1. Traffic lane detection
- Author
-
Šimunović, Bruno and Livada, Časlav
- Subjects
detekcija prometnih traka ,lane detection ,računali vid ,TECHNICAL SCIENCES. Computing. Program Engineering ,OpenCV ,HSL color space ,TEHNIČKE ZNANOSTI. Računarstvo. Programsko inženjerstvo ,computer vision ,HSL područje boja - Abstract
Problem ovoga rada je bio opis rada i kreiranje algoritma za detekciju prometnih traka. Za kreiranje algoritma se koriste OpenCV, Matplotlib i NumPy biblioteke. Algoritam je opisan u par koraka, od kojeg je najvažnija obrada slike. Slici je najbitnije prvo transformirati perspektivu, nakon toga slika se obrađuje pomoću HSL područja boja te se time reduciraju smetnje. Rezultat tih funkcija je binarna slika. Primjenom Canny algoritma detektiraju se rubovi slika kojom se završava obrada, nakon toga se pronalaze pravci koji predstavljaju prometne trake uz pomoć Houghove transformacije. Iz dobivenih pravaca stvara se jedan prosječni pravac kako bi se što bolje prometne trake prikazale. Na kraju, vraća se filtrirana slika u normalnu perspektivu i spaja se s ulaznom slikom. To je ukratko opis algoritma, a nakon toga se izvode testiranja. Takav algoritam je dovoljno brz za izvođenje u stvarnom vremenu, ali nije dovoljno efektivan za uporabu u stvarnom prometu. The problem of this paper was the description of the work and the creation of an algorithm for the detection of traffic lanes. OpenCV, Matplotlib and NumPy libraries are used for creation of the algorithm. It is described in a few steps, the most important of which is image processing. First, it is essential to transform perspective. After that, the image is processed using the HSL color space and the noise is reduced with that. This results in a binary image and the application of the Canny edge detection algorithm. To describe the image processing, routs representing traffic lanes are then found using the Hough transform. From the obtained routs, one average rout is created in order to show the best possible traffic lanes. Finally, we return the filtered image to normal perspective and merge it with the input image. This is a brief description of the algorithm, after which tests are performed. The algorithm is fast enough to run in real time, but not efficient enough for use in real traffic.
- Published
- 2021