Assessing separation of duty policies through the interpretation of sampled video sequences: A pair programming case study

In this paper we present a non-invasive technique, which can be used to interpret single camera undersampled videos in order to extract the alteration patterns of Pair Programming (PP) developers. The method uses a procedure for scene interpretation exploiting 3D face models to take into account for movement related illumination change, facial expression change and occlusion; in order to extract the PP relevant information, the scenes are sampled and then their interpretation are connected to one another also based on domain knowledge. The overall video interpretation performed in this way is robust to high frame-misclassification rates. Since the actor's identities are not relevant by themselves but only the alternating times are important, the method can be used in lower quality videos, e.g. where quality has been purposely degraded to protect privacy.