EGU General Assembly 2020
© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

On the use of LSPIV-based free software programs for the monitoring of river: testing the PIVlab and the FUDAA-LSPIV with synthetic and real image sequences

Dario Pumo, Francesco Alongi, Giuseppe Ciraolo, and Leonardo Noto
Dario Pumo et al.
  • Dipartimento di Ingegneria, Università degli Studi di Palermo, Palermo, Italy (

The development of new image-based techniques is allowing a radical change in the environmental monitoring field. The fundamental characteristics of these methods are related to the possibility of obtaining non-intrusive measurements even in adverse circumstances, such as high flow conditions, which may seriously threaten the operators’ safety conditions in traditional approaches.

Optical techniques, based on the acquisition, analysis and elaboration of sequences of images acquired by digital cameras, are aimed at a complete characterization of the river instantaneous surface velocity field, through the analysis of a floating tracer, which may be naturally present or artificially introduced. The growing availability of a new generation of both low-cost optical sensors and high-performing free software programs for image processing, is a key aspect explaining the rapid development of such techniques in recent years. The best known optical techniques are the large scale particle velocimetry (LSPIV) and the large scale particle tracking velocimetry (LSPTV).

This work is aimed to analyze and compare the performance of the two most common free software packages based on LSPIV (i.e. the PIVlab and the FUDAA-LSPIV), which use different cross-correlation algorithms. The test is carried out by analyzing several sequences of both synthetic images and real frames acquired on natural rivers under different environmental conditions (with tracers artificially introduced). An image sequences generator has been implemented ad-hoc with the aim to create, under fixed configurations, synthetic sequences of images, simulating uniformly distributed tracers moving under controlled conditions. The various configurations are characterized by different parameterization in terms of: (i) flow velocity (S=slow or F=fast flow conditions, according to a logarithmic transverse flow profile); (ii) tracer particles size (CON= disks of constant diameter; VAR=disks of variable size with given mean diameter); (iii) seeding density per frame (density: low -LD, medium -MD, high -HD).

The synthetic sequences are processed by the two software packages together with the real sequences, analyzing the errors in terms of mean value of the surface velocity field and velocity along a transverse transect, with respect to a benchmark velocity (i.e. that imposed in the image sequence generator for the synthetic sequences and that deriving from the use of traditional sensors, i.e. ADCP, for the real sequences).

How to cite: 1

Display materials

Display file

Comments on the display material

AC: Author Comment | CC: Community Comment | Report abuse

Display material version 1 – uploaded on 30 Apr 2020
  • CC1: Comment on EGU2020-10155, Jérôme Le Coz, 01 May 2020

    Thanks for this interesting comparison of PIVlab and Fudaa-LSPIV on synthetic and field data.

    Could you please tell what post-processing you applied to the velocity results, if any? Filtering spurious velocity results is usually necessary to get accurate results, especially in low-seeded flows where the lack of tracers may induce underestimated velocities with low correlation levels.


    • AC1: Replay to CC1, Dario Pumo, 02 May 2020

      Thanks for your question.

      The main purpose of this work is to compare the two software programs especially with regard to the image analysis phase, while we are less interested in the study of the influence of pre- and post-processing procedures (specific for the different software) on the image velocimetry results, since the optimal reproduction of actual velocity is not our primary objective.

      In the synthetic analysis, obviously we don’t need pre- and post-processing since the tracer is spatially distributed and moves under controlled and we do not expect spurious velocity results. In this case, we only applied the standard procedure of each software for interpolating missing data and averaging the velocity results (using all the image pairs of the sequence).

      For the real sequences, we used the same pre-processing module (that of Fudaa) for images orthorectification and stabilization, while only mild impact post-processing was applied, trying to make the same assumptions for the two programs. In particular, we applied a filter on the two velocity components, with limits given by mean(u) + or - n x SD(u) and mean(v) + or – n x SD(v) (where u and v are the two components of the raw velocity vectors, SD is the standard deviation and n is a parameter). For n we used the default parameter of PIVLab (i.e. 7). We also imposed a filter for correlation values, imposing a minimum threshold of 0.4 (i.e. the default value for Fudaa). Then we applied the same procedure used in the synthetic case for interpolating missing data and averaging the velocity results. Of course, in the real applications, a deeper calibration of the filter parameters, specific for each case study and adopted software, must be done to obtain optimal results, but this was out of the scope of this work and the comparison between the two software would be, in our opinion, “affected” by the post-processing procedure.

      • CC2: Reply to AC1, Jérôme Le Coz, 02 May 2020

        Thanks for your reply.