A Bayesian Filtering Approach to Object Tracking and Shape Recovery from Tomographic Measurement Data

Dynamic tomography of non-stationary objects from sparse tomographic data is a challenging issue arising in various areas where heterogeneous flow structures comprised of gas-solid and gas-liquid flow phases occur. Such flow patterns are quite common in pneumatic conveying, oil, food, and chemical industry. This paper addresses the reconstruction of a time-varying material distribution inside a closed pipe subject to dynamics of the object position and the object boundary in an electrical capacitance tomography (ECT) system. The main objective is to recover the shape and the current position of a moving material inclusion. The unknown quantities - shape and position - are jointly estimated applying a Bayesian recursive filter approach incorporating new measurements in each measurement update step. Consequently, the inverse ECT problem is recast as a state estimation problem. The object boundaries are modeled using a Fourier descriptor based contour model of second order which is able to cover various physically reasonable shapes. By prescribing the order of the contour model prior information is incorporated which can be interpreted as regularization as high frequency parts in the estimated contour are low-pass filtered. The proposed tomographic signal processing algorithm is validated experimentally using different test objects