Census Transform Based Optical Flow for Motion Detection during Different Sinusoidal Brightness Variations

This work is a first approach of the implementation of a specific Optical Flow algorithm for motion detection in cases where the brightness variation is represented by a sine wave, whose characteristics vary across the different image sectors. The final goal would be the implementation of such an algorithm in thermal films recording a component undergoing a sinusoidal load. Mapping the motion field all over the thermal video time history, and deriving it in order to obtain a strain map, would enable both the simultaneous measurement of stress (by performing Thermoelastic Stress Analysis) and strain by a unique video, and the possibility of stress calibration, thus linking the digital levels in output to real stress values. In this scenario, the authors present the implementation of Census Transform based Optical Flow on a simulated video, where the brightness variation is modeled like the superimposition of sine waves equal for all pixels and different Gaussian spatial distributions frame after frame. The latter is used for creating different sine patterns for each image sector.