CN-121999410-A - Exhibition hall management method and device

CN121999410ACN 121999410 ACN121999410 ACN 121999410ACN-121999410-A

Abstract

The invention discloses a method and a device for managing an exhibition hall, which belong to the technical field of computer vision, wherein the method comprises the steps of collecting video data in the exhibition hall; the method comprises the steps of determining personnel detection frames of each tracking target in a current image frame, generating a plurality of personnel prediction frames according to the personnel detection frames and a previous image frame, respectively calculating the intersection ratio between each personnel detection frame and each personnel prediction frame to generate an IOU cost matrix, respectively extracting human depth characteristics of each personnel detection frame and each personnel prediction frame to generate a characteristic cost matrix, carrying out weighted fusion on the IOU cost matrix and the characteristic cost matrix to generate a fusion cost matrix, determining the personnel prediction frames of each tracking target by solving the minimum cost matching of the fusion cost matrix, and carrying out exhibition hall management according to motion state parameters corresponding to the personnel prediction frames of each tracking target. Therefore, by implementing the invention, the problem of low tracking accuracy of exhibition hall personnel in the prior art can be solved.

Inventors

Ma Guanxiong
Duan Qiyou
GAO XIANG
HAN MENGYUAN
HUANG MENGHUA
HU YIFAN
LIU RUOPING
LUO YING
QUAN YUE
WANG XINYI
WANG YIGEN
DENG CHURAN
WU LIYA
XU CHENGHAO
XU JIANWEN
PANG PENG
Wan chan
PAN HUI
LUO XUAN
ZHAO SHUANG
Zeng Yexiang
CHENG RAN

Assignees

广东电网有限责任公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. A method of exhibition hall management, comprising: Collecting video data in an exhibition hall, wherein the video data comprises a current image frame and a last image frame; Determining a plurality of personnel detection frames of a tracking target in the current image frame, and generating a plurality of personnel prediction frames according to a plurality of personnel detection frames and the previous image frame; respectively calculating the intersection ratio between each personnel detection frame and each personnel prediction frame to generate an IOU cost matrix; Extracting human depth features of the personnel detection frames and the personnel prediction frames respectively, and calculating cosine distances between the personnel detection frames and the personnel prediction frames according to the human depth features to generate a feature cost matrix; Performing weighted fusion on the IOU cost matrix and the characteristic cost matrix to generate a fusion cost matrix; determining personnel prediction frames of all the tracking targets by solving the minimum cost matching of the fusion cost matrix; And carrying out exhibition hall management according to the motion state parameters corresponding to the personnel prediction frames of the tracking targets.
2. The booth management method of claim 1, wherein the generating a number of personnel prediction frames from a number of the personnel detection frames and the last image frame comprises: determining the motion state parameters of each personnel detection frame; Calculating a plurality of predicted motion state parameters according to the motion state parameters of each personnel detection frame and a preset state transition matrix; And generating a plurality of personnel prediction frames according to a plurality of the predicted motion state parameters.
3. The method of claim 2, wherein the calculating the intersection ratio between each of the person detection frames and each of the person prediction frames, respectively, generates an IOU cost matrix, comprising: Respectively calculating the detection frame area of each personnel detection frame, the prediction frame area of each personnel prediction frame and the intersection area between the detection frame area of each personnel detection frame and each personnel prediction frame; calculating the intersection ratio between each personnel detection frame and each personnel prediction frame according to the detection frame area, the prediction frame area and the intersection area; And generating an IOU cost matrix according to the intersection ratio between each personnel detection frame and each personnel prediction frame.
4. A method of managing a hall according to claim 3, wherein the cosine distance between each of the person detection frames and each of the person prediction frames is calculated using the following formula: in the formula, A cosine distance between the personnel detection frame i and the personnel prediction frame j; detecting human depth characteristics of the frame i for a person; human depth features of frame j are predicted for the person.
5. The method of claim 4, wherein the IOU cost matrix and the feature cost matrix are weighted and fused to generate a fused cost matrix using the following formula: in the formula, The cost matrix is fused; the cost matrix is IOU; The characteristic cost matrix; is a preset weight coefficient.
6. The method of claim 5, wherein determining the personnel prediction box for each tracking target by solving for a minimum cost match of the fusion cost matrix comprises: Solving the fusion cost matrix by using a Hungary algorithm with the minimum total cost of the fusion cost matrix as a target to obtain an optimal matching combination of a personnel detection frame and a personnel prediction frame; And for each optimal matching combination, determining the tracking personnel corresponding to the personnel detection frame as the tracking personnel corresponding to the personnel prediction frame so as to determine the personnel prediction frame of each tracking target.
7. The method according to claim 6, wherein when the exhibition is managed as a traffic system, the performing the exhibition management according to the motion state parameter corresponding to the personnel prediction frame of each tracking target comprises: Determining the motion direction and real-time coordinates of each tracking target according to the motion state parameters corresponding to the personnel prediction frames of each tracking target; Determining regional population of a plurality of exhibition hall regions based on real-time coordinates of each tracking target and the preset plurality of exhibition hall regions; Determining the total number of people in the exhibition hall according to the motion direction and real-time coordinates of each tracking target and a preset exhibition hall entrance detection line; Uploading the regional population and the headcount of each exhibition hall region to an exhibition hall management platform.
8. The method according to claim 6, wherein when the exhibition management is the robot interaction, the exhibition management according to the motion state parameter corresponding to the personnel prediction frame of each tracking target comprises: determining real-time coordinates of each tracking target according to the motion state parameters corresponding to the personnel prediction frames of each tracking target; collecting real-time robot coordinates of a plurality of preset robots; calculating the real-time distance between each tracking target and each preset robot according to the real-time coordinates of each tracking target and the real-time robot coordinates of each preset robot; when a real-time distance smaller than a preset distance threshold exists, acquiring real-time face data of a tracking target corresponding to the real-time distance; Matching the real-time face data by using a preset honored guest database, and identifying the identity information of the tracking personnel; When the identity information of the tracking personnel is a honored guest, controlling a preset robot to output customized greeting voice and customized interaction actions; when the identity information of the tracking person is a common visitor, controlling a preset robot to output welcome voice and displaying an interaction page.
9. The booth management method of claim 1, further comprising: collecting real-time temperature, real-time humidity and real-time illumination intensity in an exhibition hall; controlling the air conditioner setting in the exhibition hall according to the real-time temperature and the real-time humidity; and controlling the light setting in the exhibition hall according to the real-time illumination intensity.
10. The exhibition hall management device is characterized by comprising a data acquisition module, a prediction frame generation module, an intersection ratio calculation module, a feature extraction module, a data fusion module, a prediction frame matching module and an exhibition hall management module; the data acquisition module is used for acquiring video data in the exhibition hall, wherein the video data comprises a current image frame and a last image frame; The prediction frame generation module is used for determining a plurality of personnel detection frames of tracking targets in the current image frame and generating a plurality of personnel prediction frames according to the personnel detection frames and the previous image frame; The cross-over ratio calculation module is used for calculating the cross-over ratio between each personnel detection frame and each personnel prediction frame respectively and generating an IOU cost matrix; The feature extraction module is used for extracting human depth features of the personnel detection frames and the personnel prediction frames respectively, calculating cosine distances between the personnel detection frames and the personnel prediction frames according to the human depth features, and generating a feature cost matrix; The data fusion module is used for carrying out weighted fusion on the IOU cost matrix and the characteristic cost matrix to generate a fusion cost matrix; The prediction frame matching module is used for determining personnel prediction frames of the tracking targets by solving the minimum cost matching of the fusion cost matrix; And the exhibition hall management module is used for carrying out exhibition hall management according to the motion state parameters corresponding to the personnel prediction frames of the tracking targets.

