Abstract—A method is proposed to automatically construct a tempo-topographical model of a wireless IP-multi-camera network using some observations of known moving targets. The method is entirely unsupervised with no calibration of the cameras. This enables to extend the activity analysis across the entire camera network, constructing transition model between non-overlapping camera views, and also enabling the tracking across the blind-regions of the network. Based only on camera spatial and temporal information, this approach eliminates redundancy in the paths linking any pair of cameras by using the weighted graph technique to each node. The experimental results on real-time video streams show the feasibility of our system and its effectiveness in human activity tracking.

Index Terms—Abnormal behavior, real-time detection, multi-camera tracking, video surveillance.

K. Al-Shalfan is with the Computer Science Department, Imam Med Bin Saud University, Riyadh. Kingdom of Saudi Arabia (email: Kshalfan@gmail.com).
M. Elarbi-Boudihir is with the Applied Mathematics Department, Imam Med Bin Saud University, Riyadh. Kingdom of Saudi (e-mail: elarbi-boudihir@lycos.com).