CN116187634A - Intelligent queuing system and prediction method for same - Google Patents

Intelligent queuing system and prediction method for same Download PDF

Info

Publication number
CN116187634A
CN116187634A CN202211343051.5A CN202211343051A CN116187634A CN 116187634 A CN116187634 A CN 116187634A CN 202211343051 A CN202211343051 A CN 202211343051A CN 116187634 A CN116187634 A CN 116187634A
Authority
CN
China
Prior art keywords
queue
time
checkout
target
module
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
CN202211343051.5A
Other languages
Chinese (zh)
Inventor
苏晋鹏
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.)
Hangzhou Turing Video Technology Co ltd
Original Assignee
Hangzhou Turing Video Technology 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 Hangzhou Turing Video Technology Co ltd filed Critical Hangzhou Turing Video Technology Co ltd
Priority to CN202211343051.5A priority Critical patent/CN116187634A/en
Publication of CN116187634A publication Critical patent/CN116187634A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06312—Adjustment or analysis of established resource schedule, e.g. resource or task levelling, or dynamic rescheduling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • 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
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06315—Needs-based resource requirements planning or analysis
    • 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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Strategic Management (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Operations Research (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Data Mining & Analysis (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Human Computer Interaction (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an intelligent queuing system and a prediction method for the same, and belongs to the field of video monitoring security protection. The intelligent queuing system comprises: the system comprises a data acquisition module, an algorithm module, a strategy module, a prediction module and a display module, wherein the data acquisition module is connected with monitoring equipment, the algorithm module is connected with the data acquisition module, the strategy module is connected with the algorithm module, the prediction module is connected with the strategy module, and the display module is connected with the strategy module and the prediction module. According to the invention, the policy module defines the policy area in the image data, idle interference personnel can be effectively filtered, the data of the queue target in the policy area is obtained, and the number of queues required in the next time period is predicted by the prediction module on the basis of the data, so that the problems that the manager cannot fully utilize the existing resources, the queuing time of the client is long and the consumption experience of the client is poor due to inaccurate queue data identification in the prior art are solved.

