Moving objects detection exploiting spatial and temporal cues in dynamic scenes
Xianglong Tang · Journal of Natural Science of Heilongjiang University · 2009
Gaussian mixture models (GMM) based approaches to moving objects detection are widely used and achieve considerable success. Their performances,however,will significantly degrade in dynamic scenes due to dynamic background such as waving trees,rippling water,camera jitters,etc. The main reason is that they neglect spatial correlation among pixels,which is very critical for dealing with dynamic background. A novel moving objects detection algorithm is proposed,which effectively handles the dynamic background by exploiting spatial and temporal cues simultaneously. Specially,for each pixel,considering its temporal variation and spatial variation. The former is modeled by Gaussian mixture models. The latter is exploited to label the pixel as background if one of its neighboring pixels matches to its background model. Experimental results show that the proposed method significantly outperforms the traditional GMM method in dynamic scenes.