Image Generation for HDR Sensor Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the development of automated driving systems, the use of high dynamic range (HDR) image sensors in simulation environments fails to accurately reproduce noise, leading to incorrect evaluation of image recognition algorithms due to Signal-to-Noise Ratio (SNR) drops, which deteriorate recognition performance.
Innovation Solution
An image generation apparatus and method that generates simulation images with varying accumulation times based on light applied to an image sensor, followed by HDR synthesis, replicating the SNR drop experienced in actual HDR image sensors, allowing for accurate evaluation of image recognition algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If HDR image sensors are used in simulation environments, then the dynamic range is improved, but the noise reproduction accuracy deteriorates due to SNR drops
Solution Approach 1:
The patent changes the parameter of accumulation time for multiple simulation images, generating images with different exposure durations. By varying this parameter and subsequently performing HDR synthesis, the system reproduces the SNR drop characteristics of actual HDR image sensors, thereby improving noise reproduction accuracy while maintaining enhanced dynamic range.
2Measurement precision
If multiple simulation images with different accumulation times are generated and HDR synthesis is performed, then the noise reproduction accuracy is improved, but the processing complexity increases
Solution Approach 1:
The patent segments the image generation process into multiple steps: generating multiple simulation images with different accumulation times, performing HDR synthesis on these segmented images, and evaluating recognition performance. This segmentation allows systematic reproduction of SNR drop characteristics while managing processing complexity through structured workflow.
3Ease of operation
If conventional simulation methods are used, then the processing simplicity is maintained, but the recognition performance evaluation accuracy deteriorates
Solution Approach 1:
The patent creates copies of simulation images with varied accumulation times to replicate the behavior of actual HDR image sensors. By copying and processing multiple versions of the same scene with different exposure parameters, the system accurately evaluates recognition performance under realistic noise conditions without requiring physical hardware changes.
Data Source
AI summary
The present disclosure relates to an image generation apparatus, an image generation method, and a program that make it possible to correctly evaluate recognition performance of an image recognition algorithm.An image generation unit generates a plurality of simulation images in which a plurality of images having different accumulation times is reproduced on the basis of a physical quantity corresponding to light applied to an image sensor, and an HDR synthesis unit performs HDR synthesis on a plurality of the simulation images. The present technology can be applied to an image sensor model.