Description

Intelligent queuing system and prediction method for same
Technical Field
The invention belongs to the field of video monitoring security, and particularly relates to an intelligent queuing system and a prediction method for the intelligent queuing system.
Background
The problem that the existing market using manual checkout or intelligent checkout equipment is long in checkout when the people flow is large is solved, the problem is often found by management staff when the queuing is long, then the management staff temporarily arranges the staff to be responsible for checkout or open the exit more, and the problem that the consumption experience is poor due to overlong customer queuing time exists.
The existing solutions are as follows: 1. transmitting the video of the strategy area to a display module for display, and then identifying the queue data by a manager or a client, wherein the method has the problem of portrait privacy disclosure of the strategy area;
2. the number of people in the image is detected to judge the queue data, and the queue data identification technology has the problems that other action staff besides the queuing staff in the image can not accurately identify and predict the queue data, and the work efficiency of management staff is low.
Therefore, it is necessary to provide an intelligent queuing system capable of accurately detecting queue data to improve the working efficiency of management personnel, further reduce the queuing time of clients and optimize the consumption experience of the clients.
Disclosure of Invention
The invention aims to: an intelligent queuing system and a prediction method therefor are provided to solve the above-mentioned problems of the prior art.
The technical scheme is as follows: in a first aspect, an intelligent queuing system includes: and the data acquisition module is connected with the monitoring equipment.
And the algorithm module is connected with the data acquisition module.
And the strategy module is connected with the algorithm module.
And the prediction module is connected with the data acquisition module and the strategy module.
And the display module is connected with the strategy module and the prediction module.
In the working state, the data acquisition module collects image data acquired by the monitoring equipment and transmits the image data to the algorithm module, the data acquisition module counts the number of people entering and exiting a supermarket in the acquired image data, defines the number of people as the number of target data, and transmits the number of the target data to the prediction module. The algorithm module identifies and tracks target data in the image data and transmits the target data to the policy module.
The policy module identifies a policy region defined in the image data and defines target data in the policy region as a queue target, calculates data of the queue target, and transmits the data of the queue target to the prediction module.
The prediction module predicts the number of queues required in the next time period according to the number of target data and the data of the queue targets, calculates the number of queues required to be opened or closed, and transmits the number of queues to the display module.
And the display module is used for visually displaying the data of the queue target calculated by the strategy module and the number of the queues required to be opened or closed calculated by the prediction module.
In a second aspect, a predictive method for an intelligent queuing system includes: the prediction module performs a prediction step.
The first prediction step, the prediction module divides the data of the preset time interval into N equal parts, and the N equal parts are used as the characteristics of N dimensions to judge whether the current T moment is in the time interval or not, if so, value is met t1:t2 =1, otherwise is Value t1:t2 ＝0。
A second prediction step, wherein the prediction module defines target data at the time T as: t moment of entering target and T moment of leaving target T enter 、T leave 。
The prediction module defines target data at the time of T-1 as: t-1 enters the target at moment and leaves the target T-1 at moment T-1 enter 、T-1 leave 。
The prediction module defines target data at the time of T-2 as: t-2 is the target number of people entering at the moment T-2 and T-2 is the target number of people leaving at the moment T-2 enter 、T-2 leave 。
The prediction module defines the data identified by the policy module as: number of queues num_lane_open opened in T period T 、
Number of queues open during the T-1 time period: num_lane_open T-1 、
Number of queues open during the T-2 period: num_lane_open T-2 、
Sum of average queuing queue lengths of each queue in T time period: num_avg_queue_len T 、
Sum of average checkout queue lengths for each queue over a period of T: num_avg_checkout_len T ，
Sum of the checkout numbers for each queue over the T period: num_checkout_person T ，
wherein ,
Figure BDA0003917118220000021
Figure BDA0003917118220000022
Figure BDA0003917118220000031
i is the number of cameras.
And thirdly, combining the first and second prediction steps, constructing characteristic information of N+12 dimensions, and training a model by using a machine learning method to obtain an optimal model.
A fourth prediction step of obtaining a sum num_avg_queue_len of queue lengths to be opened at the time T+1 on the basis of the optimal model in the third prediction step T+1 。
The average queue length threshold value is preset to be delta, and the number of the queues started at the moment T+1 is predicted to be:
num_lane_open_precidt T+1 ＝num_avg_queue_len T+1 /delta。
a fifth prediction step of judging num_lane_open_predt in the fourth prediction step T+1 And num_lane_open T Whether the difference of (c) is greater than a preset difference threshold threshold=1.
diff＝num_lane_open_precidt T+1 -num_lane_open T When diff>And 1, transmitting information of the multi-open diff bar queue to a display module.
And when diff < -1, transmitting information for closing the diff bar queue to the display module.
When diff is less than or equal to-1 and less than or equal to 1, the normal information of the queue is transmitted to the display module, and the actual situation can be better simulated by adding the characteristic information of multiple dimensions, so that the model precision is improved.
In a further embodiment of the second aspect, the policy module performs a policy step;
And a strategy step I, defining a strategy area strategy-region in the image data, and defining at least a working area cashier-region, a queuing area queue-region and a processing area checkout-region in the strategy area.
A policy step II, defining the sum of target data as a set B, defining the sum of target data in a policy region strategy-region as a set C, defining the sum of residual target data as a set D, and tracking the target data in the set C to obtain an id track of the target data in the set C, wherein the initial tracking time of the target id
Figure BDA0003917118220000032
For tracking the moment of id validation, the tracking time is updated while tracking>
Figure BDA0003917118220000033
And thirdly, transmitting the id track of the target data in the set C in the second strategy step into a working area, and defining the queue as an ON state ON when the id track of the target data is always in the working area and the stay time is longer than the preset time, wherein the number of the queues under the current camera is the number of the queues which are ON at the moment: num_lane_open T +1, otherwiseOFF, the number of queues opened under the current camera: num_lane_open T +0。
And step four, transmitting the id track of the target data in the set C in the step two into a queuing area queue-region while executing the step three, when the id track of the target data enters the area, assigning a queuing state queue to the flag of the target data, determining that the target is in a semi-effective state, counting queuing time according to a tracking frame, and calculating the number of queuing people meeting the condition at the moment: num_ordering_person+1, and accumulating the time of all the target data meeting the conditions to obtain acc_queue_time.
A fifth strategy step of transmitting the id track of the target data in the set C in the second strategy step into a processing area checkout-region, when the id track of the target data enters the area, and the state of the target data is flag= queue, and the state flag = checkout is given, and the checkout time of the target is counted according to the tracking frame from the moment
Figure BDA0003917118220000041
And accumulating the time of all the target data meeting the conditions to obtain the acc_checkout_time.
