Please use this identifier to cite or link to this item:
http://hdl.handle.net/10609/110806
Title: | Segmentació de mans en imatges de profunditat |
Author: | Galmés Rubert, Bernat |
Director: | Ventura, Carles |
Tutor: | Moyà Alcover, Gabriel |
Abstract: | This document treats a hands detection method using depth information. It is achieved classifying the pixels of the image according to its probability to belong at a hand. A set of simple features are computed for each pixel, and its prediction is obtained using a Random Forest classifier. This way, real-time predictions are achieved. The aim of this work is to present the operation details of the method and analyse its behaviour. Besides, as well as problems are detected, solutions will be suggested to solve them, which will be applied in order to improve the model. Two problems detected are incorrect behaviour predicting hands on non controlled environments and the confusion of hands and face samples. One solution has been the creation of a new dataset composed by images took of a non controlled environment. Another solution has been the append of a major percentage of face samples in the training set. With the dataset change, a notable results improvement is achieved in the target environment. However, only a little variation is achieved adding the faces samples. Anyway, the behaviour of the classifier is excellent, all false predictions are caused by the classifier architecture, samples which features take the same value in some positives and false cases. |
Keywords: | real time depth information randomized decision forests hands segmentation human-computer interaction |
Document type: | info:eu-repo/semantics/masterThesis |
Issue Date: | 14-Mar-2020 |
Publication license: | http://creativecommons.org/licenses/by-nc-sa/3.0/es/ |
Appears in Collections: | Bachelor thesis, research projects, etc. |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
berngalmesTFM0120memòria.pdf | Memòria del TFM | 7,44 MB | Adobe PDF | View/Open |
Share:
This item is licensed under a Creative Commons License