| As a mean of access to information from the real physical world, Wireless Sensor Networks has been paid attention to by more and more people, and event detection is one of the important applications of Wireless Sensor Networks. Considering the characteristics of Wireless Sensor Networks, that energy consumption is limited and system is susceptible to environment influence, by a kind of efficient data fusion pattern, the event detection algorithm could reduce the data stream, by a method of event detection having good fault tolerance; the event detection algorithm would ensure the accuracy of event detection. To distinguish from the geographical location of event detection, event detection algorithm can be divided into two kinds of event-region detection and event boundary detection. In the event area, the local sensor node and its neighboring nodes have the similar observations, meanwhile, for the event boundary sensor node, it and its neighboring nodes outside the boundary sensor node have the different observations and neighboring nodes inside and outside the boundary sensor node have the significantly different observations. Taking into account the different characteristics of the events region nodes and the event boundary nodes, this paper proposed an event region detection algorithm and an event boundary detection algorithm.This paper presents a distributed event-region detection scheme for wireless sensor networks. Under the assumption that the event-region detection process follows a stationary distribution, the proposed scheme employs a kind of spatial interpolation, that is Indicator Kriging (IK), to reduce the impact of fault nodes. The approach was composed of two major tasks:(1) an indicator transformation from the raw data to the binary data;(2) the solution of conditional probability that considers the spatiotemporal correlation between the local node and its neighboring nodes. Through these, we can obtain the probability of event to determine whether the event happens. By numerical simulations, we have shown that the proposed IK scheme can result in a good detection performance. The paper also raises an event boundary detection algorithm, the two-dimensional spatial location of the sensor node and sensing data as the spatial coordinates of the corresponding three-dimensional point, so the original model changes to the three-dimensional graphics, because there are different about the sensing data between the event boundary nodes and their neighboring nodes, among their neighboring nodes, the difference is equivalent to the surface change on3D graphics. Because of the dense node, the sensor node readings inside the event boundary is similar, and the corresponding three-dimensional points can form a smooth surface, meanwhile, in the area near the boundary nodes, the corresponding three-dimensional graph represents the difference changes. Taking into account the characteristics of this algorithm, this algorithm estimates the curvature of the corresponding three-dimensional point of the sensor node in the approximate surface by the mean of surface fitting, and from the characteristics of curvature we can determine the neighboring structure of this three-dimensional node, so characteristics of neighboring structures can help us to determine out for the event border node or not.In order to verify the performance of the proposed event detection algorithm, we use MATLAB to complete the data simulation. First, we build the background of the event detection, i.e. a large number of sensor nodes are deployed randomly in a set area, these sensor nodes have a certain failure rate, which set the percentage of faulty nodes can test the fault tolerance of the algorithm. And then event detection algorithm starts running in the predetermined background, after the detection, comparing the detection result with the true situation, we can find the event detection accuracy and false positive rate, which confirms the higher performance of the proposed detection algorithms. |
