Distributed Acoustic Source Localization And Tracking Algorithm

Sound Localization and Tracking technology has been wide applied in many aspects, including Video Conference, Human-Computer Interaction, Security Surveillance, Shooter Localization and so on. Wireless Sensor Networks is quite suitable for the application of Sound Localization and Tracking for its small-size, low-power cost, and flexibility for deployment. When applied to Wireless Sensor Networks, centralized Sound Localization and Tracking algorithm has many problems, including high computing complexity in the center node, non-balanced communication between sensor nodes, and low system reliability. Distributed algorithm can avoid the problems above effectively. Existed distributed algorithms also encounter problems. It can only be applied to specific network topology, and can not compute in parallel. It is prone to trapping in local optima, and information between sensor nodes can not be fused adequately. As to above problems, this thesis researches three aspects respectively, including single sound source localization, multiple sound sources localization and single sound source tracking, and proposes distributed algorithms respectively. The Alternating Direction Method of Multipliers (ADMM) Based Distributed Sound Localization Algorithm, decomposes the Maximum Likelihood problem into each sensor node and uses Bridged Sensor Nodes and Consensus Algorithm to implement Information Fusion. The optimization target function of ADMM is non-convex, and the algorithm is prone to trapping in local optima. Multi-Resolution Search method is proposed to solve the problem. Compared to existed algorithms, the proposed algorithm can be implemented in parallel, can be applied to arbitrary network topologies, and also can avoid local optima effectively. EM Algorithm Based Distributed Multiple Sound Sources Localization Algorithm, decomposes multiple sound sources into single sound source localization problem in parallel using EM Algorithm. It applies the single sound source distributed algorithm to implement multiple sound sources distributed algorithm directly. The performance of the algorithm is close to the Centralized Multiple Sound Sources Localization Algorithm. Every sensor node can compute the location of every single sound source in parallel, and the speed of computation is faster. It also can be applied to arbitrary network topologies. Gaussian Mixture Model (GMM) Based Distributed Particle Filter Sound Tracking Algorithm, approximates the posterior probability density as a Gaussian Mixture Model, and implements Information Fusion between nodes with Consensus Algorithm. Every sensor node can obtain global posterior probability density, and only need to transfer the parameters of GMM with its neighboring nodes. Compared to existed algorithms, the algorithm reduces the transferring data between nodes, and can be applied to arbitrary network topologies. Every sensor node can compute in parallel, and the performance of the algorithm is better.