Institutional Repository
Technical University of Crete
EN  |  EL



My Space

Motion structure analysis in rivers for evaluation of dangerous events

Bacharidis Konstantinos

Full record

Year 2016
Type of Item Master Thesis
Appears in Collections


Motion estimation is a widely researched topic of Computer Vision with numerouspublished algorithms. Nowadays, Computer Vision and motion estimation are utilizedto confront real world problems. One such application, with on-going research interest,is the estimation of the motion eld of rivers using video data. The incorporation ofComputer Vision in river ow estimation can lead to the development of a low-cost, fast,accurate and above all, a non-intrusive method of measuring the river's velocity. Thesecharacteristics are of most importance since until now, this estimation required on-eldmeasurements with costly conventional equipment, such as accelerometers or doppler-based devices.Moreover, natural phenomena that have rapid occurrence, such as ash ood streams,are hard to monitor at their full extent with the use of conventional equipment. On theother hand, a video surveillance system accompanied with a motion estimation algorithmwould allow constant and accurate measurements of the ow without requiring the on-eld presence of hydrologists or any other man-held equipment, under the condition thata light-source is available. However, the development of a motion estimation algorithmfor the case of uids can be a dicult task due to the dynamic motion nature of theuid, guided by both internal and external forces, such as wind and gravity. In addition,when applying such methodologies to real world problems, such as the development of ariver monitoring system, researchers must also deal with a number of other tasks beyondthe selection of an appropriate motion estimation algorithm. These tasks span from theselection of hardware equipment to be used and the monitoring layout formation as wellas the data mining capabilities that oers, to even the derivation of a way to associatethe 2-dimensional image space with the 3-dimensional real world in order to reach to areal world velocity estimate for the river's velocity.This thesis addresses the problem of developing an Image-based monitoring layoutfor river ows presenting and discussing the methodologies behind each module. Morespecically, we propose the utilization of a stereo camera monitoring layout allowing therelation of the real world and image plane coordinate systems and thus the 2-Dimensionalimage plane motion and the 3-Dimensional river surface motion. Moreover, a novel proba-bilistic motion estimation method based on stereoscopic data is presented which evaluateseach possible displacement based on the two views of the monitored scene resulting in adense global motion eld that accurately represents the motion patterns present on the river's surface. However, this thesis does not comments upon the performance and costtrade-o of the used hardware.

Available Files