And step six, judging whether the flag state of the id track of the target data is a checkout state checkout, if not, exiting.
If so, judging whether the target leaves the processing area through the id track of the target data, if so, finishing the settlement of the number num_checkout_person+1, and if not, finishing the settlement of the number num_checkout_person+1.
A policy step seven, obtaining an average queuing length avg_queue_len, an average queuing time avg_queue_time, an average checkout number length avg_checkout_len, an average checkout time avg_checkout_time, a total checkout number num_checkout_person, a total queuing time acc_queue_time, a total checkout time acc_checkout_time, an average waiting time avg_wait_time and an average waiting number avg_wait_len of the queue targets within a certain time period T;
wherein ：
avg_wait_len＝acc_wait_time/T
avg_wait_time＝acc_wait_time/(num_checkout_person+num_queuing_person+
num_checkout_person+num_checkouting_person)
avg_queue_len＝acc_queue_time/T
avg_queue_time＝acc_queue_time/
(num_checkout_person+num_queuing_person)
avg_checkout_len＝acc_checkout_time/T
avg_checkout_time＝acc_checkout_time/
(num_checkout_person+num_checkouting_person)。
a policy step eight, repeating the policy step two to the policy step seven, calculating queuing parameter information of each queue object to obtain { avg_queue_len1, avg_queue_len2, & avg_queue_lenN },
{avg_queue_time1、avg_queue_time2、..avg_queue_timeN}，
{avg_checkout_len1、avg_queue_time2、..avg_queue_timeN}，
{avg_checkout_time1、avg_checkout_len2、..avg_checkout_lenN}，
{ num_checkout_person1, num_checkout_person2, num_checkout_person N } show the information of each queue respectively;
strategy step nine: and resetting relevant parameters every T time so as to obtain queuing information in the T time period at the next moment, filtering out some idle interference personnel by a method of dividing areas and giving different checkout states, and obtaining accurate queue target information output at the T moment.
In a further embodiment of the second aspect, the algorithm module performs algorithm steps.
The first algorithm step, the algorithm module acquires head-shoulder data from the image data, defines the sum of the head-shoulder data as a head-shoulder data set, and can detect more pedestrian targets when a pedestrian is shielded by marking the head-shoulder data so as to obtain more accurate data results
And step two, training the head and shoulder data set by using a target detection algorithm to obtain a head and shoulder detection model.
And thirdly, detecting the image data to obtain a head-shoulder frame in the image data, and then performing coordinate conversion on the head-shoulder frame to obtain a target frame of target data.
And fourthly, tracking the target frame transmitted in the third step by using a tracking algorithm, aiming at the camera code stream in the overlook view, the pedestrian tracking precision with higher precision can be obtained, and when the target is subjected to complex interference of the category of the identical attribute of shielding by people, the technical scheme can still realize stable tracking and does not lose tracking targets.
In a further embodiment of the second aspect, in the predicting step three, a SVR machine learning method by support vector machine regression is used to train the model, and a SVR method by support vector machine regression is used to train the 36-dimension feature, so that correlation between queuing information in a T time period before and after can be fully considered, and a supermarket queuing prediction queue model with higher precision is obtained.
In a further embodiment of the second aspect, in the algorithm step two, a target detection algorithm YOLOX-M algorithm based on deep learning is used for training the head and shoulder data set, the defect of detecting the insufficient pedestrians in the overlooking view of the supermarket is relieved by using the target detection algorithm YOLOX-M algorithm to detect the head and shoulder of the pedestrians in the overlooking view, and then most of the frames of the human body are obtained in a coordinate conversion mode.
In a further embodiment of the second aspect, in the algorithm step three, the target frame data is (X, Y, W, H), and the target frame data calculation formula is: x=x-w ratio_x;
Y＝y；
W＝w(1+2*ratio_x)；
H＝h(1+ratio_y)；
wherein x and y respectively represent the coordinates of the upper left corner of the target frame detected by the detection algorithm, w and h respectively represent the width and height of the frame, and the value range of ratio_x [0,0.12]. When the parameter of the value is 0, the width of the pedestrian frame is the width of the original head shoulder, and when the parameter of the value is 0.12, the width of the pedestrian frame basically covers the width of the whole pedestrian. The value range of ratio_y is between [0,1.6], when the value is 0, the height of the pedestrian frame is the width of the original head shoulder, and when the value is 1.6, the height of the pedestrian frame can basically cover all the pedestrian frames completely. In the experiment, ratio_x=0.12, ratio_y=1.6.
The parameter plays a vital role in the intelligent queuing system and the subsequent prediction algorithm. Because the targets in the working area cashier-region, the queuing area queue-region and the processing area checkout-region are subjected to tracking processing, whether the targets enter the area is judged first. The judgment basis is as follows: the center point of the target transformed bounding box is within this region. Only the average queue length and other parameters are accurate, and the overall accuracy of the prediction algorithm is more accurate. In a further embodiment of the second aspect, in the fourth algorithm step, the deep-learning-based deep_sort tracking algorithm is used for carrying out target tracking on the target frame transmitted in the third algorithm step, and the appearance characteristic advantages of the deep-sort are fully utilized to extract more abundant humanoid characteristics, so that the problem of personnel follow-up loss under the view angle is solved, and the tracking precision is integrally improved.
The beneficial effects are that: the invention discloses an intelligent queuing system, which defines a policy area in image data through a policy module, can effectively filter idle interference personnel to obtain data of a queuing target in the policy area, predicts the number of queues required in the next time period through a prediction module on the basis, can enable management personnel to fully utilize the existing resource to arrange queues and workers to reduce queuing time of clients, and solves the problems that the management personnel cannot fully utilize the existing resource, the queuing time of clients is long and the consumption experience of the clients is poor due to inaccurate identification of the queuing data in the prior art.
Drawings
Fig. 1 is a schematic diagram of the framework of the intelligent queuing system of the present application.
FIG. 2 is a schematic flow chart of a prediction method of the present application.
Fig. 3 is a schematic diagram of the conversion of head-shoulder frame coordinates into target frames according to the present application.
FIG. 4 is a schematic diagram of a policy module of the present application dividing policy areas, work areas, queuing areas, and processing areas in image data.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. It will be apparent, however, to one skilled in the art that the invention may be practiced without one or more of these details. In other instances, well-known features have not been described in detail in order to avoid obscuring the invention.
The invention discloses an intelligent queuing system capable of accurately detecting queue data to improve the working efficiency of management personnel, further reduce the queuing time of clients and optimize the consumption experience of the clients and a prediction method for the intelligent queuing system.
The following describes an intelligent queuing system and a prediction method for the same, taking a supermarket as an example.
In a first aspect, an intelligent queuing system includes: the system comprises a data acquisition module, an algorithm module, a strategy module, a prediction module and a display module.
The data acquisition module is connected with monitoring equipment, wherein the monitoring equipment can be visual equipment such as a camera.
The algorithm module is connected with the data acquisition module.
