Environment Matching Object Recognition via Key Frame Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep-learning-based object recognition technologies face challenges in adapting to new environments without requiring extensive retraining and the use of expensive GPU equipment, as they are specialized for specific scenes and struggle with efficiency in domain adaptation processes.
Innovation Solution
An object recognition apparatus and method that automatically selects a pre-trained recognition model based on environment matching by extracting key frames from input videos using adversarial autoencoders and similarity-matching networks, allowing for scene adaptation without additional learning processes, and preprocesses input videos to match the characteristics of training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep-learning-based object recognition is applied to actual fields, then high performance in object recognition is achieved, but expensive GPU equipment is required increasing expenses
Solution Approach 1:
The patent segments the deep learning system into multiple specialized recognition models, each trained for specific environments or conditions. This allows the system to use simpler, more efficient models for specific tasks rather than requiring a single large-scale model, reducing GPU requirements while maintaining recognition performance.
Solution Approach 2:
The system performs preliminary actions by pre-training multiple recognition models for different environments and conditions before actual deployment. During runtime, the appropriate pre-trained model is selected based on the current environment, eliminating the need for expensive real-time training and reducing computational requirements.
2Reliability
If deep-learning algorithm is trained with previously given training data, then the algorithm is optimized for that data, but it cannot exhibit expected performance when applied to completely different scenes
Solution Approach 1:
The patent creates a universal recognition system that functions across multiple environments by training multiple specialized models for different scenes, weather conditions, and environments. The system selects the appropriate model based on current conditions, achieving both high accuracy for specific scenes and versatility across different environments.
Solution Approach 2:
The system changes parameters by selecting different pre-trained models based on environmental parameters such as scene type, weather conditions, and lighting. This allows the system to adapt to completely different scenes without retraining, maintaining high recognition accuracy across varied conditions.
3Adaptability or versatility
If domain adaptation scheme is used to allow source domain to approach the right answer, then the algorithm can handle target domain data, but an inefficient process for collecting data and relearning is required
Solution Approach 1:
The patent performs domain adaptation in advance by pre-training multiple recognition models on source domain data for various environments and conditions. During deployment, the system directly applies the appropriate pre-adapted model without requiring real-time data collection or relearning, dramatically improving efficiency while maintaining domain adaptation capabilities.
Data Source
AI summary
Disclosed herein are an object recognition apparatus and method based on environment matching. The object recognition apparatus includes memory for storing at least one program, and a processor for executing the program, wherein the program performs extracting at least one key frame from a video that is input in real time, determining a similarity between the key frame extracted from the input video and each of videos used as training data of prestored multiple recognition models, based on a pretrained similarity-matching network, and selecting a recognition model pretrained with a video having a maximal similarity to the key frame extracted from the input video, preprocessing the input video such that at least one of color and size of a video used as training data of an initial model is similar to that of the input video, and recognizing the preprocessed video based on the initial model.