다중 센서를 이용한 실사 인체 크기 검출에 관한 연구

Abstract

Motion capture is a technology that measures the specific information such as location, speed and direction for movement of a target in 3D space, and analyzes and applies by recording it into the form that can be used in computer [1]. In recent years this motion capture technology has increased its application field including medical field and sport coaching system which helps the rehabilitation of the physically disabled as well as games, movies and animation. The motion capture method developed until now can be divided into the mechanical, optical and magnetic. However, the equipment price is expensive and it has the problem that the output size of virtual character from the motion capture system is output to a certain size that was assigned by the user without considering the actual size of the target. In the past, in order to apply the actual size of the target on the virtual character the user measured the length between each joint of the target manually. In order to solve this problem, the study that automatically measures the actual size of the target is needed. In this paper, we used Kinect sensor and the inertial sensor that provides three-dimensional location information of each joint to measure the actual size of the target automatically. Kinetic sensor is able to acquire real-time imagery information by providing both Depth images and color images of 30 frames per second. Inertial sensor is a 9-axis MEMS sensor consisting of 3-axis gyro sensor, 3-axis accelerometer sensor and 3-axis geomagnetic sensor. It has no constraint on the experimental environment with its wired and wireless communication manner and it can detect accurate motion capture data because the sensor is attached to the body of the target. Motion data files output from two sensors are BVH files and can be divided into Header section and Motion section. Header section provides the location information of each joint and Motion section provides the information of X, Y and Z axis direction of each joint. Therefore, by integrating Header section of BVH file output by detecting the actual size of the target from Kinect sensor and Motion section of BVH file output from inertial sensor, one BVH file was regenerated. As a result of experiment of two targets with different height in accordance with the proposed method, the same result with actual height was detected. Existing method has applied the size of the target to the virtual character by measuring manually, but now using Kinect sensor the motion data file can be output by measuring the actual size of the target automatically so it is expected to be able to apply to the virtual character.