Neural Network Training With Absolute Luminance Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional image processing pipelines in machine vision tasks, such as those used in self-driving vehicles and robotics, lack accurate luminance measurements, leading to inaccurate scoring during deep learning model training and reduced accuracy at the inference stage.
Innovation Solution
Incorporating calibration and calculation steps in the image signal processing pipeline to generate absolute luminance and radiance values for each pixel, enabling better object discrimination and improved light measurement across varying illumination conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image processing pipelines are used, then the processing is simple and fast, but the luminance measurement accuracy is poor leading to inaccurate object detection
Solution Approach 1:
The patent applies preliminary action by performing calibration of the image sensor before actual image processing. The calibration process establishes lookup tables and correction factors that are stored and applied during subsequent image processing operations. This preliminary calibration step ensures accurate luminance measurements without adding complexity to the real-time processing pipeline, as the complex calibration computations are performed once beforehand rather than continuously during operation.
2Reliability
If luminance calibration steps are added to the image processing pipeline, then object detection accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The calibration data including lookup tables and correction factors are computed and stored in advance during an initialization phase. During actual image processing, the system simply applies these pre-computed values through straightforward table lookups and multiplications, avoiding the need to perform complex calibration calculations in real-time. This separates the computationally intensive calibration work from the time-sensitive image processing operations.
Solution Approach 2:
The patent creates simplified representations of the calibration data in the form of lookup tables that store pre-computed luminance correction values. Instead of performing complex radiometric calculations for every pixel during image processing, the system copies and applies these pre-stored correction values directly to the image data, significantly reducing computational overhead while maintaining measurement accuracy.
3Measurement precision
If absolute luminance values are generated for each pixel, then light measurement accuracy improves, but the complexity of the processing system increases
Solution Approach 1:
The system performs sensor calibration and generates lookup tables containing luminance correction factors during an initialization phase before actual operation. This preliminary action captures all the complex radiometric calibration computations in advance, allowing the operational system to simply apply stored correction values without performing complex calculations in real-time, thus achieving accurate absolute luminance measurements without maintaining high system complexity during operation.
Solution Approach 2:
The patent introduces lookup tables as an intermediary data structure that mediates between the raw sensor data and the final luminance values. These lookup tables store pre-computed correction factors that translate sensor responses into accurate luminance measurements. This intermediary approach simplifies the processing system by replacing complex real-time calculations with straightforward table lookups and value applications.
Data Source
AI summary
Apparatuses, systems, and techniques to process luminance and/or radiance values of one or more images from one or more cameras using one or more neural networks to perform a machine vision task. In at least one embodiment, one or more neural networks determine detection difficulty levels of objects within the one or more images and performs a machine vision task based on the determined detection difficulty levels of objects within images associated with that ask.