Objectives With the development of indoor positioning technology towards continuity, precision, and intelligence, autonomous indoor positioning technology has become a research hotspot. The inertial acceleration and magnetic field values of the carrier can be obtained using micro electro mechanical system inertial measurement unit (MEMS IMU), which is an important device in positioning. However, there are difficulties in the complex noise of dynamic measurement values. For the high noise of measurement signals in low-cost MEMS IMUs, the noise separation and denoising of dynamic magnetic sequences are conducted by using blind source unmixing separation algorithm.

Methods An error model using multiplicative parameters and additive errors for MEMS IMU dynamic measurement is constructed, which expresses the error source of magnetic signals in real-time. The magnetic measurement noise disturbance law of MEMS IMU is analyzed at waist, hand and foot, then a new algorithm magnetic nonnegative matrix factorization is constructed based on the nonnegative matrix factorization with adding the reweighted sparse constraint equation. The static magnetic value were measured using an FVM-400 magnetometer in study area, and magnetic sequences of pedestrian in walking using MEMS IMU. it were conducted on blind source unmixing separation and denoise test of magnetic signals using fast independent component analysis (FastICA) and magnetic nonnegative matrix factorization (MNMF) algorithms.

Results The results show that: the performance index of FastICA and MNMF is less than 0.1 in the simulation test of magnetic disturbance unmixing. It can effectively separate the walking noise, equipment vibration noise and static magnetic sequence in the magnetic signal. FastICA has better separation effect on periodic noise, MNMF can accurately unmix the static magnetic value, and the correlation coefficient reaches 0.996 8. In the unmixing experiment of the measured dynamic magnetic sequence, the MNMF algorithm can separate signals similar to the static magnetic sequence, which keep relatively good magnetic spatial sequence feature, but the separation accuracy for other noises is poor. In addition, MNMF algorithm has obvious noise reduction on the measured dynamic magnetic sequence. The experiment shows that the matching probability of the magnetic sequence after denoising with the MNMF algorithm can be significantly improved, especially for magnetic sequence at hand after denoising. The probability reaches about 95 % with the 3-meter matching length.

Conclusions For the unmixed and separated the measured magnetic sequence using the MNMF algorithm, most irrelevant interference noise can be separated, and the unmixed signal is similar to the static magnetic sequence, which can improve the probability and accuracy of the geomagnetic matching positioning in the dynamic walking of the pedestrian. The noise separation of dynamic magnetic measurement using MNMF has certain effectiveness, which provides a new theoretical basis for magnetic measurement denoise of indoor autonomous positioning.