A Gaussian Mixture Model with Gaussian Weight Learning Rate and Foreground Detection using Neighbourhood Correlation

Moving object detection is the first and foremost step in many computer vision applications such as automated visual surveillance, human-machine interface, tracking, traffic surveillance, etc. Background subtraction is widely used for classifying image pixels into either foreground or background in presence of stationary cameras. A Gaussian Mixture Model (GMM) model is one such popular method used for background subtraction due to a good compromise between robustness to various practical environments and real-time constraints. In this paper we assume background pixel follows Gaussian distribution spatially as well as temporally. The proposed research uses Gaussian weight learning rate over a neighbourhood to update the parameters of GMM. The background pixel can be dynamic especially in outdoor environment, so in this paper we have exploited neighborhood correlation of pixels in foreground detection. We compare our method with other state-of-the-art modeling techniques and report experimental results. The performance of the proposed algorithm is evaluated using both qualitative and quantitative measures. Quantitative accuracy measurement is obtained from PCC. Experimental results are demonstrated on publicly available videos sequences containing complex dynamic backgrounds. The proposed method is quiet effective enough to provide accurate silhouette of the moving object for real-time surveillance.