Traditional surveillance systems are constrained because of a fixed and preset pattern of monitoring. It can reduce the reliability of the system and cause an increased generation of false alarms. It results in an increased processing activity of the system, which causes an augmented consumption of system resources and power. Within this framework, a human surveillance system is proposed based on the event-driven awakening and self-organization principle. The proposed system overcomes these downsides up to a certain level. It is achieved by intelligently merging an assembly of sensors with two cameras, actuators, a lighting module and cost-effective embedded processors. With the exception of low-power event detectors, all other system modules remain in the sleep mode. These modules are activated only upon detection of an event and as a function of the sensing environment condition. It reduces power consumption and processing activity of the proposed system. An effective combination of a sensor assembly and a robust classifier suppresses generation of false alarms and improves system reliability. An experimental setup is realized in order to verify the functionality of the proposed system. Results confirm proper functionality of the implemented system. A 62.3-fold system memory utilization and bandwidth consumption reduction compared to traditional counterparts is achieved, i.e. a result of the proposed system self-organization and event-driven awakening features. It confirms that the proposed system outperforms its classical counterparts in terms of processing activity, power consumption and usage of resources