Automated object detection and tracking based on clustered sensor networks

Automated visual surveillance is becoming an increasingly interesting topic for the scientific community because of changing security needs. The need for developing computer systems to provide enough information to take rapid and preventive action against security threats is great. Recent advances in VLSI design and wireless communication make possible the development of low power and small size sensor nodes that are capable of communicating with each other, without supervision from a base station, over short distances. Such developments pave the way for the design and realization of a visual surveillance system based on sensor networks.; An architecture based on a visual sensor network that performs automated object detection and tracking is presented. A processing unit attached to each camera detects objects moving in the video stream. Then the detected object is sent to a multi-agent tracking system. The information is sent to higher tier processing units for further processing. Object detection is by background subtraction.; Traditionally, background subtraction techniques have been implemented in software. However, software implementation is expensive in terms of power; therefore, it is not a good option for sensor network nodes. An efficient alternative for sensor nodes is to implement the Wronskian Change Detector in hardware as discussed in this work.; Visual information can be complemented by infrared images for concealed weapon detection. Visual information provides identity data while IR images supply localization of the concealed weapon. Image fusion incorporates the complementary information from two sources to improve detection and recognition. An image fusion scheme based on a wavelet transform is proposed.; In certain applications, it may be necessary to incorporate non-visual sensors, such as chemical sensors, in addition to visual sensors. Such non-visual sensors can be viewed as simpler nodes in the hierarchical proposed system of nodes. These sensors are organized in a two-tier network where resource and power-restricted motes form the lower tier. The upper tier is formed by stargate-class nodes designated as masters. A master selects four motes to be sentinels in order to detect an event of interest. The proposed data fusion scheme is a linear mean square estimator based on correlation coefficients.