Automated Tracking of Beaked Whales with Integrated Track Smoothing and Stitching

Abstract

Passive acoustic monitoring (PAM) is a powerful and non-intrusive tool for studying marine mammals, particularly rare and deep-diving species such as beaked whales. However, the post-processing and manual analysis of large data sets is time-intensive. Multi-target tracking (MTT) methods have significant potential in automating this process and reducing human workload, but face challenges due to the potential absence or high directionality of beaked whale vocalizations. These two effects cause high variability in target detection probability, which can cause standard MTT trajectories to fragment. In this paper, we address this issue by introducing a multi-step estimation method that combines belief propagation-based MTT with track smoothing and stitching. Using real recordings of echolocation clicks from Ziphius cavirostris (Cuvier's beaked whales), we demonstrate that track stitching can generate continuous tracks in the presence of a significant number of consecutive missed detections.