Research on robot-based surgical instrument detection and pose estimation algorithm with multi-cascade deep learning processor
10.19745/j.1003-8868.2024104
- VernacularTitle:基于多级联深度学习处理器的机器人手术器械检测和姿态估计算法研究
- Author:
Si-Qi HAN
1
;
Min-Kui CHEN
;
Li-Pu WEI
;
Qian RAN
;
Qian XU
;
Ming YU
;
Yu-Chao SUN
;
Feng CHEN
Author Information
1. 军事科学院系统工程研究院,天津 300161
- Keywords:
robotic scurb nurse;
deep learning;
surgical instrument detection;
pose estimation;
coordinate attention mecha-nism
- From:
Chinese Medical Equipment Journal
2024;45(6):1-8
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose a multi-cascade deep learning processor-based surgical instrument detection and pose estimation algorithm to facilitate the robotic scurb nurse to recognize and delivery surgical instruments.Methods The proposed multi-cascade deep leaning processor-based CYSP algorithm was hibernated with several functional modules such as YOLOX with coordinate attention block(CA-YOLOX),segment anything model(SAM)and principal component analysis(PCA).Firstly,CA-YOLOX was applied to identifying the types of the surgical instruments and completing the coarse positioning of x and y coordinates;secondly,the SAM segmenter was used to clarify the positions of the instruments in the RGB image,and the depth information and internal parameters of the camera were introduced to obtain the point cloud of the surgical instruments;finally,the center of mass,principal direction and normal direction of the surgical instrument point cloud were determined through the PCA algorithm,with which the rotation and translation(RT)matrix between the target coordinate system(surgical instrument center of mass coordinate system)and the base coordinate system of the robotic arm was solved,and the matrix was converted into a quaternion and then transmitted to the robotic arm control unit so as to drive the robotic arm to arrive at the corresponding position and pick up the instrument to complete the instrument delivery task.Migration training was accomplished on a self-constructed surgical instrument image dataset and the effectiveness of the proposed algorithm was evaluated,and instrument delivery experiments were performed on a seven-degree-of-freedom robotic arm and the success rate of the algorithm was assessed.Results The multi-cascade deep leaning processor-based CYSP algorithm had a recognition accuracy of 98.52%on the surgical instrument dataset,a success rate of 94%for the in-strument delivery experiment and average time for recognition of 0.28 s.Conclusion The multi-cascade deep leaning proces-sor-based CYSP algorithm with high reliability and practicability behaves well in facilitating the robotic scurb nurse to recog-nize and deliver surgical instruments.[Chinese Medical Equipment Journal,2024,45(6):1-8]