Background subtraction is an important first step for video analysis, where it is used to discover the objects of interest for further processing. Such an algorithm often consists of a background model and a regularisation scheme. The background model determines a per-pixel measure of if a pixel belongs to the background or the foreground, whilst the regularisation brings in information from adjacent pixels. A new method is presented that uses a Dirichlet process Gaussian mixture model to estimate a per-pixel background distribution, which is followed by probabilistic regularisation. Key advantages include inferring the per-pixel mode count, such that it accurately models dynamic backgrounds, and that it updates its model continuously in a principled way.