On the timeliness of a cluster based large scale online video surveillance