Detection And Segmentation Of Moving Objects And Their Shadows

In computer vision field,accurate detection of moving objects is indispensable in surveillance systems and intelligent transportation systems.Since cast shadows share the same movement patterns and magnitude of intensity change with the objects of interest,moving cast shadows along with moving objects induce many problems for moving object detection.Many important research and applications can get more robustness of the results with accurate segmentation of moving objects and their cast shadows,such as action analysis,object recognition,and object tracking.Thus,detection and segmentation of moving objects and their shadows are important research topics in computer vision field.When moving objects have similar chromaticity and intensity to their shadows or when they are immersed in the shadows of other moving objects,these cases make the accurate segmentation of moving objects with their shadows very difficult.Thus,it is an extremely challenging issue in identifying moving objects and their shadows.The main work of this paper includes:For the moving objects or their shadows having similar color feature to the background and background in motion,a method is proposed to detect moving regions by combination of multiple features and two background models.Firstly,the statistical model is built by learning the temporal record of color,texture and movement information computed from optical flow,then background model is constructed based on the assumption that background is subject to temporal consistency and background values will converge to ones of the first few states of the aforementioned statistical model.Finally,moving regions are extracted via the match score map between the estimated expected values of background and the observed values.Experiments were carried out based on captured videos and public datasets,the effectiveness of the method is evaluated by using the factors of recall and precision or comparison with ground truth.The results showed that the method combination of multiple features with two background models has a robust performance under moving objects with camouflages and background in motion.To accurately segment moving objects and their shadows,especially when the moving objects have similar chromaticity and intensity to their shadows or when they are immersed in the shadows of other moving objects,a two-stage method is proposed to obtain accurate regions of moving objects and their shadows using brightness ratios and movement patterns.The detected motion regions are firstly independently segmented based on local color constancy in brightness ratio(BR)space,then potential moving shadows are separated via Gaussian distribution of brightness ratios,and Gaussian distribution of brightness ratios is updated from the detected shadow pixels in the each frame.In accordance with the fact that moving objects share their movement patterns with their shadows and the relatively unchanged position relationship,moving shadows are improved via scale invariant feature transform(SIFT)flow patterns.Experiments with captured image sequences and public videos,and comparison with ground truth and several advanced methods verified the method's efficiency quantitatively and qualitatively.The results demonstrated that the method can accurately segment moving objects and their shadows,especially when the moving objects have similar chromaticity and intensity to their shadows or when they are immersed in the shadows of other moving objects.