Robust door detection in unfamiliar environments by combining edge and corner features

Camera-based indoor navigation and wayfinding can assist blind people to independently access unfamiliar buildings. In indoor environments, doors are significant landmarks and door detection plays important roles for navigation and wayfinding. Most existing algorithms of door detection are limited to work for familiar environments with restricted features without taking account of the diversity and variance of doors in different environments. In this paper, we present an image-based door detection algorithm that utilizes the general and stable features of doors - edges and corners. Furthermore, we develop a general geometric model to characterize the door shape by combining edge and corner features without a training process. To validate the robustness and generalizability of our method, we collected a large dataset of door images from a variety of environments. The proposed algorithm achieves 91.7% true positive rate with a low false positive rate of 2.9%. The results demonstrate that our door detection method is generic and robust to different environments with variations of color, texture, occlusions, illumination, scales, and viewpoints.