We propose a simple but robust method to recognize an unknown person described in natural language. In this case, a robot is given a verbal description about a person whom the robot is required to recognize. This task is challenging since humans and robots have significantly mismatched perceptual capabilities (e.g., recognizing the color of a coat). Without assuming that all linguistic descriptions and perceptual data are correct, we use a probabilistic model to ground the target person. In particular, the acceptability of color descriptions is modeled based on visual similarity and the confusion matrix of the color classifier which make the system more robust to illumination. Two groups of experiments were conducted. Our experimental results demonstrate that our system is robust to both perception and description errors.
» Read on@inproceedings{WWLrobocup17,
address = {Nagoya, Japan},
author = {Xiping Wang and Feng Wu and Dongcai Lu and Xiaoping Chen},
booktitle = {Proceedings of the Robot World Cup XXI Symposium (RoboCup)},
doi = {10.1007/978-3-030-00308-1_19},
month = {July},
pages = {228 - 240},
title = {A Robust Algorithm: Find an Unknown Person via Referring Grounding},
year = {2017}
}