A Real-Time Pedestrian Counting System Based on RGB-D

Pedestrian counting is an important task in visual surveillance. Existing computer vision based pedestrian counting methods usually require high image quality and simple background, hence their applications are limited. In this paper, we introduce an innovative RGB-D based system for real-time pedestrian counting, which is easy to install and robust to work under various conditions. The system uses a RGB-Time Of Flight (TOF) camera to capture depth and RGB images simultaneously, then detects and tracks pedestrians based on fusion of both images. We propose a Deep Convex Convolution Filtering (DCCF) algorithm for pedestrian detection in depth images in order to overcome the problem of parameter sensitiveness in traditional methods. Experimental results highlight the effectiveness and efficiency of our designed system. Our proposed system has already been put to work in public places of multiple cities successfully.