Abstract
The author previously applied extended Kalman filter for estimating the posture of depth sensor attached to a walking person or a mobile vehicle, where the posture of the sensor was included in the state vector. However, it cannot adapt to situations where the initial value of extended Kalman filter was not adequate. In various textbooks of Kalman filter, the initial value of the filter algorithm has not been discussed extensively. However, it is very important, especially for the model when the domain of the state vector has several groups and different starting point leads to different results. In this paper, the initial value setting of the algorithm is proposed. It is simple, but it has proven to work well and has a good property to adapt to various situations. By defining the reset conditions differently, it is shown that the system can detect different surfaces.
