A common issue with computing speed (and acceleration) from distance and time data is noise. The noise in this type of photogrammetric measurement has two sources: a) positional inaccuracy (how well can we compute the position of this point on a road surface with maybe a low camera angle), and b) frame timing inaccuracy. The latter is based on the accuracy of the times written by the video recording device (camera or DVR). If there is some time inaccuracy (or jitter) then the delta times used to compute speed can be off. Combining these two can lead to small inaccuracies in the speed calculation (and even worse inaccuracy in the acceleration which is a derivative of the speed). The time errors can be filtered somewhat using the time average methods described above.
To counter this noise issue, PhotoModeler has speed filtering. The idea is to filter out some of the noise by using averaging and windowing techniques of multiple speed entries over time. The size of the filter is controlled by the Filter Half Window Size setting (FHWS). A FHWS of 1 means the window is 3 speeds wide (current value +/- 1). An FHWS of 2 means the speed filter window is 5 speeds wide. The larger the FHWS, the smoother and more filtered are the speed values (and hence also the subsequent computed values like acceleration). An FHSW of 0 means that speed filtering is turned off.
The filter is done with the samples being bookmarks and not frames – so this means for filtering to work in a reasonable way, the bookmarks should have some what even spacing of frames.
The CSV file report has a row at the bottom indicating which speed filtering method was used:
Also see time averaging.