Font Size: a A A

Research On Target Tracking Technology In Case Of The Occurrence Of Occlusion

Posted on:2015-01-10Degree:MasterType:Thesis
Country:ChinaCandidate:S Y DongFull Text:PDF
GTID:2298330431983037Subject:Computer technology
Abstract/Summary:PDF Full Text Request
As computer technologies have been developed tremendously in the age of intelligence and information, target tracking technology as an important branch in computer vision also receives more and more attention. I In practical applications, such as traffic control, space exploration, national defense and even the most basic security monitoring and traffic controlling of city roads have broad application prospects, some auxiliary algorithms are often used to optimize target tracking algorithm. Such as object segmentation, trajectory filtering and prediction. Simultaneously, the scenes and lighting changes, target occlusion become a big challenge of target tracking.This paper mainly for study and improve the tracking algorithm of moving object in the video. The real-time target tracking will be adversely affected when the target has been occluded or the target appearance occurs shading changes, Based on information herenbefore, a algorithm which could make the object identification and tracking valid, even when light changes or mutual occlusion occurred is proposed.Firstly, a determination method which integrates the conditions statistical morphological background subtraction and frame subtraction is used to remove the noise randomly distributed in the vidow image, then a shading model based on HSV color space is used to remove object shadow. Secondly, a improved local Camshift algorithm which integrated the invariant moments information is proposed, this algorithm can avod the unrelated point as the target region when the image color space transforming,but also reduces the sensitivity to the light of original algorithm. Finally, a new algorithm based on regular particle filter target SIFT features is proposed to deal with the case of the occlusion tracking. This algorithm makes full use of SIFT features in describing the target, firstly establish the model of the target with the SIFT of itself, and then integrate it into the regularized particle filter method, thus, it can effectively alleviate the problem of particle degradation caused by resampling,and it has better accuracy and tracking robustness to object occlusion, scale change and the deformation.
Keywords/Search Tags:Invariant moment, CamShift, Scale invariant feature, Regularizedparticle filter
PDF Full Text Request
Related items