Depth driven people counting using deep region proposal network

People counting is a crucial subject in video surveillance application. Factors such as severe occlusions, scene perspective distortions in real application scenario make this task challenging. In this paper, we carefully designed a deep detection framework based on depth information for people counting in crowded environments. Our system performs head detection on depth images collected by an overhead vertical Kinect sensor. To the best of our knowledge, this is the first attempt to use the deep convolutional neural networks on depth images for people counting. We explored the impact of the number and quality of RPN positive anchors on the performance of Faster R-CNN and proposed a solution. Our method is very simple but effective, not only showing promising results but also efficiency as it runs in real-time at a frame rate of about 110 frames per second on a GPU.