Developing universal methods for background subtraction and object tracking is one of the critical and hardest challenges in many video processing and computer-vision applications. To achieve superior foreground detection quality across unconstrained scenarios, a novel Two Layer Rotational Symmetry Dynamic Texture (RSDT) model is proposed, which avoids illumination variations by using two layers of spatio temporal patches. Spatio temporal patches describe both motion and appearance parameters in a video sequence. The concept of key frame is used to avoid redundant samples. Auto Regressive Integrated Moving Average model (ARIMA) (Hyndman & Rob, 2015) estimates the statistical parameters from the subspace. Uniform Local Derivative Pattern (LDP) (Zhang et al., 2010) acts as a feature for tracking objects in a video. Extensive experimental evaluations on a wide range of benchmark datasets validate the efficiency of RSDT compared to Center Symmetric Spatio Temporal Local Ternary Pattern (CS-STLTP) (Lin et al., 2015) for unconstrained video analytics.
Dynamic Texture Recognition Using Time-Causal Spatio-Temporal Scale-Space Filters
