Camshift algorithm requires the target have a large color similarity between the adjacent frames , the target is easy to get lost when moving in high speed or occlusion appears. Aiming at this, an improved Camshift algorithm is proposed ,which contains a weighted color probability distribution and localsearchguiding mechanism, combining with the Kalman filter forecasting model.The experiments show that, this algorithm have a good tracking effect in complex background when object moves in highspeed or it is occluded.