Environment exploration and recognition play an important role for service robots to operate in indoor environments that have a large scale and dynamic changes. This paper presents a new geometric model of the environment that is constructed with line segments and ellipses by estimating geometric properties and extracting geometric discontinuities. Two new types of indoor environment representations, namely exploration direction point and topological opening point (TOP), are defined using this model for autonomous exploration and spatial scene segmentation. The environment marks at TOPs are extracted from 2D laser data and a novel environment recognition algorithm is developed to accomplish environment recognition tasks autonomously. The experimental results show that the proposed approach can effectively explore and recognize a large-scale indoor environment in real time.