An efficient algorithm for running max/min calculation

We present a novel algorithm for calculating the running maximum or minimum value of a one-dimensional sequence over a sliding data window. The new algorithm stores a pruned ordered list of data elements that have the potential to become maxima or minima across the data window at some future time instant. This algorithm has a number of advantages over competing algorithms, including balanced computational requirements for a variety of signals and the potential for reduced processing and storage requirements for long data windows. We show through both analysis and simulation that for an L-element running window, the new algorithm uses approximately three comparisons and O(log L) memory locations per output sample on average for i.i.d. signals, independent of the signal distribution