Trajectory Generation and Object Tracking of Mobile Robot Using Multiple Image Fusion

Abstract

In this paper, the proposed tracking method adds an adaptive appearance model based on color distributions to particle filtering. The color-based tracker can efficiently and successfully handle non-rigid and fast moving objects under different appearance changes. Moreover, as multiple hypotheses are processed, objects can be tracked well in cases of occlusion or clutter. This research proposes estimation and tracking scheme for a moving object using images captured by multi cameras. In the scheme, the state estimator has two algorithms: the particle filter that estimates the states for the linear approximated region, and SOM for the nonlinear region. The decision for the switchover is made based on the size of the position estimation error that becomes low enough for the linear region and becomes large enough for the nonlinear region. The effectiveness and superiority of the proposed algorithm was verified through experimental data and comparison. The adaptability of the algorithm was also observed during the experiments. For the sake of simplicity, this research was limited to the environment of a fixed-camera view. However, this can be expanded to the moving camera environment, where the input data might suffer from higher noises and uncertainties. As future research, selection of a precise learning pattern for SOM in order to improve the estimation accuracy and the recognition ratio, and development of an illumination robust image processing algorithm, remain.