Abstract：By using conventional Gaussian mixture model, moving object was always detected incorrectly for the situation of dynamic variation of background image and scene, because only pixel level and time domain were classified regardless of spatial information.Based on spatial neighborhood weighted Gaussian mixture model, a moving vehicle detection method was proposed. According to spatial feature of pixel, a spatial information function was defined to restrain noisy. The neighbor information weighted class probabilities of very pixels with spatial constraint were designed and proved to be meet with two criterions of normalization and spatial continuity.Regarding space and time information, the iterative renovated parameter formula and the moving detection algorithm were proposed.The experiments of moving vehicle detection for urban traffic video sequences under different climate demonstrate that the proposed method can get better classification effeciency and accuracy with low misjudgement rate.
2011, Vol. 32 
