Abstract: Long-term autonomy is the foundation of intelligence for home service robots. However, achieving long-term autonomy in home environments still presents challenges such as difficulties in understanding the environment and poor adaptability to environmental changes. Therefore, an online method for constructing and maintaining information entities is proposed to provide real-time, reliable, and comprehensive object information for robots, enabling precise environmental perception. Firstly, a form of information representation suitable for various service tasks is constructed, named information entities, to integrally represent the physical and semantic information of objects in the environment. Based on this, an information entity updating strategy is proposed based on the spatial features, semantic information, and confidence of objects, to achieve high-precision automatic information extraction for object instances. Additionally, a selective active information maintenance method is proposed by considering the semantic richness and confidence of adjacent information entities, and combining with the optimal observation points, to achieve overall perception of the environment with lower computational cost. Finally, the effectiveness of the proposed method is verified through simulation and real experiments. The experimental results demonstrate that the proposed method can accurately and comprehensively describe environmental information and achieve efficient and complete information maintenance, meeting the requirements of long-term autonomy.