Considering the problem of existed pedestrians tracking method based on head detection often made mistakes during tracking in complex backgrounds, this paper proposes an improved pedestriantracking method based on head detection to adapt to a variety of complex scenarios. This method firstlyuses the foreground segmentation technology to extract the target people, and add human body negative samples in adaboost classifier which combined with the feature of haarlike to detect head, then establish headtracking chains to fulfill the tracking of pedestrians by using the gotten head. Expriments show that the improved method reduces the error rate of head dectetion and improve the correction rate of humantracking in complex backgrounds.