Approaches to door identification for robot navigation

E. Jauregi, E. Lazkano, and B. Sierra. Abstract:
Even though great technical progress has been made in the area of mobile robotics, some fundamental control problems, such as autonomous navigation, remain unresolved. Many animals have shown that they are very good at navigating but autonomous navigation is still a complicated task for engineered robots. Therefore, research efforts have aimed at incorporating biologically inspired strategies into robot navigation models. Among these, Local navigation strategies allow the agent to choose actions based only on its current sensory input, while \emph{way finding strategies} are responsible for driving the agent to goals out of the agent's perceptual range and require the recognition of different places and correlations among them. Acquiring the appropriate set of landmarks remains a challenge. Many navigation tasks can be fulfilled by point to point navigation, door identification and door crossing. Indoor semi-structured environments are full of corridors that connect different offices and laboratories where doors give access to many of those locations that are defined as goals for the robot. Hence, endowing the robot with the door identification ability would undoubtedly increase the navigating capabilities of the robot. Doors in indoor environments are not necessarily uniform; door panels can be of different texture and colour and handles can vary in shape. Specific methods can be designed for each type of door. However, feature-based methods are more easily applicable to different object recognition tasks they are computationally expensive and therefore less attractive to be applied for real-time problems such as for mobile robot navigation. This chapter compares several approaches developed for door identification based on handle recognition, from specific methods based on colour segmentation techniques to more general ones that focus on feature extraction methods, like SIFT and (U)SURF. A new two-step multiclassifier that combines region detection and feature extraction is also presented. The objective of the approach is to extract the most relevant part of the image, the subimage that with high probability should contain the most interesting region of the image, and limit the application of the feature extraction techniques to that region. The developed algorithm improves the relevance of the extracted features, it reduces the superfluous keypoints to be compared at the same time that increases the efficiency by improving accuracy and reducing the computational time. The developed approaches are first evaluated off-line and tested afterwards in a real robot/environment system using a a Peoplebot robot from Mobilerobots. The door recognition module is integrated in a behaviour-based control architecture that allows the robot to show a door-knocking behaviour.


bibTeX entry:
@inbook{ jauregi10approaches,
chapter = "Approaches to door identification for robot navigation",
author = "E. Jauregi and E. Lazkano and B. Sierra",
year = "2010",
title = "Mobile Robots Navigation",
chapter ="12",
pages = "241--261"
}