Heterogeneous Image Recognition Model Switching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image recognition systems in broadcasting environments can only provide a single type of image recognition service at a time, limiting the ability to simultaneously process and offer diverse image recognition services, such as person and object recognition, within a single operating environment.
Innovation Solution
A method and apparatus that enable switching between different image recognition models and services based on characteristic, purpose, and time information, allowing for simultaneous provision of heterogeneous image recognition types in a single image recognition service environment by using a remote image transceiver, model executor, analyzer, camera processing, controller, and related content storage to preprocess and synchronize image recognition models for content and camera-based recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single image recognition model is used in a single processor environment, then the system structure is simple and easy to operate, but only one type of image recognition service can be provided at a time, limiting service diversity
Solution Approach 1:
The patent divides the image recognition service into multiple sections, each handling a specific recognition type (e.g., person recognition, object recognition). The processor is segmented to handle different recognition models sequentially through time-division multiplexing, allowing diverse services without requiring multiple physical processors.
Solution Approach 2:
The system dynamically switches between different image recognition models based on service requirements. The processor transitions between different recognition types (person, object, scene) by loading and executing different models as needed, enabling versatile service provision from a single processor.
2Adaptability or versatility
If multiple image recognition models are loaded simultaneously, then heterogeneous image recognition services can be provided, but memory consumption and processing overhead increase
Solution Approach 1:
The system employs periodic switching between different image recognition models rather than loading them simultaneously. Each model is activated in periodic intervals based on service demand, allowing the system to provide diverse recognition services while maintaining low memory consumption by having only one model actively loaded at any given time.
Solution Approach 2:
When switching between different image recognition models, the system discards the currently loaded model and recovers/loads the next required model. This approach minimizes memory usage by ensuring only one model resides in memory at a time, while still providing access to multiple recognition types through sequential model switching.
Data Source
AI summary
Provided is a method and apparatus for providing a section divided heterogeneous image recognition service in a single image recognition service operating environment. An apparatus for providing a section divided heterogeneous image recognition service in a single image recognition service operating environment includes a model executor configured to switch a single model of a first image recognition model and a second image recognition model to provide a heterogeneous image recognition service in a single image recognition service operating environment and to synchronize the switched model with the analyzer; the analyzer configured to extract and identify a target from an image acquired according to a model synchronized with the model executor and to transmit related content-related information to a related content storage and a transmitter based on extraction and identification results; a camera processing configured to preprocess an image acquired through the camera and to transmit the preprocessed image to the analyzer; and a controller configured to control the model executor, the analyzer, the camera processing, and the remote image transceiver based on service requirements input from a user.