A Study on the Impact of Rain on Object Detection for Automotive Applications

Visible spectrum cameras have emerged as a key technology in Advanced Driving Assistance Systems (ADAS) and automated vehicles. An important question to be answered is how these sensors perform in challenging adverse weather conditions, such as rain. Although progress has been made in determining the impact of rain on computer vision performance, previous studies have generally focused on end-to-end object detection system performance and have not addressed the specific impact of rain in detail. Moreover, the lack of image datasets with detailed labeling acquired under rain conditions means that the impact of rain remains a relatively under-researched question. The purpose of this study is to examine the impact of rain in the propagation path on perception tasks, where other factors affecting performance are removed or controlled as far as possible. This study presents the results of controlled experimental testing designed to measure the impact of rain on automated vehicle perception performance. Object detection is performed on the captured data to determine the impact of rain on performance. Four object detection algorithms, a segmentation algorithm, and an optical character recognition algorithm are used as representative examples of typical algorithms used in ADAS. It is shown that the impact of rain varies between models, and at larger distances, rain has a greater impact. In the case of the OCR algorithm, rain is shown to have a larger impact at certain distances. The findings of this study are useful for ADAS design, as they provide more detailed insight into the impact of rain on ADAS and provide guidance on potential breaking points for algorithms typically used in this type of system.