A Self-Organizing Neural Scheme for Door Detection in Different Environments

TL;DR

This paper introduces a novel, feature-based self-organizing neural network approach for robust door detection across diverse indoor environments, outperforming existing methods in accuracy and generalizability.

Contribution

The paper proposes a new door detection method using a Kohonen SOM with generic features, improving robustness and performance over prior algorithms.

Findings

Achieves over 95% detection accuracy across varied conditions

Demonstrates robustness to lighting, occlusions, and viewpoint changes

Validates effectiveness on a large, diverse dataset

Abstract

Doors are important landmarks for indoor mobile robot navigation and also assist blind people to independently access unfamiliar buildings. Most existing algorithms of door detection are limited to work for familiar environments because of restricted assumptions about color, texture and shape. In this paper we propose a novel approach which employs feature based classification and uses the Kohonen Self-Organizing Map (SOM) for the purpose of door detection. Generic and stable features are used for the training of SOM that increase the performance significantly: concavity, bottom-edge intensity profile and door edges. To validate the robustness and generalizability of our method, we collected a large dataset of real world door images from a variety of environments and different lighting conditions. The algorithm achieves more than 95% detection which demonstrates that our door detection method is generic and robust with variations of color, texture, occlusions, lighting condition, scales, and viewpoints.