Semantic Masking and Visual Feature Matching for Robust Localization