With the development of computer technology, the intelligent and automatic pedestrian detecting system continues to emerge. The pedestrian detecting system is an important part of the information management for enterprises and institutions. The traditional method has a lower accuracy due to limb occlusion and light changes. This study proposes a vertical detection method for the head features, which can ensure that the features cannot be blocked even in high flow density. Firstly, the gradient histogram feature of the foreground image is extracted and the head target is detected by SVM. The MeanShift algorithm is used to track the head target in the adjacent frame by using the color feature of the head. The pedestrian is detected according to the target track. The algorithm is applied and tested in the embedded system. The experiment results show that the method is effective and accurate.