The policy module is connected with the algorithm module, and performs a series of policy logics according to the result given by the algorithm module, so that the average queuing number length, the average queuing time average checkout number length, the average processing time and the total checkout number of supermarket shoppers in a certain period of time are completed.
And the prediction module is connected with the data acquisition module and the strategy module.
And the display module is connected with the strategy module and the prediction module.
Working principle: in the working state, the data acquisition module collects image data acquired by the monitoring equipment, the image data are transmitted to the algorithm module and the prediction module for processing, the image data are video pictures acquired by each camera in a supermarket environment, the data acquisition module counts the number of people entering and exiting the supermarket in the acquired image data, the personnel data are defined as the number of target data, and the number of the target data are transmitted to the prediction module, wherein in the embodiment, the image data can be video or code stream pictures.
The algorithm module identifies and tracks target data in the image data and transmits the target data to the strategy module, wherein the target data in the supermarket environment is space data occupied by supermarket shoppers and staff in the image.
The policy module identifies and defines a policy area in the image data, defines target data in the policy area as queue targets, calculates the data of the queue targets, and transmits the data of the queue targets to the prediction module, wherein the data of the queue targets in the supermarket environment comprises average queuing number length, average queuing time average checkout number length, average processing time and total checkout number of supermarket shoppers calculated by a series of policy logics according to the result given by the algorithm module.
The prediction module predicts the number of queues required in the next time period according to the number of target data and the number of queues required in the next time period, calculates the number of queues required to be opened or closed, and transmits the number of queues to the display module.
The display module visualizes and displays the data of the queue target calculated by the strategy module and the number of the queues required to be opened or closed calculated by the prediction module, and displays the queue prediction at the time T+1 output by the prediction module and the queuing information output by the strategy module on a screen.
By defining a strategy area in the image data through the strategy module, idle interference personnel can be effectively filtered, data of a queue target in the strategy area can be obtained, the number of queues required in the next time period is predicted through the prediction module on the basis, and management personnel can fully utilize the existing resource to arrange the queues and staff to reduce queuing time of clients.
In a second aspect, a predictive method for an intelligent queuing system includes: the prediction module performs a prediction step.
The first prediction step, the prediction module divides the data of the preset time interval into N equal parts, and the N equal parts are used as the characteristics of N dimensions to judge whether the current T moment is in the time interval or not, if so, value is met t1:t2 =1, otherwise is Value t1:t2 ＝0。
A second prediction step, wherein the prediction module defines target data at the time T as: t moment of entering target and T moment of leaving target T enter 、T leave 。
The prediction module defines target data at the time of T-1 as: t-1 enters the target at moment and leaves the target T-1 at moment T-1 enter 、T-1 leave 。
The prediction module defines the target data at the time of T-2 as: t-2 is the target number of people entering at the moment T-2 and T-2 is the target number of people leaving at the moment T-2 enter 、T-2 leave 。
The prediction module defines the data identified by the policy module as: number of queues num_lane_open opened in T period T 、
Number of queues open during the T-1 time period: num_lane_open T-1 、
Number of queues open during the T-2 period: num_lane_open T-2 、
Sum of average queuing queue lengths of each queue in T time period: num_avg_queue_len T 、
Sum of average checkout queue lengths for each queue over a period of T: num_avg_checkout_len T 、
Sum of the checkout numbers for each queue over the T period: num_checkout_person T ，
wherein ,
Figure BDA0003917118220000091
Figure BDA0003917118220000092
/>
Figure BDA0003917118220000093
i is the number of cameras.
And thirdly, combining the first and second prediction steps, constructing characteristic information of N+12 dimensions, training the model by using a machine learning method to obtain an optimal model, and better simulating actual conditions by selecting the information of the characteristic dimensions such as the number of people, the number of groups, time and the like so as to obtain better model precision, wherein the optimal model is a model with highest precision on a plurality of test sets selected by using machine learning in advance.
A fourth prediction step of obtaining a sum num_avg_queue_len of queue lengths to be opened at the time T+1 on the basis of the optimal model in the third prediction step T+1 。
The average queue length threshold value is preset to be delta, and the number of the queues started at the moment T+1 is predicted to be:
num_lane_open_precidt T+1 ＝num_avg_queue_len T+1 /delta。
a fifth prediction step of judging num_lane_open_predt in the fourth prediction step T+1 And num_lane_open T Whether the difference of (c) is greater than a preset difference threshold threshold=1.
diff＝num_lane_open_precidt T+1 -num_lane_open T When diff>And 1, transmitting information of the multi-open diff bar queue to a display module.
And when diff < -1, transmitting information for closing the diff bar queue to the display module.
And when diff is less than or equal to-1 and less than or equal to 1, transmitting queue normal information to the display module.
In a further embodiment of the second aspect, the policy module performs a policy step.
And a strategy step I, defining a strategy area strategy-region in the image data, and defining at least a working area cashier-region, a queuing area queue-region and a processing area checkout-region in the strategy area.
A policy step II, defining the sum of target data as a set B, defining the sum of target data in a policy region strategy-region as a set C, defining the sum of residual target data as a set D, and tracking the target data in the set C to obtain an id track of the target data in the set C, wherein the initial tracking time of the target id
Figure BDA0003917118220000101
For tracking the moment of id validation, the tracking time is updated at the same time of tracking >
Figure BDA0003917118220000102
And thirdly, transmitting the id track of the target data in the set C in the second strategy step into a working area, and defining the queue as an ON state ON when the id track of the target data is always in the working area and the stay time is longer than the preset time, wherein the number of the queues under the current camera is the number of the queues which are ON at the moment: num_lane_open T +1, otherwise, OFF, the number of queues opened under the current camera: num_lane_open T +0。
And step four, transmitting the id track of the target data in the set C in the step two into a queuing area queue-region while executing the step three, when the id track of the target data enters the area, queuing the flag of the target data into a state queue, determining that the target is in a semi-effective state, counting queuing time according to a tracking frame, and calculating the number of queuing people meeting the condition at the moment: and (3) num_queue_person+1, accumulating the time of all the target data meeting the conditions to obtain acc_queue_time, wherein the tracking frame of the target id has a starting time at the initial time, and the tracking frame is kept tracking all the time and the time duration is updated all the time, so that the waiting time of the target in the area can be obtained according to the current time and the initial time.
A fifth strategy step of transmitting the id track of the target data in the set C in the second strategy step into a processing area checkout-region, when the id track of the target data enters the area and the target data flag state is queue, newly giving the target state flag as a checkout state checkout, and counting the checkout time of the target according to the tracking frame from the moment
Figure BDA0003917118220000103
And accumulating the time of all the target data meeting the conditions to obtain acc_checkout_time, wherein the total waiting time acc_wait_time=acc_queue_time+acc_checkout_time
