This document presents a project that aims to automatically detect major events in soccer games by analyzing video. The system uses a Perception Concepts-Final State Machine (PC-FSM) model to first divide game footage into individual scenes using cut/dissolve detection. It then analyzes the scenes for recognized patterns that correspond to semantic events like goals or near goals. Detected events are stored in a database along with metadata and fed into a final state machine for semantic event detection. A graphical user interface allows users to view detected significant events. The document outlines the project goals, block diagram of the system, execution results of key functions, and outcomes including the creation of a database and user interface.
A Semantic content detection for soccer video based on finite state machine -Presentation
1. By Anan Atila
B.Sc. Electrical Engineer
Image and Video Processing Project
24.08.2015
2. Executive Summary
Soccer games consist of several major events.
Automatic detection of these events can allow the viewer
to choose a summary of the events that interest him/her
specifically. Repetitive behavior patterns at the match
shall be described in terms of the graphic state of the
scene, and thus the major events shall be detected
automatically.
3. Aims and Goals
The aim of the project is to propose model for analysis and detection of
semantic contents in soccer games, which relies on perception concepts and a
final state machine.
PC-FSM Model
PC – Perception Concepts
FSM – Final State Machine
To meet the requirements of the project I chose to implement the system with
several primary functions:
1) Cut/dissolve detection function – scans the video , and then switches between
cameras are identified in order to divide the game into individual scenes.
2) PCs detection function - review the types of scenes that were detected in while
searching for recognized patterns of semantic events.
3) Near goal detection function – returns goalpost-proximity parameter.
4. The data gathered during the processing are stored in a
database that contains information about the various
scenes. This information includes the scene times,
duration and its goalpost-proximity parameter. These data
are entered into a final state machine, and a detection of
semantic events is conducted. The results of the
detection are displayed through the user interface, while
enabling the user to view the significant semantic events
detected.
Aims and Goals (cont.)