Algorithm 1 (SOM):

Initialization: set the initial time step to $n=0$ and choose small random values for the initial synaptic-weight vectors ${\mathit{w}}_j(0)$, $j\in \mathcal{L}$. Sampling: set $n\leftarrow n+1$ and sample the training input pattern ${\mathit{x}}_n$, i.e., the feature vector from the n-th parking sensor. Similarity matching: find the winning neuron, whose index is $j^{☆}(n)$ at time step n by using the minimum-distance criterion: $$j^{☆}(n)=arg\underset{j}{min}\parallel {\mathit{x}}_n-{\mathit{w}}_j(n)\parallel ,j\in \mathcal{L}$$ (3) where with $\parallel \mathit{a}-\mathit{b}\parallel$, we mean the Euclidean distance between vectors $\mathit{a}$ and $\mathit{b}$. Update: adjust the synaptic-weight vectors of all neurons by using the update equation: $${\mathit{w}}_j(n+1)={\mathit{w}}_j(n)+\eta (n)h_{ji}(n)({\mathit{x}}_n-{\mathit{w}}_j(n)),$$ (4) with $j\in \mathcal{L}$, where $\eta (n)$ is the learning rate parameter at iteration n and $h_{ji}(n)$ is the neighborhood function centered on $j^{☆}(n)$ at iteration n. Adjust neighborhood: adjust the neighborhood size ($h_{ji}(n)$), the learning rate ($\eta (n)$) and continue from Step 2 if $n<n_{\mathrm{iter}}$; stop otherwise.