Video Generation Apparatus Using Intention Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video generation methods fail to consider user intentions, resulting in poorly continuous and unnatural content, making it difficult to satisfy user requirements, as they randomly select and integrate videos without analyzing semantic information.
Innovation Solution
A method that extracts intention features from video generation requests and generates target videos by filtering and fusing video clips based on object and action recognition, using neural networks to determine relevant clips and adaptively segment action clips according to user-defined lengths, ensuring the generated video aligns with user intentions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If video clips are randomly selected and integrated without analyzing semantic information, then video generation can be performed, but the generated video content is poorly continuous and unnatural
Solution Approach 1:
The patent segments video clips based on detected actions and objects, dividing the video into meaningful units that can be logically ordered. This segmentation enables continuous and natural video content by ensuring that clips are grouped according to their semantic relationships rather than being randomly selected.
Solution Approach 2:
The patent performs preliminary analysis of video clips to detect actions and objects before integration. This preliminary action includes extracting semantic information, identifying action boundaries, and categorizing clips, which ensures that the subsequent video generation process produces continuous and natural content.
2Reliability
If professional video editing software is used to edit videos, then video quality can be improved, but the software is too complex for users to manipulate
Solution Approach 1:
The patent implements automatic video generation that performs editing tasks without user intervention. The system automatically detects actions and objects in video clips, segments the clips appropriately, and integrates them into a continuous video, thereby providing professional-quality editing results while maintaining ease of use.
Solution Approach 2:
The patent changes the parameter of user interaction from manual control to automatic processing. By transforming the editing process into an automated system that operates based on detected semantic parameters (actions and objects), the patent achieves high video quality without requiring users to master complex editing software.
3Adaptability or versatility
If massive videos are stored for user selection, then video content variety can be increased, but content repetition and difficulty in finding certain content increase
Solution Approach 1:
The patent implements a feedback mechanism through action and object detection that provides structured information about video content. By analyzing and tagging clips with detected actions and objects, the system creates an organized database that enables efficient content retrieval while maintaining video variety.
Solution Approach 2:
The patent applies multi-functionality by using action and object detection for multiple purposes: segmenting video clips, organizing content for variety, and enabling searchability. This universal approach to semantic analysis simultaneously addresses content variety and search efficiency without requiring separate systems.
Data Source
AI summary
A video generation method includes obtaining action clips into which source videos are split, through action recognition with respect to the source videos, selecting target clips from among the action clips, based on correlation between clip features of at least some of the action clips and an intention feature extracted from a video generation request, and generating a target video by combining at least some of the target clips.