Design aspects and software modelling for ubiquitous real-time camera system are described in this paper. We propose system architecture using a network of inexpensive cameras and perform video processing in-network. In general, ubiquitous systems have to perform spatial and temporal calibration in advance to determine timing and coordination relationship between sensor nodes, and other application specific design considerations, such as system architecture, distributed software, control authority and communication channel. A methodology for transforming welldesigned single-node algorithm to distributed system is also proposed. Applications for ubiquitous cameras can be modelled as the composition of system finite state machine, functional services and middleware. A service oriented software architecture is proposed to dynamically reconfigure services when system state changes. We have developed a distributed gesture recognition system with true in-network processing to analyze video in real time. By exchanging data and control messages between neighboring sensors, each node can maintain broader view of the environment with integrated video processing results. Our prototype system is built on Windows machines, and uses webcams as sensors and local network as communication channel.