Presentation.md

theme: sudodoki/reveal-cleaver-theme title: PyData Berlin 2014 Talk - IPython and Sympy to Develop a Kalman Filter for Multisensor Data Fusion author: name: Thanks for attention! twitter: balzer82 url: http://balzer82.github.io/Kalman/ output: Presentation.html controls: true

--

IPython and Sympy to Develop a Kalman Filter for Multisensor Data Fusion

~ Paul Balzer ~

--

Setup

> http://tinyurl.com/PyData-Kalman <

or

git clone https://github.com/balzer82/PyData-Berlin-2014-Kalman

~

Package Versions used:

Sympy 0.7.4.1
Numpy 1.8.1
Pandas 0.14.1
Matplotlib 1.3.1
IPython 2.1.0
Seaborn 0.3.1

~

Presentation made with cleaver

--

Prolog

This talk will show you, how the IPython Notebook and Sympy helps to develop an optimal filter to fuse sensor information from different sources (e.g. IMU yawrate and GPS position) to get an optimal estimate.

--

Today's Goal

A car is driving in the city. Shading > bad GPS Signal quality > Position signal is "jumping". But position estimate could be improved by speed and turn rate measurements.

--

Topics

1. Sensor Noise
2. Filter idea in 1D
3. Multi-Dimensional Kalman Filter
4. Extended Kalman Filter

--

1. Sensor Noise

Believe it or not: No sensor on the whole world is providing the 'real value'. And even if, you do not know, if it is the real value.

sensor value = real value + some error

See Sensor-Noise.ipynb

2. Filter idea in 1D

Basic introduction to the theory behind the filter algorithm in just one dimension.

See Kalman-Filter-1D.ipynb

--

3. Multi-Dimensional Kalman Filter

Now we are going multi-dimensional with matrices.

See Kalman-Filter-CA-Ball.ipynb

--

4. Extended Kalman Filter

This is actually a real life example, which fuses GPS measurements and IMU measurements of a real vehicle.

See Extended-Kalman-Filter-CTRV.ipynb