And step six, judging whether the flag state of the id track of the target data is a checkout state checkout, if not, exiting.
If so, judging whether the target leaves the processing area through the id track of the target data, if so, finishing the settlement of the number num_checkout_person+1, and if not, finishing the settlement of the number num_checkout_person+1.
The policy step seven, obtaining an average queuing length avg_queue_len, an average queuing time avg_queue_time, an average checkout number length avg_checkout_len, an average checkout time avg_checkout_time, a total checkout number num_checkout_person, a total queuing time acc_queue_time and a total checkout time acc_checkout_time of the queue target within a certain time period T. Average waiting time avg_wait_time, average waiting number avg_wait_len.
wherein ：
avg_wait_len＝acc_wait_time/T
avg_wait_time＝acc_wait_time/(num_checkout_person+num_queuing_person+
num_checkout_person+num_checkouting_person)
avg_queue_len＝acc_queue_time/T
avg_queue_time＝acc_queue_time/
(num_checkout_person+num_queuing_person)
avg_checkout_len＝acc_checkout_time/T
avg_checkout_time＝acc_checkout_time/
(num_checkout_person+num_checkouting_person)。
a policy step eight, repeating the policy step two to the policy step seven, calculating queuing parameter information of each queue object to obtain { avg_queue_len1, avg_queue_len2, & avg_queue_lenN },
{avg_queue_time1、avg_queue_time2、..avg_queue_timeN}，
{avg_checkout_len1、avg_queue_time2、..avg_queue_timeN}，
{avg_checkout_time1、avg_checkout_len2、..avg_checkout_lenN}，
{num_checkout_person1、num_checkout_person2、num_checkout_personN}
displaying the information of each queue respectively;
strategy step nine: and resetting relevant parameters every T time so as to obtain queuing information in the T time period at the next moment, filtering out some idle interference personnel by a method of dividing areas and giving different checkout states, and obtaining accurate queue target information output at the T moment.
In a further embodiment of the second aspect, the existing method does not have a way to achieve good precision in detecting and tracking pedestrians in a top view, and especially the problem that the target id is easy to lose once the pedestrians are staggered.
To solve the above problems, the algorithm module performs algorithm steps.
The first algorithm step, the algorithm module acquires head-shoulder data from the image data, defines the sum of the head-shoulder data as a head-shoulder data set, and can detect more pedestrian targets when a pedestrian is shielded by marking the head-shoulder data so as to obtain more accurate data results
And step two, training the head and shoulder data set by using a target detection algorithm to obtain a head and shoulder detection model.
And thirdly, detecting the image data to obtain a head-shoulder frame in the image data, and then performing coordinate conversion on the head-shoulder frame to obtain a target frame of target data.
And fourthly, tracking the target frame transmitted in the third step by using a tracking algorithm, aiming at the camera code stream in the overlook view, the pedestrian tracking precision with higher precision can be obtained, and when the target is subjected to complex interference of the category of the identical attribute of shielding by people, the technical scheme can still realize stable tracking and does not lose tracking targets.
Through the method of detecting the head and the shoulder firstly, more pedestrian targets can be detected through the method of frame conversion, in addition, the frame conversion can also greatly help follow-up target tracking, the precision of pedestrian detection and tracking can be improved, frames cannot be lost when pedestrians are staggered, and the problem that target ids are easy to lose is solved.
In a further embodiment of the second aspect, in the predicting step three, a SVR machine learning method by support vector machine regression is used to train the model, and a SVR method by support vector machine regression is used to train the 36-dimension feature, so that correlation between queuing information in a T time period before and after can be fully considered, and a supermarket queuing prediction queue model with higher precision is obtained.
In a further embodiment of the second aspect, in the algorithm step two, a target detection algorithm YOLOX-M algorithm based on deep learning is used for training the head and shoulder data set, the defect of detecting the insufficient pedestrians in the overlooking view of the supermarket is relieved by using the target detection algorithm YOLOX-M algorithm to detect the head and shoulder of the pedestrians in the overlooking view, and then most of the frames of the human body are obtained in a coordinate conversion mode.
In a further embodiment of the second aspect, in the algorithm step three, the target frame data is (X, Y, W, H), and the target frame data calculation formula is: x=x-w ratio_x.
Y＝y。
W＝w(1+2*ratio_x)。
H＝h(1+ratio_y)。
Wherein x and y respectively represent the coordinates of the initial corner point of the target frame detected by the detection algorithm, and w and h respectively represent the width and height of the frame, and the value range of ratio_x is 0,0.12. When the parameter of the value is 0, the width of the pedestrian frame is the width of the original head shoulder, and when the parameter of the value is 0.12, the width of the pedestrian frame basically covers the width of the whole pedestrian. The value range of ratio_y is between [0,1.6], when the value is 0, the height of the pedestrian frame is the width of the original head shoulder, and when the value is 1.6, the height of the pedestrian frame can basically cover all the pedestrian frames completely. In the experiment, ratio_x=0.12, ratio_y=1.6
The two parameters have important roles to the intelligent queuing system and the subsequent prediction algorithm. Because the targets in the working area cashier-region, the queuing area queue-region and the processing area checkout-region are subjected to tracking processing, whether the targets enter the area is judged first. The judgment basis is as follows: the center point of the target transformed bounding box is within this region. Only the average queue length and other parameters are accurate, and the overall accuracy of the prediction algorithm is more accurate.
Through the method of detecting the head and the shoulder firstly and then through the method of frame conversion, more pedestrian targets can be detected, and in addition, the frame conversion can also greatly help follow-up target tracking.
In a further embodiment of the second aspect, in the fourth algorithm step, the deep-learning-based deep_sort tracking algorithm is used for carrying out target tracking on the target frame transmitted in the third algorithm step, and the appearance characteristic advantages of the deep-sort are fully utilized to extract more abundant humanoid characteristics, so that the problem of personnel follow-up loss under the view angle is solved, and the tracking precision is integrally improved.
The preferred embodiments of the above embodiments in combination with a supermarket environment are: the data acquisition module is responsible for collecting video pictures acquired by each camera and transmitting the video pictures to the subsequent algorithm module and the prediction module for processing.
The data acquisition module comprises the following specific steps: 1. collecting camera video real-time code streams (camera 1, camera2 … camera N) of a supermarket checkout area, and transmitting the video real-time code streams to a subsequent algorithm module;
2. the data acquisition module also collects video code streams of cameras at the entrance and exit of the supermarket to obtain data of the number of people entering and exiting in each time period T shown in fig. 2, namely data of hot people entering and exiting shown in fig. 2, and the number of people entering and exiting can be counted by a traditional or deep learning pedestrian detection algorithm to obtain the data of the number of people entering and exiting shown in fig. 2 enter 、T leave The total number of people in the period T is obtained.
The algorithm module detects and tracks pedestrians in the picture by processing the code stream transmitted by the data acquisition module.
The algorithm module comprises the following specific steps: the method comprises the steps of firstly, collecting pedestrian data in a supermarket shopping actual scene, and labeling head and shoulder of pedestrians to obtain a large-batch head and shoulder data set because cameras in the supermarket are generally at higher positions and have overlooking angles; by marking the head and shoulder data, more pedestrian targets can be detected when the pedestrian is shielded, so that more accurate data results can be obtained.