Description

Exhibition hall management method and device Technical Field The invention relates to the technical field of computer vision, in particular to a method and a device for managing an exhibition hall. Background The exhibition hall is used as an important place for product display, information exchange and customer reception, and the quality requirement of the management process is increasingly outstanding along with the continuous improvement of the intelligent requirement. However, the existing exhibition hall management technical scheme still has obvious limitation in the aspects of multi-dimensional function integration and efficiency optimization, and is difficult to meet the dual requirements of efficient operation and high-quality service of the exhibition hall. Specifically, in the aspect of personnel management and passenger flow statistics, the existing exhibition hall is used for assisting personnel identity confirmation and passenger flow counting in a manual mode, time and labor are consumed, data instantaneity is poor, and personnel motion tracks cannot be effectively tracked through technical means to achieve accurate in-out direction judgment and regional passenger flow statistics, so that managers are difficult to quickly master personnel distribution and flow conditions in the exhibition hall, and management strategies cannot be timely adjusted to optimize visiting experience. Disclosure of Invention The invention provides a method and a device for managing an exhibition hall, which can solve the problem of low tracking accuracy of exhibition hall personnel in the prior art. In order to solve the technical problems, the invention provides a management method of an exhibition hall, which comprises the following steps: Collecting video data in an exhibition hall, wherein the video data comprises a current image frame and a last image frame; Determining a plurality of personnel detection frames of a tracking target in the current image frame, and generating a plurality of personnel prediction frames according to a plurality of personnel detection frames and the previous image frame; respectively calculating the intersection ratio between each personnel detection frame and each personnel prediction frame to generate an IOU cost matrix; Extracting human depth features of the personnel detection frames and the personnel prediction frames respectively, and calculating cosine distances between the personnel detection frames and the personnel prediction frames according to the human depth features to generate a feature cost matrix; Performing weighted fusion on the IOU cost matrix and the characteristic cost matrix to generate a fusion cost matrix; determining personnel prediction frames of all the tracking targets by solving the minimum cost matching of the fusion cost matrix; And carrying out exhibition hall management according to the motion state parameters corresponding to the personnel prediction frames of the tracking targets. Preferably, the generating a plurality of personnel prediction frames according to a plurality of personnel detection frames and the previous image frame includes: determining the motion state parameters of each personnel detection frame; Calculating a plurality of predicted motion state parameters according to the motion state parameters of each personnel detection frame and a preset state transition matrix; And generating a plurality of personnel prediction frames according to a plurality of the predicted motion state parameters. Preferably, the calculating the intersection ratio between each person detection frame and each person prediction frame, and generating the IOU cost matrix include: Respectively calculating the detection frame area of each personnel detection frame, the prediction frame area of each personnel prediction frame and the intersection area between the detection frame area of each personnel detection frame and each personnel prediction frame; calculating the intersection ratio between each personnel detection frame and each personnel prediction frame according to the detection frame area, the prediction frame area and the intersection area; And generating an IOU cost matrix according to the intersection ratio between each personnel detection frame and each personnel prediction frame. As a preferred solution, the cosine distance between each of the person detection frames and each of the person prediction frames is calculated using the following formula: in the formula, A cosine distance between the personnel detection frame i and the personnel prediction frame j; detecting human depth characteristics of the frame i for a person; human depth features of frame j are predicted for the person. As a preferred scheme, the following formula is adopted to perform weighted fusion on the IOU cost matrix and the characteristic cost matrix, so as to generate a fusion cost matrix: in the formula, The cost matrix is fused; the cost matrix is IOU; The characteris