Neural Network Acuity Segmentation for Edge Vision Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-resolution image sensors generate large amounts of data, making it inefficient to transmit entire image sets from digital cameras to servers for processing and storage, particularly in applications like real-time manufacturing where latency and energy consumption are critical.
Innovation Solution
Processing image data by differentiating between focal and peripheral regions, using higher acuity in focal regions and lower acuity in peripheries, with customized kernel sizes and stride lengths, and quantization levels to reduce computational workload and energy consumption, implemented in an edge server near the camera.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If entire high-resolution image data is transmitted from digital camera to server for processing, then processing accuracy is maintained, but communication bandwidth usage and energy consumption increase significantly
Solution Approach 1:
The patent divides the image processing task into two segments: focal region processing (high acuity) and peripheral region processing (low acuity). This segmentation allows the system to apply different processing intensities to different regions, reducing overall computational load and energy consumption while maintaining accuracy where needed.
Solution Approach 2:
The patent applies local quality by using higher processing acuity specifically in focal regions where detail is critical, while using lower processing acuity in peripheral regions. This localized differentiation optimizes the balance between processing accuracy and energy consumption by concentrating computational resources only where high precision is necessary.
2Measurement precision
If entire high-resolution image data is transmitted from digital camera to server, then complete image analysis is achieved, but latency increases due to large data transmission
Solution Approach 1:
The patent extracts only the essential features from peripheral regions rather than transmitting complete high-resolution image data. By extracting key information and transmitting only necessary data to the server, the system reduces communication bandwidth usage and latency while maintaining sufficient analysis completeness.
Solution Approach 2:
The patent applies partial action by performing complete high-acuity processing only on focal regions while using reduced-acuity processing on peripheral regions. This partial processing approach reduces the total data volume that needs to be transmitted and processed, thereby reducing latency while maintaining adequate analysis capability.
3Measurement precision
If uniform high acuity processing is applied to entire image, then processing accuracy is maximized, but computational workload and energy consumption increase
Solution Approach 1:
The patent segments the image into focal and peripheral regions, applying different processing acuity levels to each segment. This segmentation enables the system to maintain high processing accuracy in critical areas while reducing computational workload in less critical areas, thereby improving overall processing efficiency.
Solution Approach 2:
The patent implements local quality by varying processing acuity across different regions of the image. High acuity processing is applied locally to focal regions where precision is critical, while low acuity processing is applied to peripheral regions, optimizing the balance between processing accuracy and computational efficiency.
4Measurement precision
If high acuity processing is applied to peripheral regions, then processing accuracy is maintained, but energy consumption and computational workload increase unnecessarily
Solution Approach 1:
The patent applies partial action by using reduced-acuity processing for peripheral regions instead of uniform high-acuity processing. This approach recognizes that high processing accuracy in peripheral regions is often excessive and unnecessary, thereby reducing energy consumption and computational workload while maintaining adequate processing quality.
Data Source
AI summary
Customization of a deep neural network model to analyze different regions of an image at different machine vision acuity levels. A graphical user interface presents an image captured by an image sensing pixel array and receives user interactions with the image to define regions of the machine vision acuity levels. Based on the user interactions with the graphical user interface, a region mask is generated to identify the regions of pixels in the image sensing pixel array. According to the region mask, unnecessary computations of low machine vision acuity are removed from the deep neural network model to generate a customized computing model of analyzing image data, captured by the image sensing pixel array, at the machine vision acuity levels.