Research And Implementation Of Motion Capture Based On Inertial Sensor

Human motion capture is an application field involving multiple technologies and disciplines involving electronic communication,human-computer interaction,sports biomechanics,and navigation.Inertial motion is a technique that records human motion through inertial sensor.With the widespread popularity of computer technology and artificial intelligence,motion capture technology has broad prospects in the fields of medical rehabilitation,film and television animation production,virtual interactive games,sports training,and intelligent monitoring.Compared with traditional equipment,the motion capture system based on inertial measurement has the advantages of good stability,anti-interference,low cost and wide application.It is a research hotspot in the field of motion capture in recent years.However,current commercial motion capture devices are expensive,not open source,and low-cost inertial sensors are characterized by accumulating errors.Therefore,this thesis designs a systemic motion capture system that is not restricted by the environment and is easy to use based on inertial measurement technology.1.For the low-cost inertial sensor,due to its inherent nature and design principle,it is subject to environmental interference.The accelerometer is corrected once by gravity reference calibration method.The field calibration method is used to correct the drift zero point.Calibration of the gyroscope and calibration of the magnetometer using the ellipse fitting method improve the measurement accuracy of the inertial sensor and the stability of the system.2.For the single sensor's large noise and integral error leading to obvious drift,this thesis analyzes and compares the popular EKF algorithm,Mahony algorithm and Madgwick algorithm,and uses the gradient descent method based filter to perform the attitude angle data fusion of the 9-axis sensor.The algorithm's higher precision and low computational complexity make the system more individual robust while achieving faster dynamic response.3.Combining the human skeleton structure and the principle based on forward kinematics,this thesis designs and builds a set of 3D virtual character skeleton models consisting of 10 nodes including head,trunk and limbs,thus completing the capture of real human body movement.And display.Based on the zero-speed detection method,the problem of jitter and frame dropping in the bone rendering algorithm is corrected.4.For the problem that there is no mutual connection between the sensors of each node,the virtual character model has joint disorder.Based on the theory of inertial navigation,this thesis proposes a multi-sensor joint calibration algorithm based on coordinate system transformation,which simplifies the commercial products.Multi-step calibration method,the model can effectively track the body's whole body movements in a single step.