TransitionProbs:
=====
* higher p(s|s) before doors?
* higher p(s|s) next to tables?
* lower p(s|s) ons stairs/corridors?

* p(s| s', s'') => difficult
  * delta components in emission vectors
* allow 0,1,2 Model
  * by setting intelligently transition probs to ZERO

EmissionProbs:
=====
* weight delta by euclidian distance to AP?
* currently best strategy: euclidian full (but top 3 manhattan is competitive)

Signal Preprocessing
=====
* Kalman filter?
* interpolate signal into the Future?

Error Pre-Estimation
=====
* evaluate gradients (sobel 3D?)
* use for optimizing AP deployment
* calculate error plane from measurement/simulation errors and compare to gradient-error

Pruning
=====
* since the cloud of UNPRUNED states are the top-results build
  we could interpret the cloud geometrically

Title
=====
* HMM-driven Localization and Tracking using Raytracing supported Rich-World-Models.