Modification and fixed-point analysis of a Kalman filter for orientation estimation based on 9D inertial measurement unit data

Abstract

A common approach for high accuracy sensor fusion based on 9D inertial measurement unit data is Kalman filtering. State of the art floating-point filter algorithms differ in their computational complexity nevertheless, real-time operation on a low-power microcontroller at high sampling rates is not possible. This work presents algorithmic modifications to reduce the computational demands of a two-step minimum order Kalman filter. Furthermore, the required bit-width of a fixed-point filter version is explored. For evaluation real-world data captured using an Xsens MTx inertial sensor is used. Changes in computational latency and orientation estimation accuracy due to the proposed algorithmic modifications and fixed-point number representation are evaluated in detail on a variety of processing platforms enabling on-board processing on wearable sensor platforms.

A cascaded two-step Kalman filter for estimation of human body segment orientation using MEMS-IMU

“…In many ambulatory biomechanical analyses, motion tracking of human body segments by accurate determination of each segment's orientation is of key importance [1]- [2]. The diverse application of body segment motion tracking ranges from rehabilitation and physical medicine to sports science [3]- [4].…”

Multi-Sensor Fusion and Error Compensation of Attitude Measurement System for Shaft Boring Machine

“…At the same time, to reduce the data acquisition and calculation burden of remote control centers and improve the ability of autonomous attitude measurements and rapid response for the SBM, it is necessary to implement the algorithm in attitude measurement units and the embedded processor of the monitoring terminal, but the implementation complexity and computational efficiency of the hardware and software must be considered. Due to the speed limit, the 8-bit and 16-bit microcontrollers only support fixed-point operation, have some difficulties in algorithm implementation, and can barely run some simple algorithms [33,34]. However, the flexibility of the algorithm is obviously improved on advanced RISC (reduced instruction set computing) machine (ARM), field programmable gate array (FPGA) and other high-speed chips.…”