Adaptive Dictionary-Based Spatio-temporal Flow Estimation for Echo PIV

We present a novel approach to detect the trajectories of particles by combining (a) adaptive dictionaries that model physically consistent spatio-temporal events, and (b) convex programming for sparse matching and trajectory detection in image sequence data. The mutual parametrization of these two components are mathematically designed so as to achieve provable convergence of the overall scheme to a fixed point. While this work is motivated by the task of estimating instantaneous vessel blood flow velocity using ultrasound image velocimetry, our contribution from the optimization point of view may be of interest also to related pattern and image analysis tasks in different application fields.