And step two, training the head and shoulder data set by using a target detection algorithm YOLOX-M algorithm based on deep learning to obtain a head and shoulder detection model with higher precision. The YOLOX-M achieves both precision and speed, is the best choice, and achieves the detection speed of 206fps on the basis of the display card RTX2080 at 640 x 640 picture input.
Detecting the code stream picture to obtain a frame of a head and a shoulder in the code stream picture, and then converting the frame of the head and the shoulder to obtain a general partial region (X, Y, W, H) containing pedestrians, wherein the specific calculation formula is as follows:
X＝x-w*ratio_x。
Y＝y。
W＝w(1+2*ratio_x)。
H＝h(1+ratio_y)。
wherein x and y respectively represent coordinates of an upper left corner point of a head and shoulder frame of the pedestrian detected by a detection algorithm, and w and h respectively represent the width and the height of the frame; ratio_x, ratio_y, and values of 0.12 and 1.6, respectively, refer to fig. 3.
And step four, performing target tracking on the conversion frame transmitted in the step three by using a deep-learning-based deep_sort tracking algorithm, and extracting features of appearance features of pedestrians by adopting a deep cosine measurement method, so that the problem of ID switching of the pedestrian tracks during interleaving is further improved.
And the policy module performs a series of policy logics according to the result given by the algorithm module to finish average queuing number, average processing time and total checkout number of supermarket shoppers in a certain time period T. Where T is assumed to be 900s.
The specific steps of the policy module are as follows: the first strategy step is to define a working area cashier-region, a queuing area queue-region, a processing area checkout-region and a strategy area strategy-region of each queue in pictures of different cameras, wherein the strategy area strategy-region comprises three other areas, the queuing area queue-region comprises a processing area, the working area cashier-region is a cashier area in a supermarket environment, the processing area is a checkout area, statistics are carried out on each single queue as shown in fig. 4, and other queues are the same.
And step two, a frame set after being output and converted by the YOLOX-M algorithm is B, a frame set of frames in a policy region strategy-region is C, the rest frames form a set D, and only the targets in the frame set C are subjected to deep-sort tracking processing to obtain the id tracks of the targets.
A third strategy step of transmitting the id frame of the second strategy step into cashier region cashier-region, and if someone is always in the region and the stay time is more than (1/5) T, considering the queue as an ON state ON, num_lane_open T +1, otherwise OFF.
A policy step four, the id frame of the policy step two is transmitted into a queuing area queue-region, if the target id enters the area, the flag of the target is given by the flag= queue, then the target is deemed to be in a semi-active state and its queuing time is counted (based on the tracking frame) and the number of queuing people that now meet the conditions is calculated: num_queue_person+1, and accumulating the time of different coincidence condition ids to obtain acc_queue_time.
And fifthly, transmitting the id frame of the policy step two into a check out-region of a check out area, if the target id enters the area and the state of the target id is flag= queue, giving the state flag = check out, counting the check out time of the target from the moment (according to the tracking frame), and accumulating the time of different conforming condition ids to obtain the acc_check out_time.
Judging whether the flag state of the target id is a checkout or not, and if not, exiting; if so, judging whether the target leaves the checkout area or not through the track line of the target, if so, finishing the checkout of the number num_checkout_person+1, and if not, performing num_checkout_person+1.
And a policy step seven, obtaining an average queuing length avg_queue_len, an average queuing time avg_queue_time, an average processing length avg_checkout_len, an average processing time avg_checkout_time and a total number of people to be processed, wherein the average processing length is an average checkout length in a supermarket environment, the average processing time is an average checkout time, and the total number of people to be processed is a total number of people to be checked out. Wherein: avg_queue_len=acc_queue_time/T.
avg_queue_time＝acc_queue_time/
(num_checkout_person+num_queuing_person)。
avg_checkout_len＝acc_checkout_time/T。
avg_checkout_time＝acc_checkout_time/
(num_checkout_person+num_checkouting_person)。
And a policy step eight, repeating the policy step two to the policy step seven, and calculating queuing parameter information of queues under other cameras. To obtain { avg_queue_len1, avg_queue_len2, }. Avg_queue_lenN,
{avg_queue_time1、avg_queue_time2、..avg_queue_timeN}，
{avg_checkout_len1、avg_queue_time2、..avg_queue_timeN}，
{avg_checkout_time1、avg_checkout_len2、..avg_checkout_lenN}，
{ num_checkout_person1, num_checkout_person2, … } num_checkout_person n } show information of each queue, respectively.
Strategy step nine: and resetting the related parameters every T time, so as to obtain queuing information in the T time period at the next moment.
The prediction module predicts the number of queues opened in the next time period (whether the queues are opened or closed more) according to the result output by the strategy module and the information such as the data transmitted by the data module, so that supermarket management staff can adjust in time, and the existing resources can be utilized to the greater extent.
The specific steps of the prediction module are as follows: the data of 24h of the whole day is divided into 24 equal parts as 24 dimension characteristics, for example, 00:00:00-01: 00. 01:00-02: 00. … 23:00-24:00, judging whether the current T moment is in the time period, if so, satisfying Value t1:t2 =1, otherwise is Value t1:t2 ＝0。
Prediction step two, T using data collection module enter 、T leave At the same time utilize T-1 enter 、T-1 leave and T-2enter 、T-2 leave And the following information for each queue obtained by the policy module:
num_lane_open T 、
num_lane_open T-1 、
num_lane_open T-2 、
num_avg_queue_len T 、
num_avg_checkout_len T 、
Figure BDA0003917118220000171
Figure BDA0003917118220000172
Figure BDA0003917118220000173
and thirdly, combining the first and second prediction steps to jointly construct the characteristic information of the 36 dimensions, as shown in a first chart. Collecting data information for 30 days in a month, performing machine learning by using a training set to train a model, evaluating the training set in the training process by using a verification set, and selecting an optimal model according to test precision by using a test set, wherein the proportions of the training set, the test set and the verification set are as follows: 8:1:1. the model may be trained using SVR machine learning methods that support vector machine regression to obtain num_avg_queue_len at time T+1 T+1 。
And fourthly, aiming at the optimal model obtained in the third prediction step. Num_avg_queue_len at time T+1 is predicted T+1 Delta=1.5 according to a preset threshold value of the average queue length. The number of the queues which are predicted to be opened at the moment T+1 is as follows:
num_lane_open_precidt T+1 ＝num_avg_queue_len T+1 /delta。
a fifth step of judging num_lane_open_precidt obtained in the fourth step of prediction T+1 And num_lane_open T Whether the difference of (c) is greater than the threshold threshold=1 set for implementation.
diff＝num_lane_open_precidt T+1 -num_lane_open T If diff>1, giving an alarm, and if diff is found, the supermarket should open a diff queue more<-1, giving an alarm, and the supermarket should close the superfluous diff lines in time.
Display mode: and displaying the queue prediction at the time T+1 output by the prediction module and queuing information output by the strategy module on a screen.
Figure BDA0003917118220000174
/>
Figure BDA0003917118220000181
/>
Figure BDA0003917118220000191
Table one: features required for predictive algorithms in supermarket embodiments
As described above, although the present invention has been shown and described with reference to certain preferred embodiments, it is not to be construed as limiting the invention itself. Various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (9)

