Object Detection of Surgical Instruments for Assistant Robot Surgeon using KNN

Abstract

Automatic object recognition is the main key to the auto robot system. As like the eye in humans, object recognition is a system of identification on a computer. With this capability, object recognition can be developed to help facilitate human work from monotonous and repetitive work. In the field of medical, object recognition can be combined with an operating robot to free the nurse's duties from the work of sending and retrieving surgical instruments during the operation process. One of the simplest recognition methods is using the K-Nearest Neighbor (KNN) algorithm, by finding the closest distance between the object and the dataset. Opencv is added as an additional library that aims to facilitate the pre-processing of image. This paper implements the KNN algorithm to distinguish the surgical instrument with a relation test set to train set is 1:1. The pre-processing section has been done with Opencv which is used to separate interest objects with a background. The classifier parameters that used are aspect ratio, solidity, and contour ratio. A region of interest (ROI) is required to limit the object area to be processed. It helps to find the principal axis to rotate the image. The aim is to extract the important information as a new orthogonal variable called main components. The KNN method test results show the robustness of automatic intraoperative object detection that can be used to improve the specialist's preventive state acknowledgment amid the robot-helped medical procedure.