Improving estimation of body lengths using extended Kalman Filter for squat movement

Abstract

Modeling and analyzing of human movements has become easier with the development of sensor technologies. Human movements can be modeled using image processing software with depth and motion sensors in 3D. Measurement errors are also observed in motion detection sensors as in most systems. Special filters have to be developed for each system in order to minimize this error rate and obtain more realistic measurements. Kalman Filter is a well-known method that is commonly used to minimize this type of measurement errors. In this study, the actual body lengths (upper arm, forearm, lower leg, upper leg) are measured and obtained from the human motion sensor. Kalman Filter and Extended Kalman Filter are applied to the obtained data from human motion sensor. All measurements are compared with the actual body lengths and error rate is calculated as using Mean Absolute Percentage Error (MAPE). Kinect data are compared with actual lengths and error rates were calculated at 20%, when the Kalman Filter is applied, the error rate decreased to 14%, while when the Extended Kalman filter is applied, it dropped to 8%. Human motion sensor data have been improved with using Extended Kalman Filter. Thus, actual measurements of candidatescan be easily obtained with only one useful sensor without taking any actual measurements by saving time and budget.