Semantic Concept Mining Based on Hierarchical Event Detection for Soccer Video Indexing

Area of Science:

• Computer Science
• Artificial Intelligence
• Multimedia Analysis

Background:

• Automated video indexing and summarization are crucial for managing large sports archives.
• Existing methods often rely on shot detection and clustering, which can be inefficient and lose temporal context.

Purpose of the Study:

• To develop a novel automated system for semantic labeling of broadcast soccer video sequences.
• To improve the accuracy and efficiency of sports video analysis through a new hierarchical classification framework.

Main Methods:

• A hierarchical, top-down video scene classification approach is proposed, avoiding traditional shot clustering.
• Sequential association mining and a novel sequential association distance are used to classify event sequences into semantic concepts (e.g., goals, saves).

Main Results:

• The proposed method accurately detects and classifies soccer events, maintaining temporal order.
• The system successfully extracts excitement clips and assigns semantic labels, enabling highlight generation.

Conclusions:

• The novel framework offers improved accuracy and temporal coherence for soccer video analysis.
• This approach facilitates efficient video retrieval, highlight extraction, and video skimming for sports content.