A method based on depth image local normal vector characteristic, described by spherical coordinate parameters ( , ), was proposed in this paper and suited to rail vehicle with the high-density passengers. This method implemented the human recognition by capturing the local normal vector, and then extracting the head contour. Depth image was a better way to solve the limitations of environmental factors effectively, such as height and illumination. Finally, some experiments were taken to test and verify the real-time performance and accuracy.