There are a number of factors that contribute to the overall accuracy of a project. One of the large factors is the precision of the point marking and the referencing of like points across photographs. With most types of points marked the user has to zoom in, and click to mark the point location. There is a fair amount of inaccuracy in this operation. Even on high contrast good targets, a human operator is accurate to only one to three pixels. When you then take into consideration that many points look different in photographs taken at different angles, the overall precision of identifying exactly the same point in object space is usually around three to ten pixels at photo scale.
One way to overcome this limitation of manual photogrammetry is to make the software mark the points in an accurate and consistent manner. It is difficult to do this on any type of natural point (e.g. corner of a window frame or other distinguishable feature) but if we know a priori what the target is going to look like, the software can do quite a good job of marking it by studying the digital image data. This is what the sub-pixel target marker does.
Note that there is a manual sub-pixel target mode which we will describe here and an automated method described in the Automatic Target Marking section. Sub-pixel target marking also applies to Coded Targets and Automated Coded Targets Project. Much of the background covered here applies to these other areas also.