Kamerově ovládané autonomní robotické vozítko

Abstract

Tato bakalářsk ́a pr ́ace se zab ́yv ́a tvorbou a programov ́an ́ım algoritmu pro de-tekci j ́ızdn ́ıho pruhu za pomoci z ́aznamu z kamery, kter ́y m ̊uˇze b ́yt pouˇzit pro ˇr ́ızen ́ı robotick ́eho voz ́ıtka, osazen ́eho kamerou a jednoˇcipov ́ym poˇc ́ıtaˇcem. C ́ılem pr ́ace je otestov ́an ́ı vyvinut ́eho algoritmu na voz ́ıtku osazen ́em poˇc ́ıtaˇcem Nvidia Jetson Nano. V souladu s c ́ılem pr ́ace bude do textu pr ̊ubˇeˇznˇe zaˇrazov ́an pˇrehled a vysvˇetlen ́ı metod pouˇz ́ıvan ́ych pro navigaci autonomn ́ıho voz ́ıtka, jeˇz vyuˇz ́ıvaj ́ı zpracovan ́ı obrazu z kamery. Pro naprogramov ́an ́ı algoritmu byl zvolen jazyk Python, jelikoˇz disponuje podporou kni-hovny OpenCV, jenˇz poskytuje mnoˇzstv ́ı pˇreddefinovan ́ych funkc ́ı a metod pro zpracov ́an ́ı obrazu. Pˇri pˇredzpracov ́an ́ı obrazu bylo vyuˇzito perspektivn ́ıch transformac ́ı, kter ́e jsou v t ́eto problematice ˇcasto pouˇz ́ıv ́any pro usnadnˇen ́ı detekce. Byla vyuˇzita zn ́am ́a bi-narizaˇcn ́ı metoda Otsu, jej ́ımˇz v ́ystupem je bin ́arn ́ı obraz, ze kter ́eho detekuje hranice j ́ızdn ́ıho pruhu metoda zvan ́a Sliding window. V ́ysledkem je k ́od pro algoritmus, kter ́y m ̊uˇze b ́yt v budoucnu pouˇzit jako z ́aklad pro v ́yvoj algoritmu na ovl ́ad ́an ́ı ˇr ́ıd ́ıc ́ı n ́apravy voz ́ıtka. Vzhledem k okolnostem pandemie byla oblast testov ́an ́ı algoritmu z ́uˇzena pouze na testovac ́ı videa z autokamery. Z ́aroveˇn byla ovˇeˇrena a vyhodnocena pouˇzitelnost nˇekter ́ych koncept ̊u pro detekci j ́ızdn ́ıho pruhu.

This thesis deals with development and programming of a lane detection algorithm which, based on camera input, can be used for steering a robotic vehicle equipped with single–board computer and camera. The goal is to test the algorithm developed, using a vehicle equipped with an Nvidia Jetson Nano computer. Throughout the thesis, an overview of camera–based navigation methods will be included to correspond with the goal. Python is the programming language of choice, thanks to the support of OpenCV library, which contains many predefined functions and methods used in image processing. The algorithm will utilize perspective transformations, which are often used in lane detection algorithms. The Otsu method was the binarization method of choice, which outputs a binary image. This binary image is then processed by the lane detection method called the sliding window. The output of the thesis is a source code for the algorithm, which can be used in future development of an algorithm for vehicle steering. Given the circumstances during the pandemic, the testing was reduced to sample test videos from a dash cam. Additionally, the usability of some lane detection concepts was proved and evaluated.