Statistical Approaches for Target Counting in Sensor Networks Open Access
Downloadable ContentDownload PDF
Recent development in sensing technology enables low cost sensor networks to solve various environment monitoring problems, such as target enumeration, target localization and target tracking, etc.. In this thesis, we mainly focus on the target counting problem: estimating the number of targets within a monitored area using wireless sensor networks. Target counting is a fundamental problem in sensor-based environment monitoring problems and it has been widely investigated by a lot of researchers.However, there are still plenty of potential research to explore in target counting. First, as the emerging of the new sensing technology, a new sensing model, the numeric sensing model has been introduced. Most existing methods fail to work in the new sensing model. In addition, there is no universally ``best'' target counting method. Each method relies on certain pre-assumptions and has its own advantages and disadvantages. Target counting algorithms that can be applied to more general situations are needed.The objective of this thesis is to explore precise target counting methods that can be applied to more general situations. We mainly consider two different network models: the energy sensing model and the numeric sensing model.Specifically, we explore the identical target counting method that can be adapted to targets of different density levels. Popular target counting methods in energy sensor network may not be scalable to dense targets. For sparse targets, it is possible to count targets via counting sensors with local maximum readings. For dense targets, precise target count estimation relies on many factors, for instance, precisely recovering the energy landscape and precise estimation of energy leakage etc.. In this thesis, we describe a Monte-Carlo simulation based target counting algorithm which inexplicitly compensates errors in measuring energy volume from all targets due to undersampling and energy leakage. An estimator based on the simulation results is used as the estimated target count.We also consider identical target counting problems in numeric sensor networks. The main source of difficulty in this type of target counting problem is the overlapping sensing regions. Some targets may be counted by multiple sensors. In this thesis, we consider two different target models: targets with deterministic detectability and targets with stochastic detectability. For targets with deterministic detectability, we explore a target counting method that can be adapted to non-uniform distributed targets. In our method, distribution of targets' positions in the monitored area is estimated based on the sensors' readings. Then the estimated count of the total targets is obtained by the method of likelihood estimation based on a sequence of binomial distributions that are derived from a sampling procedure. For targets with stochastic detectability, we explore a target counting method that can be adapted to noisy measurements. We estimate the count of targets in each overlapped coverage topology via the correlations patterns of sensors' readings. A measurement noise resistant method is proposed to achieve this purpose. Total target count is obtained based on the estimated target count in each coverage topology via the inclusive-exclusive method.