Parallel Image Recognition Device for Real-Time 4K Video Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition systems are not fast enough for real-time object detection in high-frame-rate video streams, such as 4K or 8K videos, due to the large workload required for object detection in each video frame.
Innovation Solution
An image recognition device comprising a first processing module for pre-processing video streams into an image queue and a second processing module that uses multiple processors and image recognition models to detect objects in parallel, with the ability to adjust image parameters for improved recognition speed, utilizing a communication bus for efficient data processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If object detection is performed for each video frame at 30 FPS or higher in 4K/8K video, then real-time recognition speed is required, but current image recognition systems cannot handle the large workload fast enough
Solution Approach 1:
The system segments the image recognition workload by dividing the video stream processing into multiple independent processing threads, each handling specific image frames. This allows parallel processing of different frames simultaneously, increasing overall productivity while maintaining high recognition speed for each individual frame.
Solution Approach 2:
The patent introduces a multi-dimensional processing architecture by adding thread-level parallelism and processor-level parallelism. Instead of single-threaded sequential processing, the system creates multiple threads that can execute simultaneously on multiple processors, effectively utilizing the dimensional space of parallel computation to handle large workloads at high speed.
2Productivity
If multiple processors are used for parallel processing of image frames, then productivity increases, but device complexity increases
Solution Approach 1:
The system creates universal processing threads that can handle multiple types of image frames and workloads. Each processing thread is designed to be multi-functional, capable of processing different video frames with various parameters and requirements, thereby reducing the need for specialized components and simplifying the overall system architecture while maintaining high productivity.
Solution Approach 2:
The processing threads autonomously manage their own execution and resource allocation. Each thread independently handles its assigned image frames, performs necessary processing operations, and manages memory resources without requiring constant external coordination. This self-service mechanism reduces system complexity by eliminating the need for complex inter-thread communication and synchronization protocols.
Data Source
AI summary
A method for detecting and recognizing objects in images includes obtaining a video stream and pre-processing the video stream to obtain an image queue arranged in a frame playing order and storing the image queue into a storage device. An image frame of the image queue from the storage device is read and at least one object in the image frame is detected and recognized. An image recognition device is also provided.