Digital Library of the European Council for Modelling and Simulation

Title:	Elongated boundaries detector parameters optimisation based on generation of synthetic data from aerial imagery
Authors:	Ekaterina Panfilova, Anton Grigoryev, Vladimir Burmistrov
Published in:	(2022). ECMS 2022, 36^th Proceedings Edited by: Ibrahim A. Hameed, Agus Hasan, Saleh Abdel-Afou Alaliyat, European Council for Modelling and Simulation. DOI: http://doi.org/10.7148/2022 ISSN: 2522-2422 (ONLINE) ISSN: 2522-2414 (PRINT) ISSN: 2522-2430 (CD-ROM) ISBN: 978-3-937436-77-7 ISBN: 978-3-937436-76-0(CD) Communications of the ECMS , Volume 36, Issue 1, June 2022, Ålesund, Norway May 30th - June 3rd, 2022
Citation format:	Ekaterina Panfilova, Anton Grigoryev, Vladimir Burmistrov (2022). Elongated boundaries detector parameters optimisation based on generation of synthetic data from aerial imagery, ECMS 2022 Proceedings Edited By: Ibrahim A. Hameed, Agus Hasan, Saleh Abdel-Afou Alaliyat, European Council for Modeling and Simulation. doi:10.7148/2022-0167
DOI:	https://doi.org/10.7148/2022-0167
Abstract:	The detector of elongated boundaries in the image, such as road marking lines, rails, etc., is an important component of the visual system of a highly automated vehicle (HAV). It is used by HAV to solve self-localization problems or maintain the position inside the lane. Testing and optimisation of&nbsp; computer vision algorithms include preparing of the datasets that are usually labelled manually. Synthetic data of various kinds can reduce the complexity of&nbsp; algorithms development. In this paper we describe an approach to generation of synthetic data for testing elongated boundaries detectors that works with road images obtained from the cameras mounted on the HAV and converted to bird’s eye view. This approach is based on cutting of raster aerial imagery and corresponding vector markup of target objects in the aerial imagery in various ways.<br>There is a class of elongated boundaries detectors, the first stage of which is the background suppression on the image and thus highlighting of the target lines. For them, we propose a method for creating a dataset consisting of images with a suppressed background, specifically, images of white elongated lines on a black background. The lines’ shape will be similar to the target lines of the detector. With such a dataset you can tune the parameters, which affect stages of the algorithm, following the suppression of the background in the image.<br>In this paper we also consider elongated boundaries detector.&nbsp; Its parameters fix the geometric model of the target lines. Automatic optimisation of the quality of such a detector is possible using the Optuna toolkit, but it requires a large dataset. In this paper, we propose an approach to optimisation of the detector on a synthetic dataset. The effectiveness of this approach is confirmed by testing on real data.
Full text: