A system for improving the accuracy of road situation recognition in an unmanned ground vehicle
Students Name: Skorobreshchuk Serhii Yuriiovych
Qualification Level: magister
Speciality: System Programming
Institute: Institute of Computer Technologies, Automation and Metrology
Mode of Study: full
Academic Year: 2020-2021 н.р.
Language of Defence: ukrainian
Abstract: Skorobreshchuk S.Y., Golembo V.A. (supervisor). A system for improving the accuracy of road situation recognition in an unmanned ground vehicle. Master’s thesis. - National University "Lviv Polytechnic", Lviv, 2020. Extended abstract Improving the orientation of unmanned vehicles will free people from stressful driving, optimize traffic and reduce future accidents. Many discoveries and developments have been made in the field of self driving cars. [1] However, one of the limiting factors is the presence of ordinary drivers on public roads. If you look closely at their driving style, it is immediately apparent that anyone deviates from the rules to one degree or another. Therefore, at the moment it is impossible to create an autopilot that will be able to withstand the unpredictable behavior of other people on the road. And that is why the sooner a model that is able to move around the city with high accuracy is implemented - the sooner all transport will be autonomous and free humanity from most of the issues of large cities that exist at the moment [2]. Object of research: means of assessing the road situation in unmanned ground vehicles. Subject of research: increasing the accuracy of recognizing the road situation in unmanned ground vehicles. The aim of the study is to develop and test the impact of image processing algorithms for the selection of significant objects on the accuracy of recognition of the road situation in the unmanned ground vehicle. 5 Research results: - built a test model of an unmanned ground vehicle, a training track, wrote software that allows you to easily collect input data for training, process them, build a neural network model and finally control the developed device [3]. The whole process of debugging, installation of all software and basic modules of the software part is described; - the algorithm of allocation of lines of a road marking as the basic elements of an estimation of a road situation chosen for carrying out the given research is developed [4]; - it was found that the recognition is slightly impaired if we apply the developed processing algorithms in conditions close to training [5]. This result is explained by the fact that previous processing algorithms in addition to noise can cut off some important information and need to be improved. However, in all other cases, pre-processing of images to highlight significant elements has reduced road recognition error by an average of 6%, which is a fairly good improvement for neural networks used as a basis for research. Keywords - unmanned ground vehicle, image processing, neural networks. List of used literature sources. 1. Holembo V.A., Hrebenyak A.V. Navihaciya v kolektyvi avtonomnyh aparativ // Visnyk Nacionalnoho universytetu “Lvivska politehnika” “Kompyuterni systemy ta merezhi”. – 2010. – № 688. – pp. 77–83. 2. Melnyk A. O. Naukovi napryamy stvorennya bahatorivnevoyi platformy kiberfizychnyh system // Kiberfizychni systemy dosyahnennya ta vyklyky : materialy tretoho naukovoho seminaru, 13–14 June, 2017 , Lviv. – 2017. – pp. 4–9. 6 3. Shtuchni nejronni merezhi : obchyslennya / M.A. Novotarskyj, B.B. Nesterenko // Praci Instytutu matematyky NAN Ukrayiny. – T50. – Kyyiv: In-t matematyky NAN Ukrayiny, 2004. – 408 p. 4. David A. Forsyth, Jean Ponce. Computer Vision: A Modern Approach, 8(1-2) – 2012. – pp 50-440. 5. Deng L. Deep learning: Methods and applications / Deng L. and Yu D. // Foundations and Trends in Signal Processing, 7(3–4) – 2014. – pp. 197–387.