Method of Updating Shadow Model for Shadow Detection based on Nonparametric Bayesian Estimation

Detecting shadows is needed for object detection methods because shadows often have a harmful effect on the result. Shadow detection methods based on shadow models are proposed. The shadow model should be updated to detect shadows which are not included in the learning data. In this paper, a new method for updating the shadow model for shadow detection is proposed. The proposed method models shadows by the Gaussian mixture distribution. The parameters of the distributions are estimated by the Dirichlet Process EM (DPEM) algorithm, which is a nonparametric Bayesian scheme. Shadows are detected by the probability density calculated with the shadow model. The detected result is improved by using the color segmentation method. The data which are not included in the learning data for the current shadow model are obtained by comparing the results. The DPEM algorithm is applied to the frequency distribution of the data to obtain initial parameters. After that, it is applied to the frequency distribution of the data which are obtained as shadow data in past frames and the new shadow model is constructed. Results are demonstrated by experiments using real video sequences.