Surround-View Fisheye BEV-Perception for Valet Parking: Dataset, Baseline and Distortion-Insensitive Multi-Task Framework

Zizhang Wu, Yuanzhu Gan, Xianzhi Li, Yunzhe Wu, Xiaoquan Wang, Tianhao Xu, Fan Wang

Abstract: Surround-view fisheye perception under valet parking scenes is fundamental and crucial in autonomous driving. Environmental conditions in parking lots perform differently from the common public datasets, such as the imperfect light and opacity, which substantially impacts on perception performance. Most existing networks based on public datasets may generalize suboptimal results on these valet parking scenes, also affected by the fisheye distortion. In this article, we introduce a new large-scale fisheye dataset called <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F</b> isheye <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">P</b> arking <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">D</b> ataset ( <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FPD</b> ) to promote the research in dealing with diverse real-world surround-view parking cases. Notably, our compiled <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FPD</b> exhibits excellent characteristics for different surround-view perception tasks. In addition, we also propose our real-time distortion-insensitive multi-task framework <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F</b> isheye <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">P</b> erception <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Net</b> work ( <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FPNet</b> ), which improves the surround-view fisheye BEV perception by enhancing the fisheye distortion operation and multi-task lightweight designs. Extensive experiments validate the effectiveness of our approach and the dataset's exceptional generalizability.