A Pattern Matching Scheme for Accurate Medicine Localization in Automatic Pharmacy Inventory

A vision based automatic pharmacy inventory is an unstructured environment in the sense that a random face of medicine box confronts the camera according to how it is put on shelf. This poses a challenge to traditional image content based object detection algorithms. In this paper we propose a shapebased medicine box localization framework. Instead of using feature point or texture based method which depends on image content, the proposed method adopts a pattern matching scheme, which takes advantage of the object’s shape to build a synthetic pattern, for object localization. The superiority of our method lies in three-folds: 1) the proposed method is robust to frontal image content variation and does not depend on single image to make identification and localization; 2) the pattern matching scheme is robust to object perspective transform; 3) the method is insensitive to noises introduced by clutter environment, such as unbalanced illumination and reflection. To evaluate the proposed method, we test it in the real application environment. Large scale experiment shows the accuracy and robustness of our method