Intelligent Transport Systems, From Research and Development to the Market Uptake. 4th EAI International Conference, INTSYS 2020, Virtual Event, December 3, 2020, Proceedings

Research Article

Performance Evaluation of Object Detection Algorithms Under Adverse Weather Conditions

Download
729 downloads
  • @INPROCEEDINGS{10.1007/978-3-030-71454-3_13,
        author={Thomas Rothmeier and Werner Huber},
        title={Performance Evaluation of Object Detection Algorithms Under Adverse Weather Conditions},
        proceedings={Intelligent Transport Systems, From Research and Development to the Market Uptake. 4th EAI International Conference, INTSYS 2020, Virtual Event, December 3, 2020, Proceedings},
        proceedings_a={INTSYS},
        year={2021},
        month={7},
        keywords={Object detection Adverse weather Autonomous driving},
        doi={10.1007/978-3-030-71454-3_13}
    }
    
  • Thomas Rothmeier
    Werner Huber
    Year: 2021
    Performance Evaluation of Object Detection Algorithms Under Adverse Weather Conditions
    INTSYS
    Springer
    DOI: 10.1007/978-3-030-71454-3_13
Thomas Rothmeier1, Werner Huber1
  • 1: Ingolstadt University of Applied Sciences

Abstract

Camera systems capture images from the surrounding environment and process these datastreams to detect and classify objects. However, these systems are prone to errors, often caused by adverse weather conditions such as fog. It is well known that fog has a negative effect on the camera’s view and thus degrades sensor performance. This is caused by microscopic water droplets in the air, that scatter light, reduce contrast and blur the image. Object detection algorithms show severely worse performance and high uncertainty when exposed to fog. However, they need to work safe and reliable in all weather conditions to enable full autonomous driving in the future. This work focuses on the evaluation of several state-of-the-art object detectors in normal and foggy environmental conditions. It is shown that the detection performance deteriorates considerably when exposed to fog. Further, the results suggest that some algorithms are more robust towards fog than others.