Tracking Correction Method for Rapid and Random Protein Molecules Movement

Satoshi Kamiya, Keisuke Toida, Taka-aki Tsunoyama, Kazuhiro Hotta; Proceedings of the Asian Conference on Computer Vision (ACCV), 2024, pp. 3022-3037

In recent years, there has been a increasing demand for tracking protein molecules with the focus on immune system researches. However, machine learning-based single-particle tracking (SPT) faces the challenges in accuracy due to the rapid and random movement of molecules as well as detection errors. To address these issues, we use frame interpolation to pseudo-decrease the speed of movement and perform two-stage matching to achieve stable tracking. We also use an optimization algorithm that concatenates short tracks. This approach has achieved higher performance on the CD47 dataset and the PTC dataset than conventional baselines.