In almost all algorithms I have been exposed to it is taken for granted that the data will always be available. The motion data often had missing samples, sometimes for a single data point and sometimes for large portions of the capture. Every piece of processing I did had to be aware that it could not rely on a give point existing. A great help in solving this problem was the Masked Array extension to Numeric Python[MA]. This provides a facility to operate on arrays with an associated mask that represents missing or invalid values. These masks are propagated through arithmetical operations, so adding a valid element to a masked element will result in a masked element. If care was not taken a single invalid point could invalidate all of my data, but once those dependencies were resolved the results were much more reliable.