Video Retrieval Using Semantic Concept Group Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video retrieval methods based on semantic concept detection struggle with accuracy when handling complex video query information that includes scene changes, as they typically consider simple semantic concepts separately, leading to inadequate retrieval results.
Innovation Solution
The method determines a semantic concept group sequence from video query information, incorporating both semantic and sequential information to retrieve videos, which reflects the overall correlation degree between the query and the video content, including scene changes and other sequential information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If video retrieval is performed based on simple semantic concepts separately, then the retrieval process is simple, but the retrieval accuracy is insufficient for complex video queries
Solution Approach 1:
The patent segments the video retrieval process into multiple hierarchical levels: simple semantic concept detection, semantic concept group formation, and sequence-level matching. Each level handles specific aspects of the query, allowing the system to maintain operational simplicity while improving accuracy through structured multi-stage processing
Solution Approach 2:
The patent adds a temporal sequence dimension to traditional semantic concept matching. By introducing semantic concept groups with sequential relationships and matching them against video sequences, the system transitions from static concept matching to dynamic sequence-aware retrieval, significantly improving accuracy for complex queries involving scene changes
2Device complexity
If multiple semantic concepts are considered separately, then the computational complexity is low, but the ability to capture scene changes and sequential information is insufficient
Solution Approach 1:
The patent merges multiple semantic concepts into semantic concept groups that preserve both individual concept meanings and their sequential relationships. By combining detection results of multiple simple semantic concepts into grouped sequences and matching them against video sequences, the system captures scene changes and temporal information without proportionally increasing computational complexity
Solution Approach 2:
The patent performs preliminary detection of simple semantic concepts and organizes them into semantic concept groups with sequential relationships before the actual video retrieval process. This preprocessing step structures the query information in advance, enabling efficient sequence-level matching during retrieval while avoiding the need for complex real-time analysis
Data Source
AI summary
The present application discloses various video retrieval methods and apparatuses. One of the video retrieval methods comprises: determining a semantic concept group sequence according to video query information, the semantic concept group sequence comprising: at least two semantic concept groups and sequential information between different semantic concept groups therein, and each semantic concept group in the at least two semantic concept groups comprising at least one simple semantic concept; and retrieving videos at least according to the semantic concept group sequence. The technical solution provided in the present application can better meet actual application demands of complicated query of videos.