Statistical - Spatial technique for video object detection

Abstract

Object detection is the most important stage due to its great impact on next stages. Moving objects detections in real environment have several challenges such as dynamic background, illumination changes (gradual, sudden) and noise etc. Traditional techniques of background modeling do not have the ability to face these challenges but can deal with the static background. Based on statistical, spatial and temporal features we have proposed hybrid technique to model background and detecting moving object, it can deal efficiently with challenges in real environment. The proposed system consists of two stages, the first one is construction of statistical model by computing mean and standard deviation for each pixel and spatial model that are created by calculation Center Symmetric Local Binary Patterns which consists of a group of histograms for each pixel. The second one is foreground detection by checking each pixel in current frame with predefined thresholds if the value of pixel raises out of these thresholds then the statistical model will be applied otherwise the spatial model is applied. The contribution of this paper is creating model hybrid between statistical and spatial features which lead speed and accuracy cause statistical features are fast in calculation and spatial features characterized by accuracy. Experimental results proved that our proposed method lead to approximate 2% increase in the accuracy comparing with the traditional benchmark methods by using the exact dataset.