Robust Vehicle Detection Across Diverse Weather Conditions Using Advanced Ensemble Learning Techniques

• Ratnapal Kumarswami Mane,
• Poonam Sharma,
• Ankush D Sawarkar,

Vol. 19, No. 12, December 31, 2025
10.3837/tiis.2025.12.017, Download Paper (Free):

• autonomous vehicles
• Ensemble learning
• fuzzy vector-rank partition matching filter
• fast curvelet-wrapped autofocus transform
• Viola-Jones' steerable-pyramid detection

Abstract

Autonomous vehicle perception systems struggle with detecting other vehicles under adverse weather conditions, demanding robust solutions capable of handling environmental variability. This research presents a novel ensemble learning model that uniquely integrates advanced preprocessing, Recurrent Neural Network (RNN) -based contextual aggregation, and Generative Adversarial Network (GAN) -driven feature refinement to achieve robust vehicle detection across diverse weather conditions. A fuzzy vector-rank partition matching filter enhances vehicle images from the DAWN2020 dataset to eliminate background noise. A fast curvelet-wrapped autofocus transform sharpens the images while enhancing relevant features of the vehicle. For feature extraction, Viola-Jones' steerable-pyramid detection captures intricate details of the vehicles. The processed images are then fed into an ensemble detection model combining outputs from SqueezeNet Inception-v3, EfficientNet-B7, RNN, and GAN, ensuring the model adapts effectively to the high variance from different weather conditions. The model significantly outperforms contemporary methods with a reduced root mean square error and a rapid inference time of 15.94 ms. Additionally, cross-dataset validation is performed using the BDD100K dataset. These results highlight its capability to detect vehicles precisely under complex weather conditions.

Statistics

Cite this article