Weather and meteorological optical range (MOR) perception is crucial for smooth and safe autonomous driving (AD). This article introduces two deep learning-based architectures, employing early and intermediate sensor fusion and multi-task strategies, designed for concurrent weather and MOR classification in AD. Extensive experiments employing the publicly available FogChamber dataset demonstrate that the proposed early fusion architecture, characterized by its lightweight design and simplicity, achieves an accuracy of 98.88% in weather classification and 89.77% in MOR classification, with a competitive memory allocation of 5.33 megabytes (MB) and an inference time of 2.50 milliseconds (ms). In contrast, the proposed intermediate fusion architecture prioritizes performance, achieving higher accuracies of 99.38% in weather classification and 91.88% in MOR classification. However, it requires a more substantial memory allocation of 54.06 MB and exhibits a longer inference time of 15.55 ms. Compared to other state-of-the-art architectures, the proposed methods present a competitive balance between accuracy performance, inference time, and memory allocation, which are crucial parameters for enabling autonomous driving.
Weather and Meteorological Optical Range Classification for Autonomous Driving
IEEE Transactions on Intelligent Vehicles ; 9 , 10 ; 6637-6647
01.10.2024
4679548 byte
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Weather Quiz - Meteorological marvels
Online Contents | 2006
LONG RANGE WEATHER CHARACTERIZATION FOR AUTONOMOUS VEHICLE SENSORS
Europäisches Patentamt | 2024
|Visual weather simulation using meteorological databases
AIAA | 1993
|