Matching topographic features in 2-D images for model-based recognition in manufacturing applications

A robot vision system is presented for the recognition (identification and pose estimation) of 3D objects in modern manufacturing applications. Recognition is accomplished by matching topographic features obtained from the model images to similar scene features extracted from the scene image. The model features, extracted from intensity model images, are processed off-line and are compiled into the model database along with a set of attributes and invariance indices. Matching is performed between the scene features and the model features of different invariance levels, yielding the object identity and estimates of its pose in the scene. This system is tested through a set of objects with curved surfaces and complex surface characteristics that make their recognition using the classical techniques quite challenging. These objects are modeled using the proposed brightness models. Satisfactory recognition results are obtained and presented.