The final Total Error in processing is useful for gauging project quality. The final error is displayed as the last bar in the Total Error Dialog as well as in the Project Status Report.
Total Error is a statistical measure that is calculated in processing (bundle adjustment). In bundle adjustment we need to set a scaling value for all expected parameters precisions. This value is called sigma-naught (s0) (or more properly a priori sigma-naught (a priori means 'before')). In PhotoModeler we set this scale to 1.0. When all assumptions of the adjustment are met, the output sigma should be 1.0 also. This means that the total error value on the last iteration should be 1.0 if everything matches our expectations. What does it mean when the final total error is not 1.0?
When the final total error is less than 1.0, we can say that the data model (including camera parameters etc.) is good and the marking is more precise than the assumptions. These assumptions are shown in the Preference Dialog - Process / Default Point Standard Deviation Weights. For example, if the standard deviation for all the sub-pixel marked points in a project is 0.1 pixels and the final total error of processing that project is 0.5, then one could say that the actual marking precision is on average actually 0.05 pixels (0.1 * 0.5) and not 0.1. In other words you got better marking precision than first assumed. This is only a first assumption though because the camera calibration, for example, could perhaps be improved even further and then maybe the final total error would have been even smaller.
When the final total error is over 1.0 then some of the initial assumptions are incorrect. Those incorrect assumptions could be that the camera was properly calibrated, or that the marks are marked to the assumed precision (defaults are 1.0 pixels (one standard deviation) for manually marked points, and 0.1 pixels for sub-pixel points). When the total error is very large (let's say over 10) then there are probably gross errors too - such as mis-referenced points or a very badly calibrated camera.
In conclusion, when the final total error is 1.0 or smaller then we probably have a reasonably good project and have met our expectations for marking quality and camera calibration. Note, it does not mean that it can't be improved further. When the final error is between 1.0 and approximately 10.0 then perhaps our marking precision assumption is wrong or the camera is not well calibrated. When the final error over 10.0 we may have a poor camera calibration but as well possibly some gross errors such as mis-referenced points or edges (or badly defined constraints). High total errors should be investigated (look for gross errors and consider how the camera was calibrated).
Note that we typically hold different project types to different standards. While a high-accuracy industrial survey may be unsuitable with a total error over 2.0, an accident reconstruction road project, or large architectural survey may be fine with a total error as high as 10.0. The higher error says that you could not meet your assumptions of marking points to 1.0 pixel precision (which is accepted when manually marking points on blurry road features, window corners, etc.). The higher error does not necessarily mean the project itself is bad or inaccurate. In Inverse Camera projects we accept even higher total error values (perhaps up to 15.0). Regardless, when the total error is very high (over 20) you should investigate the sources of errors (mis-marked or mis-referenced points, camera incorrectly calibrated, etc.). If you are carrying out very high accuracy work (target dots in a precise industrial survey for example) then search your project for problems at lower total error values (over 2.0).
See also Total Error Dialog.