Due to uneven underground illumination and high similarity between pedestrian characteristics and background, pedestrian detection technology based on computer vision is facing great challenges in underground application. Faster region convolutional neural networks(RCNN) was proposed for pedestrians detection of coal mine underground. Faster RCNN pedestrian detection algorithm uses region proposal network(RPN) to generate candidate regions. RPN shares convolutional layer with Fast RCNN, so as to improve network training and detection speed. A dynamic self-adaptive pooling method is adopted to perform self-adaptive pooling operation for different pooling domains in the process of image feature extraction, so as to improve detection accuracy. The experimental results show that the algorithm has better detection effect for pedestrian image in different environments.