In [14], a method for the time synchronization of a multiple-camera system is proposed without using an external clock signal. The basic idea is to use co-occurrence of appearance changes of objects in motion that are observed on different views. Specifically, the spatial integral over the image plane of temporal derivatives of brightness is used as a temporal feature of a video sequence. Although a great amount of efforts have been devoted to the image-based synchronization technique, they are not universal and may not be applicable in the real world applications due to the innate limitations, such as prerequisite LED auxiliary, arbitrarily tilting or stationary cameras, specific texture of background, or restrictive motion of objects.

Actually, camera synchronization with external clocks or triggers is still needed in the practical viewpoint. Generally, there are three categories of state-of-the-art techniques. The first is to use dedicated wires to transfer the reference signal. Many of the industrial vision sensors are equipped with dedicated electrical inputs/outputs to synchronize trigger signals, in which one of the vision sensors��or a dedicated signal emitter device��acts as a master, and the others are operated in synchronization with the trigger signal emitted from the master. A major problem in this classical and widely-used means is that deployment of synchronization wires is cumbersome in some situations��short wires may impose constraints on spatial configuration of vision sensors; long wires may cause unstable synchronization.

The second solution is to use wired standard bus such as IEEE1394 and Ethernet. Instead of dedicated synchronization wires, some systems allow synchronization through standard electronic buses used for image transfer such as IEEE 1394 [24] and Ethernet [25,26]. These systems bring higher flexibility, but they still require wired connections and are unsuitable for wireless vision sensor networks. The third type is to employ wireless communication protocols for synchronization in sensor network field. The principal difficulty in time synchronization of wireless network systems lies in nondeterminism in wireless media access time [27]. Due to this nondeterminism, it is difficult to make certain when a synchronization packet started to propagate from the sender.

RBS [28] introduced a receiver-receiver synchronization scheme to remove the effect of the sender nondeterminism, but Cilengitide requires many message exchanges between receivers to achieve high precision. TPSN [29] and FTSP [30] suppress this nondeterminism by time stamping at the media access control (MAC) layer, but they inherently require special MAC implementations. It is also possible to equip a dedicated receiver of radio or optical reference synchronization signal, but at the cost of additional equipments.