Object Detection and Recognition Using Template Matching with SIFT Features Assisted by Invisible Floor Marks

For simultaneously localizing and mapping (SLAM) an indoor mobile robot, a method to process a monocular image of entire environmental view is proposed. To ensure that an object can be searched for, invisible floor marks are proposed for modifying the environment and which are useful in narrowing the search area in an image. Specifically our approach involves: 1) narrowing the searched area using invisible floor marks, 2) extracting features based on scale-invariant feature transform (SIFT), 3) using template matching with SIFT features assisted by partial templates and the spatial relationship to the floor, and 4) verifying object recognition with an AdaBoost classifier using Haar-like features based on object shape information. A robot is localized relative to the floor using the floor marks, then, objects in a clattered image are extracted and recognized, and 3D solid models of them are mapped on the floor to build a highly structured 3D map. Recognition was over 80% successful, including tables and chairs and taking several tens of seconds per 640 × 480 pixel image.