Adaptive shadow detection based on GMM and MRF

作者：Min, Hua-Qing^*; Lü, Ju-Mei; Luo, Rong-Hua; Chen, Cong

来源：Journal of South China University of Technology(Natural Science Edition), 2011, 39(7): 115-120.

DOI：10.3969/j.issn.1000-565X.2011.07.019

• Adaptive thresholds
• Detection accuracy
• Gaussian Mixture Model
• Markov Random Fields
• Misclassifications
• Parameter learning
• Shadow detections
• Spatial dependence

Shadows of moving objects often reduce the accuracy and the effectiveness of tracking and recognition of moving objects. In order to distinguish the objects from their shadows, an adaptive shadow detection method is proposed based on the Gaussian mixture model (GMM) and the Markov random field (MRF). In this method, first, an improved GMM, which adaptively adjusts the parameter learning ratio, is proposed to remove the light shadow of an object. Then, a new approach to shadow detection is put forward based on the spatial dependence information about the integrated neighborhood of MRF. Moreover, in order to improve the MRF-based shadow detection accuracy and effectiveness, the information capacity is used to select color features, and the results of a coarse shadow detection obtained from the adaptive threshold-based segmentation are employed to initialize the parameters of MRF. Experimental results indicate that the proposed approach effectively avoids the misclassification existing in the shadow detection and improves the detection accuracy.

• 出版日期2011-7
• 单位华南理工大学