Kalman filtering for tracking video objects based on motion estimation

Abstract

A video tracking algorithm that used the prediction of object centroids based on Kalman filtering was preposed. Firstly, video objects were segmented and their centroids were calculated. Then, Kalman filtering was used to predict the object centroids in next frame with the information of centroids and motion vectors in successive two frames of a video sequence, so It was able to track multiple non-rigid moving objects fast, efficiently and automatically. Experimental results show that the proposed method is robust to deal with the problems in multi-targets tracking such as new target entering, target leaving, scaling and deforming of non-rigid objects.