Video Identification via Image and Optical Flow Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video identification methods fail to effectively improve accuracy by not considering changes between adjacent frames, leading to suboptimal video recognition performance.
Innovation Solution
A video identification method that extracts images and optical flows from videos, using a first machine learning model with a larger depth than a second model to classify images and optical flows, respectively, and fuses the classification results to enhance video identification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single machine learning model is used to classify video frames directly, then the identification process is simple and fast, but the accuracy is insufficient because changes between adjacent frames are not considered
Solution Approach 1:
The patent segments the video identification task into two independent classification problems: one for image content and another for optical flow. Each problem is handled by a dedicated machine learning model, allowing each model to be optimized for its specific input type while working together to achieve higher overall accuracy.
Solution Approach 2:
The patent merges the classification results from two different machine learning models (image classifier and optical flow classifier) through a fusion mechanism. This combination allows the system to leverage complementary information from both models to achieve superior video identification accuracy compared to using a single model.
2Measurement precision
If a deep machine learning model is used to extract more feature information, then the identification accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent divides the feature extraction task into two separate streams: one for image features and another for optical flow features. Each stream uses a machine learning model with appropriate depth for its specific requirements, avoiding the need for an overly deep model to handle both tasks simultaneously.
Solution Approach 2:
The patent applies machine learning models with different depths to different tasks based on their specific requirements. The image classification model uses a deeper architecture to capture complex visual patterns, while the optical flow model uses a shallower architecture sufficient for motion analysis, optimizing the balance between accuracy and computational cost.
Data Source
AI summary
The present disclosure provides a video identification method, a video identification device and a storage medium. The video identification device extracts an image and an optical flow from a video, classifies the image by using a first machine learning model to obtain a first classification result, classifies the optical flow by using a second machine learning model to obtain a second classification result, wherein a depth of the first machine learning model is larger than a depth of the second machine learning model, and fuses the first classification result and the second classification result to obtain the identification result of the video.