This paper is concerned with the problem of distributed extended object tracking, which aims to collaboratively estimate the state and extension of an object by a network of nodes. In traditional tracking applications, most approaches consider an object as a point source of measurements due to limited sensor resolution capabilities. Recently, some studies consider the extended objects, which are spatially structured, i.e., multiple resolution cells are occupied by an object. In this setting, multiple measurements are generated by each object per time step. In this paper, we present a Bayesian model for extended object tracking problem in a sensor network. In this model, the object extension is represented by a symmetric positive definite random matrix, and we assume that the measurement noise exists but is unknown. Using this Bayesian model, we first propose a novel centralized algorithm for extended object tracking based on variational Bayesian methods. Then, we extend it to the distributed scenario based on the alternating direction method of multipliers (ADMM) technique. The proposed algorithms can simultaneously estimate the extended object state (the kinematic state and extension) and the measurement noise covariance. Simulations on both extended object tracking and group target tracking are given to verify the effectiveness of the proposed model and algorithms.