Analysis and processing of results

General

The analysis step is accessible when all the image processing has been done. It can be done from only a few images to have a quick idea of the velocity fields. This step allows a statistical analysis of the results in order to extract average velocities, turbulent distributions and to make additional filtering to those already carried out in the displacement methods. This step is accessed through the menu Process >> Analysis. A dialog box opens.

Analysis options panel

In this analysis step, the displacements in pixels per image are transformed into meters per second. The processing depends on the choice of pre processing chosen method. For example, it is a simple multiplication and division by the resolution and the frequency of the video in the case without transformation. For each calculation cell, the standard deviation of the distribution of the measurements after filtering is calculated and can be visualized in the following step.

Statistical analysis cell

The parameter cell size allows to define the size (in meter) of the cell for spatial averaging. Its value depends on the precision of the expected results. To have a high spatial resolution, this number must be small. But the computation time will then be greater, the statistical convergence lower (less measurement samples in the area). On the other hand, a value that is too large reduces the spatial resolution but speeds up the calculations.

Import of the DEM

In order to quickly estimate a flow from velocity measurements, it is possible to import a bathymetry. To do this, select the DEM box and enter a value for the free surface elevation. The bathymetry must be in .XYZ format. It is possible to enter the bathymetry file in the interface that will open when you press the Compute button. A summary visualization of the bathymetry appears. If it seems correct, we can close the corresponding graph.

If you use a pre-processing based on the position of the camera (camera pose or UAV), you must give a value to the free surface level. This is the relative position with respect to the vertical axis of reference (takeoff position for the UAV).

Filtering

The first possible filtering concerns the number of detections on a trajectory. If the number of detections is too low (few consecutive detections of the same particle), we consider that the measurement is unreliable. In this case these movements are removed from the calculation of the average.

To filter out the outliers, we use the quantiles of the displacement distribution in the cell. The threshold 90 % means that only measurements between 10 % and 90 % of the value distribution are taken into account. This parameter can thus evolve between 50 and 100 %. However, it is important to ensure that the number of samples remaining is significant.

Uncertainty analysis

If a homography is chosen (2D only for this version), the uncertainty due to the position of the GRPs on the ground and on the image is taken into account. A monte carlo simulation is performed with predefined distributions of the projection file data. The number of simulations (10000) and the parameters of the distributions can be modified in the interface. Each of these homographies is then applied to a regular grid in order to have the displacement uncertainty for each zone of the ROI. The uncertainty of the velocity measurement is then obtained by multiplying the uncertainty by the value of the measured displacement to obtain an uncertainty in m/s.