Large meeting place seat positioning method based on deep learning

Abstract

The invention provides a large-scale meeting place seat positioning method based on deep learning, which is used for positioning seats with different scale proportions in a large-scale meeting place video stream, and is characterized by comprising the following steps: step S1, obtaining a video stream of an overhead view angle of a meeting place to be detected; s2, adopting a network combined by ResNet50 and FPN as a Backbone, and connecting the Backbone to an FAM module and an ODM module to construct an initial convolutional neural network; s3, training the initial convolutional neural network based on a stochastic gradient descent method and a minimization loss function to obtain a large meeting place seat detection model; and S4, carrying out seat detection on an image in the video stream every set frame by adopting a large meeting place seat detection model, and outputting positioning detection results corresponding to all seats in the meeting place to be detected. According to the invention, seat targets with various sizes, angles and proportions in different color modes, such as indoor light and night city infrared light, shot by different camera positions can be efficiently identified, and the method has the advantages of strong generalization and high precision.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a method for locating seats in a large venue based on deep learning. Background technique [0002] Scientific statistical methods of meeting venue data and intelligent methods of maintaining order of meeting venues are indispensable links in the modernization of the conference management industry. The large venue is a place for various activities such as large-scale lectures, performing arts performances and regular large-scale conferences. For performing arts activities, box office data is an important criterion for income statistics, and it is an intuitive representation of the success of artistic activities; for lectures and speech activities, the real attendance rate, audience status during the performance, such as the number of departures, etc. Data is an important basis for evaluating the audience experience from more dimensions, and provides an important referen...

AI Technical Summary

Problems solved by technology

[0003] Seat positioning is an important part of large-scale venue data collection. The traditional expert system-based seat positioning for large venues is highly dependent on manual presets. Due to the large number of seats, manual input of information is very time-consuming, and the preset information is difficult to adapt to. The position of the camera changes, and the maintenance cost is high

With the continuous development of computer vision technology, especially the excellent performance of deep learning in the field of target detection, it has been widely used in many fields such as smart medical care and automatic driving. The seat positioning method based on deep convolutional neural network has the advantages of easy Deployment, ease of migration, and accurate positioning have multiple advantages. However, most target detection methods do not have the ability to frame oblique targets, and perform poorly on venue seat targets with highly diverse attributes such as size and scale. Improvements to target detection methods are needed.

Images