Surveillance Camera Autocalibration based on Pedestrian Height Distributions

We propose a new framework for automatic surveillance camera calibration by observing videos of pedestrians walking through the scene. Unlike existing methods that require accurate pedestrian detection and tracking, our method takes noisy foreground masks as input and automatically estimates the necessary intrinsic and extrinsic camera parameters using prior knowledge about the distribution of relative human heights. Our algorithm is computationally efficient enough for online parameter estimation. Experimental results on both synthetic and real data show the robustness of our method to camera pose and noisy foreground detections.