Detekce a rozpoznání dopravních značek v obrazech pomocí hlubokých neuronových sítí

Abstract

The aim of this diploma thesis was to detect and recognize traffic signs using neural networks. Using a camera located in a vehicle, it is possible to obtain images from which are traffic signs detected and recognized. This thesis is divided into a theoretical part and the actual implementation of a program for detection and recognition of traffic signs.

The theoretical part describes convolutional neural networks and detection methods that could be applied in this problem. In the actual implementation part, the functionality, accuracy and speed of detection and recognition of traffic signs were experimentally verified. This section describes two methods of detecting and recognizing traffic signs. The first method is using a single network for detection and recognition. The second method is detection of traffic signs using one network and second convolutional neural network for recognition of the found signs. The final chapter contains problems that can occur by putting this application into full operation and describes the pros and cons of this system.