A novel statistical measure is proposed for robust object detection and segmentation, which is named 'Radial Reach Filter (RRF)'. The capability of detecting new objects from a time-series image is a important problem of vision systems such as video surveillance, multimedia processing. The basic method of detecting new objects is simple background subtraction that is to subtract current image from a background image. However, simple background subtraction is susceptible to illumination change such as shadows. And when the brightness difference between new objects and a background is small, simple background subtraction cannot detect the difference. We realize robustness under ill-conditioned environment by the "RRF pair-point" with a fixed brightness difference in the shape of radiation. A local texture is evaluated by evaluating brightness sign change defined in each pair. Thus, robust background subtraction was realized. Experiments with image sequences from the real world show the effectiveness of the proposed methods.
展开▼
