An Autonomous Intervention Strategy for Robotic Soft Endoscope Guidedby Anatomical Features

An autonomous intervention strategy for soft endoscope in upper digestive tract is studied based on the previously designed YunSRobot endoscopic robot. The anatomical features of digestive tract are extracted from the endoscopic images based on the Faster-RCNN (faster-regions with convolutional neural network feature) algorithm and mathematical morphology image processing. Then an autonomous endoscope manipulation and orientation strategy is designed to perform autonomous endoscope intervention in the upper digestive tract. Comparative experiments are carried out on a highly simulated phantom of upper digestive tract, and the results show that the intervention success rate of the proposed method is 100%. Compared with the master-slave control based intervention by the unprofessional operator, the average intervention duration of the autonomous intervention is reduced from 262.01 s to 197 s, and the maximum insertion force is decreased from 11.8 N to 9.6 N. Compared with master-slave control based intervention by the professional operator, the maximum insertion force is almost the same, and the average intervention duration is only 34.33 s longer. The autonomous intervention strategy proposed also can be applied to soft endoscope intervention in other natural orifice such as respiratory and urinary tracts.