PERIODICITY DETECTION OF LOCAL MOTION (WedPmPO1)

Author(s) : Xiaofeng Tong (Chinese Academy of Sciences, China) Lingyu Duan (Institute for Infocomm Research, Singapore) Changsheng Xu (Institute for Infocomm Research, Singapore) Qi Tian (Institute for Infocomm Research, Singapore) Hanqing Lu (Chinese Academy of Sciences, China) Jinjun Wang (Institute for Infocomm Research, Singapore) Jesse Jin (University of Newcastle, Australia)

Abstract : In this paper, we present an automatic motion periodicity detection approach based on local motion analysis under complex condition. Periodicity is a significant characteriza-tion for compact representation and a proper temporal scale for periodic motion analysis. We concern local motion on interesting region rather than global motion. The task is challenging for local motion is usually buried in clutters with global motion and noises. Unlike most existing methods that assume the moving object region has been labeled or the camera is static, we apply robust local motion estimation technology and object localization method to extract object motion. The object motion is characterized by both local motion probability based confidence and motion vectors after global motion compensation. Then, we compute the autocorrelation serial of object motion energy and find local maximum points in it. With the set of indices of local maxi-mum points, we can estimate the basic periodicity through least-square fitting. We have applied this method in swim-ming videos and got encouraging results.

Menu