Beyond the Smudge: Simulating Opaque and Transparent Automotive Camera Lens Soiling

The performance and safety of Advanced Driver Assistance Systems (ADAS) are critically dependent on the integrity of camera-based vision systems. Environmental contamination of the camera lens, such as mud, dirt, and dust, can severely degrade image quality, leading to compromised system reliability. Real-world collection of contaminated image data is expensive and infrequent, necessitating robust simulation techniques to emulate lens soiling for ADAS model training. In this paper, we present a novel augmentation technique that simulates opaque and transparent lens contamination using Perlin noise. Our proposed method employs a lightweight Convolutional Neural Network (CNN) to generate contamination shapes, effectively modeling natural patterns observed in real-world datasets. The simulation of contamination scenarios enhances ADAS robustness under challenging conditions, contributing to improved safety. Experimental results demonstrate the effectiveness of the method in replicating realistic contamination effects, with two key heuristics quantifying information loss due to lens soiling. This work provides a foundation for future research in camera degradation simulation for autonomous systems.