Psc Pel Per Ps Phl Phr Pkl Pkr (a) (b) (c) Figure 3 (a) The feature vectors extracted from the 3D camera [6]. (b) The Graphical User Interface (GUI) of the system summarizing different users' actions [6]. (c) The system ability to search for a certain event within a certain time stamp and play it back [6]. "tracking" to obtain a bounding box and trajectory of the target which can allow basic event detection such as a target that is entering a private area, etc. Nevertheless, this does not allow the recognition of the human's behavioral features (e.g., distinguishing whether the surveillance target is opening her/ his car door or trying to steal it). Hence, the available surveillance systems require humans to closely monitor the cameras output. For unattended cameras, when an event of interest happens, humans have to watch the whole video sequence to identify the event in question. Thus, in order to allow AmI to be realized in domestic and open spaces (where privacy is not a limitation), there is a need to develop intelligent vision systems. Such systems should be capable of detecting complex FEBRUARY 2015 | IEEE CompUtAtIonAl IntEllIgEnCE mAgAzInE 47