Object tracking is an essential step of a video processing pipeline. In the Fish4Knowledge project, recognizing fish trajectories allows to provide information to higher-level modules, such as behavior understanding and population size estimation. However, video quality limitations and appearance/motion characteristics of fish make the task much more challenging than in typical human-based applications in urban contexts. To solve this problem, robust appearance and motion models must be employed: this chapter describes an approach devised to tackle the fish tracking problem in this project, and presents and evaluation of the tracking algorithm in comparison with state-of-the-art techniques.
|Data di pubblicazione:||2016|
|Appare nelle tipologie:||2.1 Contributo in volume (Capitolo o Saggio)|