Multi-object recognition and tracking with feature points matching and spatial layout consistency

In this paper, we propose a simultaneous textured object recognition and localization approach designed to real-time mobile devices applications. Local interest points are extracted and classified with the help of a strong matching algorithm. The matching procedure is based on a strict verification of the spatial layout consistency of the considered interest points. The proposed approach is well suited for real-time multi objects tracking and yields precision/recall rates superior to 90%.