The traditional environment maps built by robots include both metric ones and topological ones. These maps are navigation-oriented and not adequate for service robots to interact with or serve human users who normally rely on conceptual knowledge or semantic contents of the environment. Therefore, the construction of semantic maps becomes necessary for building an effective human-robot interface for service robots. This paper aims to build a 3D environment map with an RGB-D sensor and extract semantic information from the RGB images to help blind people navigate at home. A novel approach is presented to diagnose and eliminate errors during semantic extraction.