Detekce semaforů v obrazech

Abstract

The purpose of this thesis is to create an application, which can detect traffic lights in images, more precisely find the traffic light in image and determine its state. To create this detector we will use convolutional neural networks, which is state-of-the-art method for detecting objects in images. In first part, we will describe basic principles of convolutional neural networks, layers, which build the network and how this method works as whole. Next, we will describe how we obtained our dataset, which is required for training of these networks and how we prepared that dataset. In last part, we will go through our implementation, we will compare different tested convolutional neural network architectures and finally, we will test our detector on real-life data.