Local estimation fusion for tracking objects under occlusion

The aim of this work is to present an algorithm that solves the track fusion problem for object tracking under occlusion. The approach uses local estimates of the object positions. These estimates are obtained by Kalman filters using a constant velocity motion model. The sensors process its own information with different tracking algorithms and send the position estimates to a central node, where the fusion is done by a simple convex combination. The contribution of this work is the correction of the input data to the filters using a comparison between the detected and estimated position using a threshold. This assures that the weighted data are better than the detected. The algorithm is tested with real images obtained from similar sensors. It behaves better than the one using a global Kalman filter. The results show that the probability of missed objects with our approach is less than in each sensor.