Scalable Operators for Feature Extraction on 3-D Data

Real-time extraction of features from range images can play an important role in robotic vision tasks such as localisation and navigation. Feature driven segmentation of range images has been primarily used for 3D object recognition, and hence the accuracy of the detected features is a prominent issue. Feature extraction on range data has proven to be a more complex problem than on intensity images due to both the irregular distribution of range images. This paper presents a general approach to the development of scalable derivative operators using a finite element framework that can be applied directly to processing regularly or irregularly distributed range image data. The gradient operators of varying scales are evaluated with respect to their performance on regular and irregular grids.