Exploring energy efficiency of hardware-architectures for IMU based orientation estimation

Abstract

Various applications demand mobile, accurate and reliable motion capturing. Wireless inertial measurement units provide non-reactive and undisturbed real-time tracking of human movements. The use of customized sensor fusion methods is crucial for achieving the highest orientation accuracy. Due to the high dependency of movement data on orientation estimation accuracy based on inertial measurement data a comparison of the various algorithms presented in literature is hampered. Furthermore, the execution time and energy efficiency of the algorithms on programmable, heterogeneous and ASIC platforms is an important design space parameter. Therefore, this chapter compares the orientation estimation accuracy of inertial sensor fusion algorithms based on a single data set using a highly accurate optical motion capturing system as reference. The data set comprises of RAW data of from a commercial and from a custom developed wireless measurement unit. The provided energy consumption, execution time and achievable accuracy analysis of eleven different sensor fusion algorithms enables the selection of the optimal orientation estimation algorithm with regard to application demands and processing architecture. Furthermore, a customized processor core with the presented Kalman filter based orientation estimation hardware accelerator demonstrates techniques for increasing energy efficiency in respect of achievable throughput rate.