Support vector learning approaches for object localization in acoustic wireless sensor networks

Object tracking, whose goal is to estimate the location of a target of interests, is one of the key issues in applications of wireless sensor networks (WSNs). Recently, various target tracking methods were proposed, especially using learning techniques such as neural network and support vector machine (SVM). This paper presents two SVM-based learning approaches for target tracking using WSNs. In the first approach, a black-box relationship between the acoustic measurements and the location of object is learned using least-square support vector regression (LSSVR). The other approach is multi-class classification with cell decomposition, which employ posterior probability regression with Platt's method to learn the sensor model. We describe both approaches and evaluate their performance in terms of the accuracy and robustness. Experimental results show that the direct regression approach is more accurate and robust to the sensing noise than the posterior probability regression approach. The localized aspects of the posterior regression can be advantageous in terms of scalability.