Model Precision

Measures the precision of models by walking the user through measuring a selected feature in all the models that show it. Model manager must be active to use this tool.

The model precision allows a user to select a feature of interest in the current stereo model. It then searches for all other models that show that feature and open each in sequence to allow additional measurements of the same feature. After all observations have been made the comparison is reported in a textbox and optional to a file. In order to use the tool you must first open multiple models in the Model Manager (ModMan).

2. Open the Model Precision Application (ModPre). Optionally, before beginning the measure sequence, you can search for a better starting model using the "Center Model" and "Next Model" buttons.

ModPreMenu1stPt

The model precision application menu.

F1, or left click	Pick a feature to measure in the models.
F2	Re-center the view in the best model.
F3	Toggle the view through all the models that show the current location.
F7	Allows acces to the user parameters discussed above. Parameter descriptions are given in step 3 below.

ModPreParams

Parameters:

Random Motion Distance	When swapping models the cursor is shifted randomly in all three axes in order to keep measurements independent. This can be disabled by setting this value to zero, but leaving it at least several times the GSD is highly recommended.
Show model data	There is an optional block of data that shows the model name, the residuals from the mean measurement, the pixel radius (percent of the distance from the image center to the corner), and the image names for each model (see the example in step 6).
File recording mode	User can chose to record nothing to the output file, append to output file, or overwrite the output file
Output file	The path to the output file

4. Pick a feature that can be repeatedly measured. In the example below the user pick the head of the traffic arrow.

5. After measuring the first feature the application will drive to other models for additional measurements. The menu options change for remeasurement of the feature.

ModPreMenuRemeasure

F1, or left click	Re-observe the feature at the cursor location, and then move to the next model.
F3	If a point is too close the edge of the model or obscured you can skip a model.
F4	If a point is determined to be undesirable this restarts the application without saving data.
F5	This restarts the application, but saves the data that had been recorded.
F7	Allows access to the user parameters discussed in step 3 above.

Your progress through the stack of models is show on the main window as shown below.

ModPreProgress

6. The precision statistics are shown in the text box (and optional saved or appended to a file). Below is an example with model specific data turned on. Root mean squared error, or RMSE, is with regard to the differences of individual measurements from the mean. The maximum differences are the maximum deviations from the mean in each axial direction. RadL and RadR are descriptions of how far a point was away from the center of the raw image. 100% is as far away as possible, 0% is exactly in the center.