Particle Filter with Multiple Motion Models for Object Tracking in Diving Video Sequences

Abstract

This paper addresses the problem of object tracking in diving video sequences by particle filter. Because the diversity of motions in diving video sequences such as bouncing on the springboard, somersaulting in the air increases the difficulty to construct particle motion model, this paper presents an object tracking method in diving video sequences by particle filter with multiple motion models. In this method, video paragraphing based on Hough transform technique and the knowledge of critical frame is proposed to divide a whole diving video sequences into several sub-sequences, and construct a particle motion model for each sub-sequences. The particles are predicted by multiple motion models to adjust to athlete motions in diving video sequences. Experiments demonstrate that the object tracking method is efficient in diving video sequences.