Počítání objektů ve videosekvencích

Abstract

The motive for creation of this thesis was to create a web interface for counting objects in video sequences using Python scripts, that can work with neural networks. The detection results are displayed to the user in the form of graphs, images and videos. The thesis is divided into 4 parts. First part introduces the issue of detection, object tracking and counting the crossing of a line by an object. The second part describes the principle of generating datasets using GAN networks. These are used to simplify data collection for neural network training. The third part deals with the implementation of the web interface. Server is built on ASP.NET Core technology. Client is a PWA created using the Blazor WebAssembly framework. The conclusion is devoted to a comparison of several methods for detecting and tracking objects through the scene.

Description

Subject(s)

Information system, web server, PWA client, Blazor WebAssembly, ASP.NET Core platform, MariaDB, EF Core, Python scripts, running scripts on the server, script settings from the web, TensorFlow, PyTorch, Keras, GAN, SSD, YOLO, DeepSORT, model training, line crossing detection, object tracking, object classification

Citation