Motion Detection Based on Directional Rectangular Pattern and Adaptive Threshold Propagation in the Complex Background

This paper presents a Directional Rectangular Pattern (DRP) based complex background modeling method to detect the moving objects in a video sequence. Different from Local Binary Pattern (LBP) encoding the binary result of first-order derivative between the central point and its neighborhoods, Directional Rectangular Pattern is proposed to encode the binary result of first and second order derivative direction in all neighborhoods among a rectangular region. To model the distribution of the DRP micro-patterns, the DRP integral histograms are used to extract the discriminative features to represent the input videos. The local gray-level feature based Gaussian Mixture Model (GMM) is exploited to calculate an adaptive threshold for the histogram similarity measure to decide which part/pixel is background or moving object. Experimental results on two public videos are used to testify the effectiveness of the proposed method by comparing with LBP, GMM based background modeling methods.