Kalman Filter: An Introduction
Kalman Filter is an optimal estimation algorithm which is used in various disciplines today, right from tracking position & velocity of a vehicle, to keeping track of the spacecraft’s internal combustion chamber’s temperature, to estimating the exchange rate of a particular currency. Rudolph E. Kalman developed it in 1960 & one of its variants was used in the trajectory estimation around the moon in the Apollo program by NASA.
The reason that makes Kalman filter so noteworthy is that it can take noisy measurements as input & provide very accurate measurement of the true value quickly. It does not wait for a lot of data to be generated to calculate an accurate estimate, instead, it keeps on updating its estimate simultaneously as the data is being generated.
In a nutshell, Kalman filter consists of two main stages state prediction & measurement update. We keep on iterating over these two steps & eventually, the estimate converges to a real value. It is even capable of processing measurements which are generated at irregular time intervals.
Why do we need Kalman Filter?
Kalman filter takes input from one or more sensors & tries to provide an accurate estimate which is close to value of the target. The real world itself is very random & some external factors which may not be considered by our sensors can affect our measurements. Often sensory data also exhibit imperfect measurements.
Kalman filter is simply a mathematical model that can develop a surprisingly accurate estimate of the true value being measured, by taking into consideration measurements of different values. Of course, there are other algorithms that can be used instead of Kalman filter to get a good estimate of the value being measured, but what makes Kalman filter exceptional is, with only a few measurements it can provide a fairly good estimate.
What are some real-world applications of Kalman Filter?
- Navigation system’s
- Vision‐based real‐time vehicle tracking system.
- 3D modelling
- Speech Enhancement
- Radar tracker
- Weather forecasting
We dive into nitty-gritty details of some variants of Kalman Filter in further articles whose links are mentioned below
Common Types of Kalman Filters:
- Standard Kalman Filter (KF) — linear (output proportional to the input)
- Extended Kalman Filter (EKF) — nonlinear (used in robotics)
- Unscented Kalman Filter (UKF) — highly nonlinear (used where EKF fails to converge)
