At present, the problem of pedestrian detection has attracted increasing attention from researchers in the field of computer vision. And the Faster R-CNN (faster regions with convolutional neural network) is regarded as one of the most important techniques in this problem. However, the recognition effect of this method is still inadequate in some respects, such as fitting the diversity of pedestrians' clothing or the light intensity, which can lead to false and miss detection for pedestrians in large probability. We try to improve the recognition rate by improving the sample contrast in this paper and propose the modified Faster R-CNN method with automatic color enhancement (ACE). Firstly, we collect considerable quantity of head-and-shoulder image samples as the training dataset. Then, the ACE is performed on the dataset to improve the image contrast. Next, the dataset is input into the Faster R-CNN. Finally, the effectiveness of this method is verified with the actual data collected from the subway station in Beijing.