Rastreamento e contagem de peixes utilizando filtro preditivo

Abstract

This work addresses the problem of estimating the number of sh that swim by a an observation window installed in a sh transposition mechanism, also known as a sh ladder, which are built to assist them in their upriver journey to spawning grounds. Our system and a sh classier module compose a complex computer vision based systemfor monitoring and evaluating the adequacy and eciency of sh ladders. Due to the opulent biodiversity of Brazilian ecosystem, this represent a formidable task, whose main concern is with species that are either in extinction or economically relevant. The target site is the hydroelectric generation plant of Igarapava, on the Rio Grande river in Minas Gerais, Brazil. Our method compares positively to existing ones in that it is able, besides counting, to acquire relevant information. Characteristics of dierent sh species that may be obtained are swimming ability, time of migration and peak ow rates. Our approach is based on image sequence analysis and on the use of a bayesian multiple-blob tracker (BraMBLe). This integration enables tracking shes reliably evenunder severe environmental changes, such as variations in illumination parameters and water characteristics. Also, sh trajectories in the eld of view and other necessary information for their accurate counting are achieved. Fish migration are estimated by dividing the scene in regions of interest that are constantly monitored by the system. Our methodology was successfully validated with videos streams acquired at the shladder, and it was able to attain overall accuracy as high as 80.95%.