Feature modeling and tracking in a time-resolved sequence of combustion images

Abstract

Computer-assisted flow-field analysis and interpretation require appropriate data models and digital image processing tools. In recent work1 we used frame-to-frame correlation to compute the flow of fluid boundary features in a sequence of laser shadowgraph images. The generality of our approach was limited by the use of a block representation for the shadowgraph structure. A new shape model for spatially resolved images of turbulent flows affords a more powerful vehicle for such tasks as tracking and classifying flow-field features. Each frame in a sequence of combustion images is initially reduced to a skeleton, which preserves the important structural information concerning fluid boundaries etc. Skeletons are then represented by a model consisting of nodes and branches. A node is defined as the junction where two or more skeletal lines (branches) meet. Several options for modeling branches are investigated including polynomials and Fourier descriptors. The node-branch representation of each image is used for frame-to-frame feature tracking. Certain branch models are also appropriate for flow classification.