Hierarchical camera auto-calibration for traffic surveillance systems

• Camera self-calibration procedure with 5 levels of hierarchy and 12 different cases.
• Completely autonomous system, independent of the position of the camera.
• Focal average error < 5%, pitch average error < 3.2°, roll average error < 0.7°.
• Occlusion management ratio = 91% single frame and 95% multi-frame.
• From a total of 2269 objects, the system has obtained a detection rate of 93.3%.

In this paper, a hierarchical monocular camera auto-calibration method is presented for applications in the framework of intelligent transportation systems (ITS). It is based on vanishing point extraction from common static elements present on the scene, and moving objects as pedestrians and vehicles. This process is very useful to recover metrics from images or applying information of 3D models to estimate 2D pose of targets, making a posterior object detection and tracking more robust to noise and occlusions. Moreover, the algorithm is independent of the position of the camera, and it is able to work with variable pan-tilt-zoom (PTZ) cameras in fully self-adaptive mode. The objective is to obtain the camera parameters without any restriction in terms of constraints or the need of prior knowledge, to deal with most traffic scenarios and possible configurations. The results achieved up to date in real traffic conditions are presented and discussed.