Advances in the field of computer vision enable smart cameras to cooperatively analyse scenes without human intervention. Large networks of autonomous, self-organising PTZ (pan, tilt, zoom) cameras require algorithms and protocols that make way for cooperation between multiple smart cameras (SCs). This paper introduces a distributed algorithm for object tracking with multiple SCs (DMCtrac). The focus lies on PTZ management issues arising in large SC systems rather than on computer vision algorithms. DMCtrac enables SCs to observe objects throughout an area under surveillance by using their PTZ abilities to follow these objects. The algorithm has been evaluated by simulation of SC systems with up to 50 SCs as used for example for people tracking and traffic analysis. Results show, that DMCtrac is able to cope with large numbers of SCs and objects and is robust towards real world disturbances like communication failure.