Vision-Based Fail-Safe for Autonomous Driving Visibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving technologies face challenges with reduced visibility due to weather conditions and occlusions, leading to inaccurate object detection and classification, which can result in deficient performance and safety issues.
Innovation Solution
A vision-based machine learning model using convolutional neural networks processes image sensor data to determine visibility information and trigger corrective actions, such as adjusting driving behavior or switching to manual control, without relying on costly or error-prone sensors like radar.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If costly sensors like radar are used for autonomous driving, then object detection capability is improved, but system cost increases
Solution Approach 1:
The patent extracts and removes the radar sensor from the autonomous driving system, relying solely on image sensors (cameras) for object detection. This elimination of costly radar hardware while maintaining detection capability through vision-based machine learning models directly resolves the contradiction between detection precision and system cost.
Solution Approach 2:
The patent uses multiple image sensors to capture visual information from different perspectives, creating a comprehensive visual representation of the environment. This visual copying approach replaces the need for expensive radar sensors while achieving comparable or superior object detection accuracy through neural network processing.
2Measurement precision
If traditional sensors are used in adverse weather conditions, then object detection is performed, but detection accuracy deteriorates due to reduced visibility
Solution Approach 1:
The patent employs dynamic neural network models that adapt to changing visibility conditions in real-time. The system adjusts its detection parameters and processing strategies based on current weather conditions, allowing it to maintain high detection accuracy despite adverse conditions like fog, rain, or snow that reduce visibility.
Solution Approach 2:
The system incorporates feedback mechanisms where detection results and visibility conditions are continuously monitored and used to adjust processing parameters. This feedback loop enables the neural network to compensate for reduced visibility by adjusting detection thresholds and focusing computational resources on critical detection tasks, thereby maintaining accuracy in adverse weather.
3Quantity of substance
If image sensors are used for autonomous driving, then system cost is reduced, but reliability deteriorates in reduced visibility conditions
Solution Approach 1:
The patent merges multiple image sensors and combines their data streams through neural network processing to create a robust detection system. This fusion of multiple visual inputs compensates for the limitations of individual sensors in reduced visibility conditions, maintaining reliability while using only cost-effective image sensors instead of expensive radar.
Solution Approach 2:
The system dynamically changes processing parameters such as detection thresholds, confidence levels, and processing intensity based on visibility conditions. In adverse weather, the neural network adjusts these parameters to maintain reliable detection performance, ensuring that autonomous driving reliability is preserved even when using only image sensors rather than costly multi-sensor systems.
Data Source
AI summary
Systems and methods for fail-safe corrective actions based on vision information for autonomous driving. An example method is implemented by a processor system included in a vehicle, with the method comprising obtaining images from image sensors positioned about the vehicle. Visibility information is determined for at least a portion of the images. Adjustment of operation of an autonomous vehicle is caused based on the visibility information.