Fast Segmentation and Modeling of Range Data via Steerable Pyramid and Superquadrics

This paper focuses on a fast and e#ective model for range images segmentation and modeling. The first phase is based on the well-known Simoncelli's steerable pyramid, useful to distinguish image information from noise. Gradient modulus and phase information is then exploited for achieving edges characterizing objects. Modeling is faced through superquadrics recovery. In this case a fast and simple procedure to estimate their free parameters is proposed. Achieved results on simple objects show that our model is simple, fast and robust to noise.