1. An intelligent queuing system, comprising:
The data acquisition module is connected with the monitoring equipment;
the algorithm module is connected with the data acquisition module;
the strategy module is connected with the algorithm module;
the prediction module is connected with the data acquisition module and the strategy module;
the display module is connected with the strategy module and the prediction module;
in a working state, the data acquisition module collects image data acquired by the monitoring equipment and transmits the image data to the algorithm module, the data acquisition module counts the number of people entering and exiting a supermarket in the acquired image data, defines the number of people as the number of target data, and transmits the number of the target data to the prediction module;
the algorithm module identifies and tracks target data in the image data and transmits the target data to the strategy module;
the strategy module identifies a strategy area defined in the image data, defines target data in the strategy area as a queue target, calculates the data of the queue target, and transmits the data of the queue target to the prediction module;
the prediction module predicts the number of queues required by the next time period according to the number of target data and the data of the queue targets, calculates the number of queues required to be opened or closed, and transmits the number of queues to the display module;
And the display module is used for visually displaying the data of the queue target calculated by the strategy module and the number of the queues required to be opened or closed calculated by the prediction module.
2. A predictive method for an intelligent queuing system as claimed in claim 1, comprising:
the prediction module executes a prediction step;
the first prediction step, the prediction module divides the data of the preset time interval into N equal parts, and the N equal parts are used as the characteristics of N dimensions to judge whether the current T moment is in the time interval or not, if so, value is met t1:t2 =1, otherwise is Value t1:t2 ＝0；
A second prediction step, wherein the prediction module defines target data at the time T as: t time of entering target number and T time of leaving target number enter 、T leave ；
The prediction module defines target data at the time of T-1 as: the target number of people entering at the moment T-1 and the target number of people leaving at the moment T-1 are T-1 enter 、T-1 leave ；
The prediction module defines target data at the time of T-2 as: t-2 is the target number of people entering at the moment T-2 and T-2 is the target number of people leaving at the moment T-2 enter 、T-2 leave ；
The prediction module defines the data identified by the policy module as: number of queues num_lane_open opened in T period T 、
Number of queues open during the T-1 time period: num_lane_open T-1 、
Number of queues open during the T-2 period: num_lane_open T-2 、
Sum of average queuing queue lengths of each queue in T time period: num_avg_queue_len T 、
Sum of average checkout queue lengths for each queue over a period of T: num_avg_checkout_len T 、
Sum of the checkout numbers for each queue over the T period: num_checkout_person T ，
wherein ,
Figure FDA0003917118210000021
Figure FDA0003917118210000022
/>
Figure FDA0003917118210000023
i is the number of cameras;
a third prediction step, combining the first and second prediction steps, constructing characteristic information of N+12 dimensions, and training a model by using a machine learning method to obtain an optimal model;
a fourth prediction step of obtaining a sum num_avg_queue_len of queue lengths to be opened at the time T+1 on the basis of the optimal model in the third prediction step T+1 ，
The average queue length threshold value is preset to be delta, and the number of the queues started at the moment T+1 is predicted to be:
num_lane_open_precidt T+1 ＝num_avg_queue_len T+1 /delta；
a fifth prediction step of judging num_lane_open_predt in the fourth prediction step T+1 And num_lane_open T Whether the difference of (a) is greater than a preset difference threshold threshold=1;
diff＝num_lane_open_precidt T+1 -num_lane_open T when diff>1, transmitting information of a multi-open diff strip queue to a display module;
when diff < -1, transmitting information for closing the diff strip queue to a display module;
and when diff is less than or equal to-1 and less than or equal to 1, transmitting queue normal information to the display module.
3. The prediction method of an intelligent queuing system according to claim 2, further comprising:
The policy module executes a policy step;
a strategy step I, defining a strategy area strategy-region in image data, and defining at least a working area cashier-region, a queuing area queue-region and a processing area checkout-region in the strategy area;
a policy step II, defining the sum of target data as a set B, defining the sum of target data in a policy region strategy-region as a set C, defining the sum of residual target data as a set D, and tracking the target data in the set C to obtain an id track of the target data in the set C, wherein the initial tracking time of the target id
Figure FDA0003917118210000031
For tracking the moment of id validation, the tracking time is updated while tracking>
Figure FDA0003917118210000032
And thirdly, transmitting the id track of the target data in the set C in the second strategy step into a working area, and defining the queue as an ON state ON when the id track of the target data is always in the working area and the stay time is longer than the preset time, wherein the number of the queues under the current camera is the number of the queues which are ON at the moment: num_lane_open T +1, otherwise, OFF, the number of queues opened under the current camera: num_lane_open T +0；
And step four, transmitting the id track of the target data in the set C in the step two into a queuing area queue-region while executing the step three, when the id track of the target data enters the area, assigning a queuing state queue to a flag of the target, determining that the target is in a semi-effective state, counting queuing time according to a tracking frame, and calculating the number of queuing people meeting the condition at the moment: num_ordering_person+1, and accumulating the time of all target data meeting the conditions to obtain acc_queue_time;
A fifth strategy step of transmitting the id track of the target data in the set C in the second strategy step into a processing area checkout-region, when the id track of the target data enters the area and the target flag state is queue, newly giving the target state flag as a checkout state checkout, and counting the checkout time of the target according to the tracking frame from the moment
Figure FDA0003917118210000033
Accumulating the time of all the target data meeting the conditions to obtain acc_checkout_time, wherein the total waiting time acc_wait_time=acc_queue_time+acc_checkout_time;
judging whether the flag state of the id track of the target data is a checkout state checkout or not, if not, exiting;
if yes, judging whether the target leaves the processing area through the id track of the target data, if yes, finishing the settlement of the number num_checkout_person+1, and if no, finishing the settlement of the number num_checkout_person+1;
a policy step seven, obtaining an average queuing length avg_queue_len, an average queuing time avg_queue_time, an average checkout number length avg_checkout_len, an average checkout time avg_checkout_time, a total checkout number num_checkout_person, a total queuing time acc_queue_time, a total checkout time acc_checkout_time, an average waiting time avg_wait_time and an average waiting number avg_wait_len of the queue targets within a certain time period T;
wherein ：
avg_wait_len＝acc_wait_time/T
avg_wait_time＝acc_wait_time/(num_checkout_person+num_queuing_person+num_checkout_person+num_checkouting_person)
avg_queue_len＝acc_queue_time/T
avg_queue_time＝acc_queue_time/(num_checkout_person+num_queuing_person)
avg_checkout_len＝acc_checkout_time/T
avg_checkout_time＝acc_checkout_time/(num_checkout_person+num_checkouting_person)；
a policy step eight, repeating the policy step two to the policy step seven, calculating queuing parameter information of each queue target, obtaining { avg_queue_len1, avg_queue_len2,..avg_queue_lenn }, { avg_queue_time1, avg_queue_time2,..avg_queue_time n }, { avg_checkout_len1, avg_checkout_len2, { avg_checkout_lenn }, { avg_checkout_time1, avg_checkout_time2, { num_checkout_person1, num_checkout_person2, num_person } respectively displaying information of each queue;
strategy step nine: and resetting the related parameters every T time, so as to obtain queuing information in the T time period at the next moment.
4. The prediction method of an intelligent queuing system according to claim 2, further comprising:
the algorithm module executes algorithm steps;
the first algorithm step, the algorithm module acquires head-shoulder data from the image data, and defines the sum of the head-shoulder data as a head-shoulder data set;
training the head-shoulder data set by using a target detection algorithm to obtain a head-shoulder detection model;
detecting the image data to obtain a head-shoulder frame in the image data, and then performing coordinate conversion on the head-shoulder frame to obtain a target frame of target data;
And fourthly, performing target tracking on the target frame transmitted in the third algorithm step by using a tracking algorithm.
5. The prediction method of an intelligent queuing system according to claim 2 wherein in the prediction step three, a model is trained using a SVR machine learning method supporting vector machine regression.
6. The prediction method of an intelligent queuing system according to claim 4 wherein in algorithm step two, the head-shoulder dataset is trained using a deep learning based object detection algorithm YOLOX-M algorithm.
7. The method according to claim 4, wherein in the third algorithm step, the target frame data is (X, Y, W, H), and the target frame data has a calculation formula:
X＝x-w*ratio_x；
Y＝y；
W＝w(1+2*ratio_x)；
H＝h(1+ratio_y)；
wherein x and y respectively represent the coordinates of the initial corner point of the target frame detected by the detection algorithm, and w and h respectively represent the width and the height of the frame.
8. The prediction method of the intelligent queuing system according to claim 7, wherein the value range [0,0.12] of the ratio_x is that the width of the frame of the pedestrian is the width of the original head and shoulder when the parameter of the value is 0, the width of the frame of the pedestrian substantially covers the width of the whole pedestrian when the parameter of the value is 0.12, the value range of the ratio_y is [0,1.6], the height of the frame of the pedestrian is that of the original head and shoulder when the value is 0, and the height of the frame of the pedestrian substantially completely covers all the frames of the pedestrian when the value is 1.6.
9. The prediction method of an intelligent queuing system according to claim 4 wherein in algorithm step four, the deep-learning-based deep_sort tracking algorithm is used to track the target frame entered in algorithm step three.
CN202211343051.5A 2022-10-31 2022-10-31 Intelligent queuing system and prediction method for same Pending CN116187634A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211343051.5A CN116187634A (en) 2022-10-31 2022-10-31 Intelligent queuing system and prediction method for same

