Study of Dynamical Processes with Tensor-Based Spatiotemporal Image Processing Techniques

Image sequence processing techniques are used to study exchange, growth, and transport processes and to tackle key questions in environmental physics and biology. These applications require high accuracy for the estimation of the motion field since the most interesting parameters of the dynamical processes studied are contained in first-order derivatives of the motion field or in dynamical changes of the moving objects. Therefore the performance and optimization of low-level motion estimators is discussed. A tensor method tuned with carefully optimized derivative filters yields reliable and dense displacement vector fields (DVF) with an accuracy of up to a few hundredth pixels/frame for real-world images. The accuracy of the tensor method is verified with computer-generated sequences and a calibrated image sequence. With the improvements in accuracy the motion estimation is now rather limited by imperfections in the CCD sensors, especially the spatial nonuniformity in the responsivity. With a simple two-point calibration, these effects can efficiently be suppressed. The application of the techniques to the analysis of plant growth, to ocean surface microturbulence in IR image sequences, and to sediment transport is demonstrated.