Begin () { Load dataset D Iterate over every image of the dataset for each image i in dataset D: { Extract HOG features from I, AppendI to features matrix F } Apply nonlinear Bayesian filter on matrix F for filtering. $P\left(x,t\right)\approx {\sum }_{k=1}^N{w}_t^{\left(k\right)}\delta (x-z_t^{\left(k\right)})$; Where × is a feature, Returns region Matrix. Annotate the regions from the region matrix. Calculate matrix feature O from region matrix $O\left(x,y,t\right)=\propto O\left(x,y,t-1\right)+D(x,y,t)$, t is time,$\propto$ is update rate $0<\propto <1$11.1 StoreO into a motion feature matrix If (motion in motion feature matrix >0) {message == Abnormal }else{message = = normal } } End ()