Javascript must be enabled for the correct page display

Multimodal detection and recognition of persons with a static robot

Rombouts, J. (2009) Multimodal detection and recognition of persons with a static robot. Master's Thesis / Essay, Artificial Intelligence.

[img]
Preview
Text
AI-MAI-2009-J.O.Rombouts.pdf - Published Version

Download (8MB) | Preview

Abstract

In the context of the SnackBot project at Carnegie Mellon's Robotics Institute the feasibility of using simple soft-biometric features to recognize persons at various distances and orientations with respect to a static robot in an indoor office setting was investigated. The features that were investigated were person height and several color features (1D and 2D chromaticity histograms in HSV, nRGB, YCrCb and CIE-Lab color-spaces with several bin-sizes) extracted from torso or head+torso. Two different segmentation techniques that fused information from a planar-laser scanner and a stereo camera were implemented: one using static background modeling (BGM) and the other using the distance measurements from the camera (DS). A data set containing thirty subjects in several poses and distances w.r.t the robot in two different environments was created for testing. The recognition accuracy was evaluated with three different supervised machine learning techniques: kNN, SVM and Random Forests. Experiments with both BGM and DS systems show that both systems attained high performances (95%+) for close locations (±1.3 meters from robot) and that performance decreased with distance and changing illumination. Overall the BGM yielded the best results. When trained on only a single feature-vector per person (e.g. only frontal pose) the recognition of other poses was found to be quite robust.

Item Type: Thesis (Master's Thesis / Essay)
Degree programme: Artificial Intelligence
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 15 Feb 2018 07:28
Last Modified: 15 Feb 2018 07:28
URI: http://fse.studenttheses.ub.rug.nl/id/eprint/8503

Actions (login required)

View Item View Item