Segmented Video Retrieval Using Key Frames and Similarity Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video retrieval methods rely on unreliable human-added metadata, leading to inefficiencies and high costs in processing vast amounts of video data, as they fail to comprehensively recognize and process various information such as video and temporal context.
Innovation Solution
A method and apparatus utilizing machine learning to generate retrieval query vectors and video retrieval vectors based on key frame information, enabling comparison and selection of target retrieval video data through similarity scoring and temporal adjacency analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional retrieval methods use human-added metadata (hashtags, titles, descriptions) to retrieve video information, then the retrieval process is simple to implement, but the retrieval quality and reliability deteriorate due to dependence on unreliable metadata
Solution Approach 1:
The patent replaces the mechanical/manual process of human-added metadata with an automated machine learning system. The ML model automatically generates retrieval vectors from video content, substituting the manual metadata creation process and eliminating its reliability issues while maintaining operational simplicity through automation.
Solution Approach 2:
The video retrieval system performs self-service by automatically analyzing video content and generating retrieval vectors without human intervention. The machine learning model independently processes video data, extracts features, and creates searchable representations, making the system self-sufficient and eliminating dependence on unreliable human-added metadata.
2Productivity
If conventional methods process exponentially increasing video data using traditional metadata-based approaches, then the processing cost and time increase significantly, but the comprehensive recognition capability remains limited
Solution Approach 1:
The patent segments video data into smaller manageable units and processes them through the machine learning model. By dividing the exponentially increasing video data into segments that can be efficiently encoded and stored as retrieval vectors, the system achieves scalable processing capacity without linearly increasing processing time or costs.
Solution Approach 2:
The patent transforms video data from its original complex multi-modal format into compressed retrieval vectors through machine learning encoding. This parameter transformation changes the data representation from raw video frames and audio to condensed feature vectors, enabling efficient storage and rapid retrieval while maintaining comprehensive recognition capabilities.
3Device complexity
If video retrieval systems rely on hashtags and video titles for information retrieval, then the system complexity remains low, but the ability to comprehensively recognize visual and temporal context deteriorates
Solution Approach 1:
The machine learning model performs multiple functions simultaneously: it extracts visual features, temporal context, and semantic information from video content, and generates a unified retrieval vector that encompasses all these aspects. This multi-functional approach comprehensively recognizes visual and temporal context while maintaining manageable system complexity through a single integrated model.
Data Source
AI summary
In order to implement the foregoing object, an exemplary embodiment of the present disclosure discloses a video retrieval method performed by a computing device. The video retrieval method may include: receiving retrieval query data for one or more video data; generating a retrieval query vector based on the received retrieval query data; comparing the retrieval query vector with a plurality of video retrieval vectors each representing target retrieval video data segmented based on key frame information; and selecting one or more target retrieval video data by the result of the comparison between the retrieval query vector and the plurality of video retrieval vectors.