Most surveillance camera systems are still controlled and monitored by humans. Smart surveillance camera systems are proposed to automatically understand the scene captured, identify the objects of interest, detect the abnormality, etc. However, most surveillance cameras are either wide-angle or pan-tilt-zoom (PTZ). When the cameras are in the wide-view mode, small objects can be hard to be recognized. On the other hand, when the cameras are zoomed-in to the object of interest, the global view cannot be covered and important events outside the zoomed view will be missed.
In this paper, we proposed a system composed of a wide-angle camera and a PTZ camera. The system is able to capture the wide-view and the zoomed-view at the same time, taking the advantages from both views. A real-time human detection and identification algorithm based on a neural network is developed. The system can efficiently and effectively recognize humans, distinguish different identities, and follow the person of interest using the PTZ camera. A multi-target multi-camera (MTMC) system is developed based on the original system. In the MTMC system, multiple cameras are placed at different places to look at different views. The same person shown in any camera can be recognized as the same person while different persons can be distinguished among all the cameras.