3-GWD : A Textural and AI-Based Approach for Real-Time Detection of Weather Disturbances Applied to Autonomous Vehicle

We propose 3-GWD, a new and innovative solution for the particle atmospheric disturbers' detection that can be used to weight the data in a fusion process and adjust the sensor reliability. The proposed approach, based on 3D Gray-Level Co-occurrence Matrices (3D-GLCMs), is generic and could be used for a significant number of sensors (LiDAR, IR camera, Neuromorphic camera…). We demonstrate that this method leads to the identification of very relevant patterns in the symbolic space that shallow Artificial Neural Network (ANN) can classify with a great accuracy. We also discuss the importance of all-environment dataset by creating a reality- enriched simulated database with controlled disturbances and numerous environments. Using a single optic camera produces promising results with a high prediction score. We end our study with an attempt to compare our approach with previous works using the DAWN and the MCWCD datasets.