile loitering detection based on non-verbal cues using dense trajectory descriptors
Schmuck et al. Robocentric conversational group discovery
Lee et al. Fast people counting using sampled motion statistics

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