Deep Learning Visibility Estimation for Autonomous Sensor Usability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for autonomous driving face challenges in accurately determining visibility distances due to compromised sensor data from weather and other conditions, leading to ineffective real-time deployment and inaccurate classification of sensor usability.
Innovation Solution
Utilizing deep neural networks trained with real-world, augmented, and synthetic data to estimate visibility distances, enabling accurate determination of sensor data usability for various autonomous tasks by adjusting reliance on sensor data based on computed visibility distances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional feature-based computer vision techniques are used to detect visibility issues, then individual visual features can be identified, but the system complexity increases and real-time deployment becomes ineffective
Solution Approach 1:
The patent merges multiple feature detection tasks into a single deep learning model that simultaneously performs visibility condition classification and visibility distance estimation. This unified approach integrates what would otherwise require separate analysis of edge features, color analysis, and other visual features into one cohesive system, reducing overall complexity while maintaining detection accuracy.
Solution Approach 2:
The patent replaces conventional mechanical computer vision techniques (feature extraction, edge detection, manual classification) with a deep learning-based perceptual system. This substitution allows the system to process visual information more efficiently, achieving real-time performance by leveraging the pattern recognition capabilities of neural networks rather than sequential feature analysis.
2Loss of information
If conventional systems classify reduced sensor visibility causes, then visibility conditions can be identified, but accurate determination of sensor data usability is not provided
Solution Approach 1:
The patent adds a new dimension to visibility analysis by estimating actual visibility distance in addition to classifying visibility conditions. Instead of merely categorizing weather conditions (rain, fog, snow), the system provides quantitative distance information that enables more nuanced decisions about sensor data usability, allowing the system to distinguish between light drizzle (usable) and dense fog (unusable).
Solution Approach 2:
The patent changes the output parameter from binary classification (blind vs. not blind) to continuous visibility distance estimation. This parameter transformation allows for more granular assessment of sensor data quality, enabling the system to adjust reliance on sensor data based on the estimated distance rather than making all-or-nothing decisions.
3Ease of operation
If conventional systems treat each type of compromised sensor visibility equally, then classification is simplified, but sensor data usability determination becomes inaccurate
Solution Approach 1:
The patent applies local quality assessment by evaluating visibility conditions independently for different regions of the sensor field of view. Instead of treating the entire image as uniformly compromised, the system can determine that certain regions (e.g., within 100 meters) remain usable even when other regions are obscured, allowing for more reliable and nuanced sensor data utilization.
Data Source
AI summary
In various examples, systems and methods are disclosed that use one or more machine learning models (MLMs)—such as deep neural networks (DNNs)—to compute outputs indicative of an estimated visibility distance corresponding to sensor data generated using one or more sensors of an autonomous or semi-autonomous machine. Once the visibility distance is computed using the one or more MLMs, a determination of the usability of the sensor data for one or more downstream tasks of the machine may be evaluated. As such, where an estimated visibility distance is low, the corresponding sensor data may be relied upon for less tasks than when the visibility distance is high.