In the following work the multi-mode inertial tracking system is proposed. The system tries to bypass the constraints regarding the device placement and typical usage scenario. The system consists of four principal modules. The first one is the data pre-processing and calibration module: it automatically calibrates the accelerometer, gyroscope and magnetometer sensors. The simple calibration is employed for the accelerometer and the gyroscope, while the magnetometer is being calibrated in realtime using Stochastic Gradient Descent approach. The second module recovers the trajectory of the pedestrian and is able to separate the forward and backward steps of the user. The third module is an orientation estimation filter with magnetometer disturbance detector. For orientation estimation a well-known complementary filter is employed. To adaptively tune the magnetometer filter gain the residuals between the virtual gyroscope and the real one are continuously handled. The fourth module is responsible for the power management of the whole system. The key features of the proposed system is that it does not restrict typical user behaviour, automatically adapts to the used sensors and could be easily fused with any existing localization technology. The first section of the paper is devoted to the practical aspects of implementation of such system for embedded devices.