DOIONLINE 
DOIONLINE Publish In |
International Journal of Advances in Electronics and Computer Science-IJAECS |
 Journal Home Volume Issue |
||||||||
Issue |
Volume-9,Issue-2 ( Feb, 2022 ) | |||||||||
Paper Title |
6D Posture Estimation In Medical and Nursing Care Using Monocular Color Images | |||||||||
Author Name |
Yoshimoto Kurihara, Tad Gonsalves | |||||||||
Affilition |
Department of Information & Communication Sciences, Faculty of Science and Technology, Sophia University, Tokyo, Japan | |||||||||
Pages |
16-20 | |||||||||
Abstract |
Abstract - Current object posture estimation research estimates the position and angle of an object by storing a 3D model of the object in advance in a computer and comparing it with the model. However, such a module is too large to be processed in a mobile robot, such as a nursing care robot.In this study, we aimed to develop a module that can be processed in a mobile robot, and succeeded in creating a module that is simpler, smaller, and faster than existing research. Our 6D posture estimation module estimates the class of an object, its coordinates, and the rotation angle in 3D space from a single monocular RGB image. We developed two types of modules: one using EfficientNetV2-M as the inference network and the other using AlexNet, and tested their accuracy against each other. In addition, we compared the estimation accuracy of different camera locations in order todetermine the best location to install the camera under indoor conditions. The highest accuracy was obtained when the camera position was set to 30 degrees and the EfficientNetV2-M-based module was used, with a classification accuracy of 100% and a 6D posture loss of about 0.051.Since the distance from one end of the room to the other in the image was about 9meters, simple conversion of this 6D posture loss into coordinates and rotation angle resulted in a coordinate error of about 3.8cm per direction and a rotation angle error of about 1.5° per direction. Keywords - 6D posture estimation, Efficient Net, Alex Net, Deep learning, Image recognition | |||||||||
| View Paper | ||||||||||
