Detection and Tracking of Objects in Low Contrast Conditions

We present an efficient object detection and tracking technique using still cameras in low contrast conditions. The tracking algorithm involves background subtraction using Gaussian Mixture Model (GMM). Our method involves updating the parameters of the Mixture Model using a combination of an online k-means approximation technique and the ExpectationMaximization (EM) algorithm. We have shown experimentally that our proposed method yields results with higher accuracy and superior performance in situations where foregroundbackground contrast is low, as compared to established techniques involving only either one of k-means or EM algorithm to update mixture parameters.