Абстрактный

Region Based Sampling Method for Tracking Abrupt Motion of an Object

E Ramya (II ME), K Manivannan

The problem of tracking can be defined as identifying the motion of the target in consecutive frames in the video. Conventional tracking methods fail to track abrupt motions because the motion is not smooth and uniform. A novel tracking algorithm based on Monte Carlo sampling method to deal with abrupt motions is introduced. The algorithm efficiently handles smooth and abrupt motions with the help of marginal likelihood and density-of-state term. The N-fold way WLMC method captures abrupt motions more quickly with the rejection free property of the algorithm. It helps estimate the dos value with a less number of samples without obtaining samples from the whole state space. The target tracking is guided by considering the measurement of the adjacent targets. A context aware sampling method which considers the position of the objects which are adjacent to the target helps track the target in case of occlusions. Experimental results show that the algorithm tracks the target in scenes containing complex scenarios such as occlusions, fast and abrupt changes in position and scale