A new method of segmenting one-dimensional signals

Abstract

Presents a new method of segmenting a one-dimensional signal into a set of features of the following type, line, curve, hole, discontinuity, and corner. A set of requirements for the segmentation process result from the application area, which in this case is robot arc welding. Experiments with several existing methods and their advantages and disadvantages are described leading to the introduction of a new method called the Incremental Simultaneous Filtering and Fitting Approach (ISFFA). The algorithm is able to partition signals into features of the above type reliably in the presence of noise and missing data, yet is reasonably economical to implement.