Real-time detection of moving objects in a video sequence by using data fusion algorithm

Abstract

The moving object detection and tracking technology has been widely deployed in visual surveillance for security, which is, however, an extremely challenge to achieve real-time performance owing to environmental noise, background complexity and illumination variation. This paper proposes a novel data fusion approach to attack this problem, which combines an entropy-based Canny (EC) operator with the local and global optical flow (LGOF) method, namely EC-LGOF. Its operation contains four steps. The EC operator firstly computes the contour of moving objects in a video sequence, and the LGOF method then establishes the motion vector field. Thirdly, the minimum error threshold selection (METS) method is employed to distinguish the moving object from the background. Finally, edge information fuses temporal information concerning the optic flow to label the moving objects. Experiments are conducted and the results are given to show the feasibility and effectiveness of the proposed method.

Contour Detection of Multiple Moving Objects in Unconstrained Scenes using Optical Strain

“…The Jaccard index represents how well the predicted foreground aligns to the groundtruth. It is defined as: (e) Result from [14].…”

“…We compare our model with the methods proposed by Patel and Parmar [13] and Tang et al [14], which also use the optical flow as input to obtain the moving objects of the scene. Both methods work on the spatial domain, taking into account only one frame each time.…”

“…2) Tang et al [14]: A model that works with both, the image and the optical flow magnitude, is presented. It is proposed to compute the edges over the images using an edge detector method.…”

“…All authors acknowledge the Auvergne-Rhône-Alpes region for the funding of this work. (e) Result from [14]. P = 0.78, R = 0.07, F 1 = 0.14 (f) Result from [13].…”

“…Let us note that, under the assumption that the boundaries are well detected and complete, both methods are complementary as it is straightforward to obtain the object shape from it boundaries and vice-versa. The main drawback of these methods is their sensibility to noise, which is dissipated when the models are contour-based [14].…”