The filter toolkit is a set of blueprints which can used to process signals. It allows us to get smoother values from a signal, and thus allows us to easily process this signal.
An example of raw signals is what we can get from the accelerometer or from the magnetometer. Using a filter allow us to have more stable values from these inputs. However, it’s not limited to this kind of use case: everytime you have a raw or unstable behavior, and you want to smooth it, then the Filter Toolkit can be used.
Note: A basic knowledge of signal processing is advised.
In the Filter Toolkit, all filters are designed as Actor Components. To start using them, you just need to add the desired filter as an Actor Component of the Actor whose behavior requires filtering.
Once it’s done, the parameters of the filter can be adjusted in the Details panel (when the component is selected).
Now, you’re ready to use the filter: just call the GetValue method of the filter, providing your signal as input, and get the filtered output.
More details about the different filters and the way they work is given below.
Principle of the filters
All the filters in the Filter Toolkit are ActorComponent. They must be added to an actor before being used. Every filter proposes a set of parameters.
The only function of all filters is GetValue which takes an input (usually a float from a signal) and returns a filtered output.
The state of the filter is changing after each input, so you can’t expect the same result by calling GetValue two successive times.
If you are processing a temporal signal, the period between two successive calls of GetValue should be constant (the sampling of the input signal is supposed to be regular).
Low-pass threshold filter
If the difference between the current input and the last output is above a threshold (which is a parameter of the filter).
High-pass threshold filter
If the difference between the current input and the last output is below a threshold (which is a parameter of the filter).
The output is the difference between the current input and the previous input.
The output is the mean of the last N inputs where N is the size of the filter (the only parameter).
A mean filter for vectors is provided: it applies a mean filter on each component of the input vector.
The output is the median of the last N inputs where N is the size of the filter (the only parameter).
It works like the mean filter, except that all previous inputs are weighted according to the values in an array (the kernel).
Technically we have:
Dynamic kernel filter
It works like the kernel filter, excepts that the GetValue takes two parameters: the input and the weight of this input. The jernel is then dynamically updated using the provided weights.
It is an infinite impulse response filter, according to this definition. The two parameters of this filter are the input kernel (the weights applied to the input) and the output kernel (the weights applied to the output).
The Filter Toolkit also contains a PID controller component. The purpose of the PID controller is to estimate the good input to apply to a system in order to have this system produce a given output. This component takes 4 parameters:
Target: the desired output
Kp: the proportional gain applied to the error (the difference between the target and the input)
Ki: the integration gain applied to the error
Kd: the derivative gain applied to the error