To deal with the problem of sample degeneration and sample impoverishment in traditional particle filter，a weight selected method is presented.The proposal distribution for each particle involves sampling from the state-space model a number of times.In order to improve the accuracy of infrared object tracking，the grey-scale feature and gradient feature are combined to establish observation model based on infrared image characteristics，and the influence of each characteristic on tracking is adjusted by depending on confidence.Simulation results show that the improved algorithm is robust，and can track infrared object stably under complex background.