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211343051.5A CN116187634A (en) 2022-10-31 2022-10-31 Intelligent queuing system and prediction method for same

Publications (1)

Publication Number Publication Date
CN116187634A true CN116187634A (en) 2023-05-30

Family

ID=86444856

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211343051.5A Pending CN116187634A (en) 2022-10-31 2022-10-31 Intelligent queuing system and prediction method for same

Country Status (1)

Country Link
CN (1) CN116187634A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117968208A (en) * 2024-03-29 2024-05-03 中建安装集团有限公司 Environment system control method and control system

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117968208A (en) * 2024-03-29 2024-05-03 中建安装集团有限公司 Environment system control method and control system

Similar Documents

Publication Publication Date Title
US8582816B2 (en) Method and apparatus for video analytics based object counting
CN111191576B (en) Personnel behavior target detection model construction method, intelligent analysis method and system
CN104123544B (en) Anomaly detection method and system based on video analysis
CN102542289B (en) Pedestrian volume statistical method based on plurality of Gaussian counting models
CN104637058B (en) A kind of volume of the flow of passengers identify statistical methods based on image information
CN103632158B (en) Forest fire prevention monitor method and forest fire prevention monitor system
CN110348312A (en) A kind of area video human action behavior real-time identification method
WO2017122258A1 (en) Congestion-state-monitoring system
CN106991668B (en) Evaluation method for pictures shot by skynet camera
CN107679471B (en) Indoor personnel air post detection method based on video monitoring platform
CN102982313B (en) The method of Smoke Detection
JP2019505866A (en) Passerby head identification method and system
KR20080021804A (en) Target detection and tracking from overhead video streams
CN106339687A (en) Self-adaptive pedestrian street-crossing signal calculating method based on video
CN106981150A (en) A kind of supermarket&#39;s intelligent anti-theft system and method
WO2023155482A1 (en) Identification method and system for quick gathering behavior of crowd, and device and medium
WO2020114116A1 (en) Pedestrian detection method based on dense crowds, and storage medium and processor
WO2021022698A1 (en) Following detection method and apparatus, and electronic device and storage medium
CN109558790B (en) Pedestrian target detection method, device and system
CN115841651B (en) Constructor intelligent monitoring system based on computer vision and deep learning
CN116187634A (en) Intelligent queuing system and prediction method for same
CN114926781A (en) Multi-user time-space domain abnormal behavior positioning method and system supporting real-time monitoring scene
CN110674703A (en) Video tripwire alarm counting method and flow in intelligent monitoring
CN110889347B (en) Density traffic flow counting method and system based on space-time counting characteristics
Song et al. A reliable counting vehicles method in traffic flow monitoring

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