We proposed a Frenet-Serret descriptor to classify stress categories based on color dynamics of pixels stored in photoelasticity videos. A collection of image compression models of disc and ring with a monotonic incremental load was generated. For each pixel, a temporal curve was created using color changes each frame. A descriptor histogram with Frenet Serret parameters was used to train a neuronal network; it classified in four types of stress zones (concentrated, high, medium, low). With the proposed method, a dynamic differentiation was possible in the field of stress without considering traditional digital photoelastic procedures.
|