Augmented (or mixed) reality (AR) and deep learning are among the most popular fields of technology these days. AR can revolutionize entertainment, education or even industry. Real-time footage of the real world is a fundamental part of these applications, thus, the scientific field of image processing is able to extend AR, making it more interactive and realistic. Deep learning has been dominating the field of image processing since the appearance of convolutional neural networks (CNN) which makes it suitable to apply in environments such as augmented reality. During my work, I studied these technologies as an intern of INDE R&D company. I attempted to implement a feature that is able to segment human bodies on AR images. The methods I studied are based on CNN-s, or a traditional model like GrabCut, a Gaussian Mixture Model based method. Beside the segmentation task, I integrated an algorithm, which estimates the depth distances of the people on the input image as well. Using these features combined, occlusion handling becomes available which increases the dimensionality of the AR application.