Long Term Trajectory Prediction of Moving Objects Using Gaussian Process

Abstract

Long term trajectory prediction of moving objects has many applications in robotics. There are several intelligent techniques such as MLP and ANFIS which have been applied on prediction problems. But, for online training using small size deterministic dataset, the above techniques fail to apply. In this paper we use less parametric nonlinear technique called Gaussian process for long term trajectory prediction of moving objects. Our simulation results show that Gaussian process approach can be successfully applied by using recursive and direct long term prediction strategies. It is also more robust to noise and can be generalized based on small size dataset.