Object detection and tracking using statistical and stochastic techniques

This paper proposes a multilevel structure for object detection and tracking in simple and complex environments. The foreground object is obtained using self-adaptive Gaussian Mixture Model (GMM) for dealing with the illumination changes, repetitive motion of the targets and clutters in the scenario. To obtain the robust and flexible target tracking, synergizing combinations of the two random modeling techniques are used. One is the Pseudo-2D Hidden Markov Models (P2DHMMs) for modeling the outline of the object and detects the human. The other is the Kalman Filter which uses the P2DHMM output to track the detected human.