Object tracking using virtual particles driven by optical flow and Kalman filter

Abstract

In the field of robot vision, object tracking plays an important role. We propose a method of tracking object, where Kalman filter and optical flow are used to track moving object. To apply the Kalman filter to object tracking, virtual particles which consist of pairs of velocity and position are introduced. The spatio-temporal differentiation method is used to derive the optical flow. And, to increase particles which contribute tracking, weighting and resampling processes are used. We used sythesized image and captured image to test the proposed method. Experimental results show that using Kalman filter makes object tracking accurate and using weighting and resampling enables particles to concentrate on the moving object.