The invention discloses an expected maximum window width variable nuclear particle filtering-based tracking method, which comprises the following steps of: describing a moving object by using a motion parameter-containing state equation; obtaining a particle set at the next moment by using the state equation based on the initialized particle set; calculating the covariance, posterior probability, initial weight, characteristic value, global nuclear widow width and the like of the particle set according to an expectation-maximization (EM) algorithm, and a Cholesky algorithm; and repeating the whitening process to obtain the position and weigh value of each particle in the particle set so as to further obtain the position and size of the object in the current frame. By the method, the moving object in an image is tracked robustly in real time; and compared with the similar method, the method can achieve the same particle filtering tracking effect with less particles and improve the execution efficiency of the algorithms.