Non-rigid Object Tracking By Support Vector Regression Particle Filtering

Tracking technique is one of the most active research subjects in the area of computer vision research. With the rapid development of modern computer science and information technology, and with the marvelous renovation of image identification, tracking technique comes out into the open and is of great practical value in the field of military defense, medical research, traffic monitoring, astronomical prediction, intelligent supervision etc.Particle filter based on the Bayesian estimation of nonlinear filtering have unique advantages in dealing with non-linear moving targets. In order to improve the robustness of tracking, namely more accurate approximating posterior probability of the tracking target, two aspects have been studied as follows: (1) the weight of the particle is calculated by observing function. But the observed values will be affected by environmental noise, such as background confusion or the target deformation, Support Vector Regression (SVR) is used to enhance the effectiveness of particle weight and lower the noise by re-estimating particle weight; (2) increasing the number of particles can improve the precision of the tracking, but a large number of particles will greatly increase the complexity of the calculation which will reduce the efficiency of tracking. In order to solve this problem, distribution algorithm is used to calculate the predicting values of the particles which have the characteristics of independent and identically distribution.The research results are as follows: 1. Identifying and inhibiting the noise by the SVR is researched; 2. Re-weighting the power of particles by the SVR is added to improve the accuracy in the athletes real-time tracking system; 3.With the Characteristics of independent and identically distribution of the particles, distribution algorithm is used to improve the tracking robustness of our real-time tracking system. The experimental results show that our approach is effective.