Search papers, labs, and topics across Lattice.
This paper introduces a cost-effective air pressure system designed to mitigate rain interference on automated vehicle cameras, addressing the limitations of existing solutions like hydrophobic coatings and expensive industrial systems. The system uses a directed airflow to clear water droplets from camera lenses, improving perception in rainy conditions. Experimental results demonstrate a significant increase in pedestrian detection accuracy, improving from 8.3% to 41.6% when using a Deep Learning model.
A simple air pressure system can boost pedestrian detection accuracy in rainy conditions by 5x, offering a low-cost alternative to expensive sensor upgrades for autonomous vehicles.
Recent advances in automated vehicles have focused on improving perception performance under adverse weather conditions; however, research on physical hardware solutions remains limited, despite their importance for perception critical applications such as vehicle platooning. Existing approaches, such as hydrophilic or hydrophobic lenses and sprays, provide only partial mitigation, while industrial protection systems imply high cost and they do not enable scalability for automotive deployment. To address these limitations, this paper presents a cost-effective hardware solution for rainy conditions, designed to be compatible with multiple cameras simultaneously. Beyond its technical contribution, the proposed solution supports sustainability goals in transportation systems. By enabling compatibility with existing camera-based sensing platforms, the system extends the operational reliability of automated vehicles without requiring additional high-cost sensors or hardware replacements. This approach reduces resource consumption, supports modular upgrades, and promotes more cost-efficient deployment of automated vehicle technologies, particularly in challenging weather conditions where system failures would otherwise lead to inefficiencies and increased emissions. The proposed system was able to increase pedestrian detection accuracy of a Deep Learning model from 8.3% to 41.6%.
