The \(k\)-Nearest Neighbors (\(k\)NN) algorithm is the machine learning-core of the proposed algorithm. For a given histogram that is derived from the current camera image, the \(k\)-Nearest Neighbors (\(k\)NN) algorithm measures the similarity of this histogram to all histograms in the training dataset and outputs the \(k\) most similar training histograms and the corresponding \(x,y\)-positions.

While the \(k\)NN algorithm is one of the simplest machine learning algorithms, it offers several advantages: it is non-parametric, allowing for the modeling of arbitrary distributions. Its capability to output multiple predictions allows for neat integration with the proposed particle filter. Its simplicity comes together with transparency: it allows for spotting the possible sources of error such as wrongly labeled training examples.