For Full-Text PDF, please login, if you are a member of IEICE,|
or go to Pay Per View on menu list, if you are a nonmember of IEICE.
Anomaly Detection of Knot-Tying Action in Surgical Training by Probabilistic Frame Anomalousness Estimation
Yoko OGAWA Tadashi MATSUO Nobutaka SHIMADA Yoshiaki SHIRAI Yoshimasa KURUMI Masaru KOMORI
D - Abstracts of IEICE TRANSACTIONS on Information and Systems (Japanese Edition)
Publication Date: 2018/03/01
Online ISSN: 1881-0225
Type of Manuscript: Special Section PAPER (Special Section on Student Research)
anomaly detection, motion recognition, state transition model, surgical training system,
Full Text(in Japanese): PDF(2.4MB)
>>Buy this Article
We propose an anomaly detection method for self-training system of surgical procedures. The method stochastically estimates anomalousness of input frames by temporal filtering based on 3D point cloud registration. Proposed motion model consists of regular and anomaly state pairs assigned to each model frame, and it allows the system to detect anomaly frames with frame-wise matching. The system adapts the model parameters for each input sequence, then it detects anomaly frames with more major errors preferentially. We constructed the model from a professional surgeon's and a beginner's knot-tying motion, and showed effectiveness of our method for detecting anomaly frames.