Raindrops on the Windshield: Performance Assessment of Camera-based Object Detection
Sinan Hasirlioglu, Fabio Reway, Tim Klingenberg, Andreas Riener, Werner Huber · 2019
Camera systems capture images from the surroundings and process these data for detecting and classifying objects. One particular category responsible for faults and failures is adverse weather conditions. In this case, it is known that rain, for example, degrades a sensor’s performance due to absorption and scattering on falling raindrops. In addition, object detection with in-vehicle cameras is affected by raindrops on the windshield since the outside view is disturbed. To counteract, windshield wipers are installed to improve visual information for human drivers but also benefit image sensors.This paper investigates the image quality and the correctness of object detection within one wiping action under rainy weather conditions. The results suggest that the image quality and the object detection performance decrease as the raindrops accumulate on the windshield. Further, it is shown that there is a trade-off between performance and time. Therefore, adaptive weighting data fusion should be implemented so that decisions are made based on the frames in which the sensors’ data is the most reliable.