Aiming at the unstable characteristics of the particle filter algorithm which is easily affected by background noise in complex dynamic scene while it is tracking target, an improved Kalman particle filter (KPF) target-tracking algorithm is put forward. The method of using embedded Kalman particle filter is used to predict the predicted status value of particle filter secondarily. And the secondary sampling technique is used to enhance particle richness, and thus eliminates the influence of background noise to a certain extent. Besides in order to meet the requirements of Kalman on linear motion and to eliminate the effect of rapid background change on tracking accuracy, the gray projection algorithm is promoted to calculate the background migration for motion compensation. The experiment results show that the improved Kalman particle filter algorithm can effectively track the moving object in the complex dynamic scene, which proves that the propposed KPF algorithm has high precision, strong robustness and good real-time performance.