A Multi-Target Tracking Method for Phytoplankton Based on Adaptive Buffer Intersection-Over-Union

In this study, we propose a novel multi-target tracking method based on Adaptive Buffer Intersection over Union (A-BIOU), aiming to effectively track phytoplankton, especially those targets with irregular motions and indistinguishable appearance. To address the challenge of tracking irregular moving targets, we first extend the space for detection and trajectory matching by adding buffers to compensate for the bias of Kalman filtering in motion estimation. However, improperly setting the buffer size may lead to the problem of misassociation of different targets. To cope with this problem, we propose a new association strategy. Specifically, we classify trajectories into high-density region trajectories and low-density region trajectories by Gaussian density estimation and perform the association in two steps. In the first step of association, we match the high-density region trajectories with the detection using a smaller buffer Intersection over Union that is adaptively adjusted; while in the second step of association, we match the remaining trajectories with the detection using a larger buffer Intersection over Union that is adaptively adjusted. Although, our method is relatively simple, it works well in the multi-target tracking task of phytoplankton. We validated this paper's method using Peridinium umbonatum var. inaequale as an experimental subject, and the experimental results show that the evaluation metrics of IDF1, ID
SW
, HOTA, and MOTA of multi-target tracking reach 83.91, 317, 76.93, and 92.87, respectively, with an accuracy that is significantly better than that of existing multi-target tracking methods. This provides an effective means for accurately obtaining phytoplankton community movement parameters and developing comprehensive toxicity analysis techniques based on phytoplankton movement and behavior characteristics.