Dear Community, I am wondering if there are articles/books/references proposing implementations of Kalman Filters for positioning based on a single accelerometer ? (no GPS, no camera, no gyro, no odometery...)
if you have only acceleromter measurement, then your system is not observable. For example assumming Constant Acceleration model (A = [1 T *T/2; 0 1 T; 0 0 1], C = [0 0 1]), then observability matrix is M = [0 0 1; 0 0 1; 0 0 1]. You can estimate position, but covariance will constantly growing.
In this case you can determine the position each time stamp (epoch) as a predicted value only and there is will not be any measurement model that you need for update stage. In this case, I think you can a condition like zero velocity update (ZUPT), but your positioning accuracy will still low.
your positioning accuracy will still low ,so, you can only determine the position each time as a predicted value only and there is will not be any measurement model that you need for update stage
thank you and good luck
consult this work: Filtre de Kalman discret à la modélisation HydrologiqueFebruary 2019