Analysis of Object Detection Under Different Weather Conditions in Simulated and Real Environment

Pragati Jaiswal, Axel Vierling, Karsten Berns

Abstract: In modern robotic applications, like autonomous driving in an outdoor environment, a very important task is robust object detection. The object detection model tends to perform poorly under weather conditions including rain. Therefore, in order to have good performance, the data required for the training of these models has to be from a wide variety of conditions, like different lighting, weather, locations, etc. Any model not trained on a different variety of data is more likely to overfit on the dataset it is trained on and hence have poor performance over other situations. To avoid this situation often simulation data is used. The main goal of this paper is to analyse the performance of the object detection model on the various simulated and real weather conditions in order to achieve good performance under the real rain condition with a model trained only on simulated rain data.

External IDs:doi:10.1007/978-3-031-04870-8_52