Study on the application of different image preprocessing algorithms in image segmentation using deep learning techniques

Abstract

Semantic segmentation is one of the most important fields of computer vision due to its applicability. In this case, semantic segmentation will be applied to images taken from vehicles knowing the importance of autonomous driving for our future. This project aims to achieve precise and fast results in the field of semantic segmentation with the complication of using less powerful and more affordable hardware than the one used nowadays. Deep learning techniques will be used to solve this problem. More concretely, a specific model of convolutional neural network will be trained and in charge of making the predictions, a U-Net. Different parameters of the U-Net will be changed to study how they affect the results. Furthermore, various image sizes, color spaces or reduction methods will be applied to study their impact on the speed and accuracy of the U-Net predictions. Finally, all those results will be compared in order to make a final decision in which is the best combination and which fields impact the most and how.