Adaptive Video Frame Extraction via Two-Stage Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video frame extraction methods often result in missing or duplicate frames, especially for videos of varying lengths, due to fixed time intervals for sampling, which can lead to inefficient and non-representative thumbnail generation.
Innovation Solution
A two-time sampling method is employed, where a first sampling stride adjusts based on video length to create a key frame timestamp table, and a second sampling stride is determined by the quantity of key frames and to-be-extracted frames, allowing for adaptive extraction of key frames, followed by hard decoding and conversion to generate thumbnails.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a fixed time interval is used for sampling video frames, then the extraction process is simple, but frames may be missing or duplicate for videos of varying lengths
Solution Approach 1:
The patent applies dynamics by making the sampling stride adaptive rather than fixed. The first sampling stride is dynamically adjusted based on video length to generate a key frame timestamp table, and the second sampling stride is dynamically determined based on the quantity of key frames and to-be-extracted frames. This dynamic adjustment ensures reliable frame extraction for videos of varying lengths without causing missing or duplicate frames.
Solution Approach 2:
The patent changes the sampling parameters (strides) based on video characteristics. The first sampling stride is adjusted according to video length, and the second sampling stride is determined by the quantity of key frames and to-be-extracted frames. This parameter adaptation resolves the contradiction by making the extraction process reliable for different video lengths while maintaining a relatively simple overall structure.
2Adaptability or versatility
If a fixed sampling method is used for all videos, then the processing is efficient, but the thumbnail representation is poor for videos of different lengths
Solution Approach 1:
The patent implements adaptability through dynamic sampling strides while maintaining efficiency. The first sampling stride adapts to video length for generating the key frame timestamp table, and the second sampling stride adapts to the desired number of thumbnails. This dynamic approach ensures good thumbnail representation for videos of different lengths without significantly compromising processing efficiency.
Solution Approach 2:
The patent segments the sampling process into two distinct stages: first sampling to generate the key frame timestamp table with adaptive stride based on video length, and second sampling to extract specific frames based on desired quantity. This segmentation allows each stage to be optimized independently, achieving both adaptability and efficiency.
3Productivity
If traditional frame extraction methods are used, then the process is straightforward, but processing time is excessive
Solution Approach 1:
The patent segments the extraction process into two phases: first sampling to create a key frame timestamp table (reducing the search space), and second sampling to extract the actual frames needed. This segmentation dramatically improves processing speed by avoiding exhaustive search while maintaining a relatively simple overall methodology.
Solution Approach 2:
The patent performs preliminary action by first generating the key frame timestamp table before extracting the actual frames. This preliminary step organizes the data structure and identifies key frame positions, enabling faster subsequent extraction and improving overall processing speed without adding excessive complexity.
Data Source
AI summary
A video frame extraction method performed by a computing device includes: obtaining a key frame timestamp table of a video based on a first sampling stride, the first sampling stride being adjusted according to a length of the object video; obtaining a timestamp of a key frame to be extracted from the key frame timestamp table based on a second sampling stride, the second sampling stride being determined according to a quantity of key frame timestamps in the key frame timestamp table and a quantity of to-be-extracted key frames; reading, according to the timestamp, frame data of the key frame; and generating, based on a hard decoding result of the frame data, a thumbnail corresponding to the key frame.