Consistent Identification of Laparoscopic Tools by Direction- and Angle-Guided Tracking for Robot-Assisted Surgery
Youqiang Zhang, Minhyo Kim, Jun Seok Park, Sangrok Jin
Abstract
Maintaining consistent tool identities (IDs) is essential for accurate navigation and automation in minimally invasive laparoscopic surgery. However, short- and long-term tool exits, re-entries, and crossover between visually identical tools remain major challenges for robust identity tracking. We propose a direction- and angle-guided ID consistency method that integrates keypoint-based pose estimation with geometric reasoning and multi-cue association. Based on this multi-cue information, we construct a three-stage matching framework that enables reliable ID recovery under complex operating conditions. Experiments were conducted on the Cholec80 surgical video dataset. In addition, real-time validation was performed using a two-degree-of-freedom mechanism with remote center motion. The proposed method improved the ID tracking performance across all evaluation metrics, including the HOTA, AssA, and IDF1, while reducing ID switches and enhancing consistency in multi-tool scenarios. These results validate the effectiveness of our approach and highlight its potential for enhancing ID stability in robotic-assisted surgical systems.
BibTeX
@inproceedings{ral2026_consistentidenti,
title = {Consistent Identification of Laparoscopic Tools by Direction- and Angle-Guided Tracking for Robot-Assisted Surgery},
author = {Youqiang Zhang and Minhyo Kim and Jun Seok Park and Sangrok Jin},
booktitle = {RA-L 2026},
year = {2026}
}