Simulation System for Autonomous Driving Weather Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic driving systems face challenges in simulating real-world driving conditions, particularly severe weather conditions, which limits the collection of test data and recognition accuracy due to the impracticality of reproducing such conditions and the inability of camera images to provide three-dimensional profiles of objects.
Innovation Solution
A simulation system that generates realistic images using computer graphics (CG) to replicate severe weather conditions and combines multiple sensors like LiDAR and millimeter wave sensors to improve recognition rates, allowing for synchronization control and deep learning recognition in a virtual environment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera images are used for object recognition in automatic driving systems, then the system can detect objects such as vehicles, walkers and traffic signals, but the recognition rate is substantially changed by external factors such as weather conditions and time zone, leading to increased misdetection and undetection
Solution Approach 1:
The patent creates virtual copies of real-world driving scenes including severe weather conditions through computer graphics technology. By generating synthetic training data that replicates various weather conditions (rain, snow, fog, night, backlight), the system can train recognition algorithms without requiring actual physical presence in those conditions, thereby resolving the contradiction between maintaining recognition accuracy and avoiding weather-related detection errors.
2Measurement precision
If the number of samples for learning is increased to improve recognition rate with deep learning technique, then the recognition rate improves, but it has a limit to extract learning samples during actually driving on a road and it is not realistic to carry out driving test and sample collection after meeting severe weather conditions
Solution Approach 1:
The patent generates virtual training samples through computer graphics rendering of driving scenes under various weather conditions, eliminating the need for physical driving tests in severe weather. This copying approach allows unlimited generation of diverse training data without the constraints of real-world sample collection feasibility.
Solution Approach 2:
The patent performs preliminary generation of training samples covering all possible weather conditions before actual model training. By pre-generating comprehensive datasets including rare severe weather scenarios that would be impossible to capture during normal driving, the system eliminates the need for subsequent real-world sample collection under difficult conditions.
3Loss of information
If camera images are used for outside information detection, then the system can recognize objects, but three-dimensional profiles cannot be obtained which is insufficient for fully automatic driving
Solution Approach 1:
The patent enhances two-dimensional camera images with three-dimensional information through depth estimation algorithms and multi-view geometry techniques. By synthesizing depth maps and 3D spatial relationships from 2D image data, the system recovers three-dimensional profile information without requiring additional LiDAR or radar sensors, thus resolving the contradiction between information completeness and system complexity.
Data Source
AI summary
This vehicle synchronization simulation device and means is provided with: a means for calculating positional information of the own vehicle; a means for transmitting the own vehicle positional information to a server means; a means for converting the own vehicle positional information into a specific data format and transmitting the same; a means for transferring the data via a network or a transmission bus inside of a specific device; a means for receiving the data and generating an image; a means for recognizing and detecting a specific object from the generated image; and a means for changing/correcting positional information of the own vehicle using the information resulting from recognition. The server means is provided with a means that synchronously controls three means, a own vehicle position calculating means, a transmitting/receiving means, and an image generating/recognizing means.