Autonomous Driving Tone Mapping for HDR Bandwidth Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in processing high dynamic range (HDR) images due to limited data transfer bandwidth and processing power, leading to inefficiencies in online and offline image processing workflows, which can impact safety and computational resources.
Innovation Solution
A unified approach for tone mapping HDR images, combining global and local tone mapping to reduce bit depth, allowing efficient online processing by neural networks and offline processing for human consumption, while optimizing bit allocation based on pixel values and end-task requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If HDR images are processed with high bit depth to preserve image quality, then measurement precision is improved, but productivity deteriorates due to limited data transfer bandwidth and processing power
Solution Approach 1:
The tone mapping process is segmented into two distinct stages: global tone mapping that preserves HDR quality for neural network processing, and local tone mapping that converts to LDR for human viewing. This segmentation allows different parts of the processing pipeline to use appropriate bit depths for their specific purposes, resolving the contradiction between maintaining image quality and improving processing efficiency
Solution Approach 2:
The patent applies local tone mapping to specific regions of the image where human viewing is required, while maintaining global HDR quality in other regions for machine processing. This local quality adjustment allows the system to optimize for human perception in display regions without compromising the overall HDR information needed for autonomous driving decisions
2Productivity
If data duplication is avoided to improve productivity, then loss of information increases due to single-processing-path constraints
Solution Approach 1:
The globally tone-mapped HDR image serves multiple functions: it is used for online neural network processing and simultaneously serves as the source for offline local tone mapping to LDR. This multi-functionality eliminates the need for data duplication while ensuring that both processing paths have access to the complete HDR information needed for their respective tasks
3Productivity
If global tone mapping is applied to reduce bit depth for efficient processing, then productivity is improved, but manufacturing precision deteriorates due to bit allocation limitations
Solution Approach 1:
The patent changes the parameter of bit depth distribution by applying global tone mapping that strategically allocates bits across different luminance ranges. This parameter transformation allows the system to work with reduced bit depth while maintaining perceptually relevant precision, resolving the contradiction between computational efficiency and precision
Data Source
AI summary
A system is provided that includes an image sensor coupled to a vehicle, and control circuitry configured to perform operations including receiving, from the image sensor, an input stream comprising high dynamic range (HDR) image data associated with an environment of the vehicle, and processing the input stream at the vehicle by applying a global tone mapping, followed by offline image processing that can include applying a local tone mapping to the globally tone mapped images of the same input stream.