A Method for Identification of Objects in Cluttered Scenes Using Local Operators and Range Gating

Identification of known objects is the central issue in many applications of machine vision. We discuss a detection method for objects (both 2D and 3D) randomly located in complex cluttered scenes where partial occlusions are possible. Moreover, the images can be captured in a turbid medium that further degrades visibility. The images are acquired using a gated imaging system so that the medium backscattering noise is minimized and only the scenes components within a predefined distance from the camera are captured in the images, while the rest of the scene remains invisible. The database of known objects is built in reference scale using local interest points (shape features) extracted from template images presenting an object of interest from various viewpoints. By matching interest point detected (using relative scale) in gated images to the local features from the database, known objects can be identified even if they are only partially visible. The paper briefly discusses the proposed methodology and explains how the complexity of vision-based navigation algorithms could be dramatically reduced (with the corresponding improvement of the robustness) by adopting the proposed approach. Key-Words: object identification, gated imaging, moment invariants, local shape, relative scale