การจำแนกท่าทางการเคลื่อนที่ของมือจากภาพ 2 มิติ และการวัดพิกัด 3 มิติ สำหรับการควบคุมภาพในห้องผ่าตัด

Abstract

This research presents hand movement classification from 2D images for image control in the operating room and uses stereo vision techniques to track hand movements in 3D to measure the distance from the camera to the hand. Compare the camera's best viewing distance. The work process can be divided into 3 processes: 1) the process of creating 3D image data using stereo vision techniques, 2) the process of detecting the position of the hand in a 2D system, and 3) the process of classifying hand gesture commands from the image. 2D in medical image control in steps 2 and 3, the experiment will be divided into 2 cases: wrist classification and wrist classification cases. To compare the accuracy of positioning various gestures in the experiment. There are three experiments to measure the performance of the system: 1) a depth measurement experiment with stereo vision techniques, 2) a hand movement detection experiment, and 3) a hand gesture classification experiment. Experiment 1 had an accuracy in measuring the distance from the camera to the hand at 63.569% at a camera distance of 140 centimeters. In experiment 2, in the case of not classifying the wrist, the error value between the centroid coordinate and the hand center coordinate (Template coordinate) had similar values. The maximum of 2.426 centimeters is the camera at 60 centimeters (Camera 1). In the case of classifying the wrist, the error value that is closest to 0.924 centimeters is the camera at 100 centimeters (Camera 2). And in Experiment 3, a total of 3 groups of gestures can be classified. (11 movements) In the case of wrist classification, it has the highest accuracy of gesture classification, with accuracy 95.4%, sensitivity 93.6%, and specificity 98.8%. The camera distance with the best classification is 60 cm (camera 2) in this case. that does not classify the wrist gives the highest accuracy, which is accuracy 92.3% sensitivity 87.9% specificity 100%, with the camera distance that has the best classification being 60 cm (camera 2). From the experiment, it can be concluded that Stereo vision systems can measure the distance from the camera to your hand. And it can detect the position of the hand and classify hand movements in a 2D system.