Video Recognition Model Switching with Pre-Input Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video recognition systems face challenges in accurately recognizing events due to changes in video recognition environments, leading to potential decreases in recognition accuracy when switching between recognition models.
Innovation Solution
A video processing system that includes first and second video analysis models and a switching mechanism to transition between them based on changes in recognition environments, with pre-inputting data preceding the switching timing to maintain recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If video analysis models are switched according to changes in video recognition environment, then adaptability to different environments is improved, but recognition accuracy may decrease during the switching transition
Solution Approach 1:
The patent applies preliminary action by inputting video data before the actual switching timing to the destination model in advance. This allows the destination model to process and analyze the video data ahead of time, so that when switching occurs, the model is already ready to continue recognition without interruption or accuracy loss. The switching control unit determines the switching timing based on environment changes and proactively prepares the destination model by feeding it video data preceding the switching timing.
2Reliability
If video data is input to both source and destination models during switching, then recognition continuity is improved, but processing load increases
Solution Approach 1:
The patent applies partial action by inputting only a specific portion of video data (data preceding the switching timing by a predetermined period) to the destination model, rather than continuously inputting all video data to both models simultaneously. This selective approach ensures sufficient data is available for continuous recognition while avoiding excessive processing load that would result from duplicating all video processing to both models throughout the entire switching period.
Data Source
AI summary
A video processing system (10) includes a recognition model (M1) that analyzes a video corresponding to a first video recognition environment; a recognition model (M2) that analyzes a video corresponding to a second video recognition environment; and a switching unit (11) that switches a recognition model for analyzing video input data to be input, from the recognition model (M1) to the recognition model (M2) according to a change from the first video recognition environment to the second video recognition environment in the video input data, in which the switching unit (11) inputs video input data including data preceding a switching timing by a predetermined period to the recognition model 10 (M2) according to the change from the first video recognition environment to the second video recognition environment in the video input data.