Detekce návěstidel a klasifikace jejich stavu scény s využitím knihovny OpenCV v jazyce Python

Tomáš VANĚČEK

Bachelor's thesis

Detection of traffic lights and classification of their state in the scene image using the OpenCv library in Python

Abstract:

Tato práce byla zaměřena na detekci návěstidel a klasifikace jejich stavu.Práce je vytvořená pomocí knihovny OpenCV v programovacím jazyce Python. Po získání záznamů z kamery modelové lokomotivy Smart Train jsme tyto záznamy použily pro natrénování kaskádového klasifikátoru. Stav návěstidla jsme zjišťovali nalezením barev světel v klasifikátorem detekované oblasti.

Abstract:

The topic of this work is to create an algorithm capable of detecting the model of traffic light and classifying its state. The project is written in Python and for its realization we will use the OpenCV library, which focuses on computer vision and machine learning issues. The images in which we detect the presence and state of traffic light are taken by the camera of a moving locomotive for the Smart Train model track. A cascade classifier is used for detection. For its training were used 5800 images that contained a traffic light and 11600 images that did not contain the traffic light. The resulting classifier achieves high accuracy when using images from train camera. Algorithms for finding limited ranges of color values of HSV model are used to detect its states. The lit colors of the traffic lights are detected correctly, however the colors of the unlit lights are also detected. The resulting application is designed to simplify work on solving similar topics.

Keywords

Počítačové vidění, Kaskádový klasifikátor, OpenCV, Python, HAAR, HSV, AdaBoost, Detekce objektů, Detekce návěstidel, Smart Train

Language used: Czech

Date on which the thesis was submitted / produced: 4. 5. 2023

Accessible from:: 31. 12. 2999

Thesis defence

• Supervisor: Ing. Pavel Mautner, Ph.D.

University of West Bohemia

Faculty of Applied Sciences

Bachelor programme / field:
Computer Science and Engineering / Information Technologies