Automatic calibration of fish-eye cameras from automotive video sequences

Abstract

A technique for calibrating a lens in the automotive environment to compensate for radial distortion introduced by wide-angle or fish-eye lenses, without the need for a dedicated calibration environment, is proposed. At present, car manufacturers are endeavouring to introduce systems that provide the driver with views of the car's surroundings that are not directly visible (blind zones). To achieve this, wide-angle/fish-eye lens cameras are fitted to many modern vehicles to maximise the field of view. However, fish-eye lenses introduce undesirable radial distortion to the resulting images that can be compensated for by post-processing the images. Calibration of the camera is important for fish-eye compensation, because each camera has different intrinsic properties. However, in some situations, calibration via specific calibration set-up can be undesirable. For example, in automotive mass production, where time and space on a production line have a direct impact on cost, even minutes spent on calibration is costly. In these situations, automatic calibration can reduce production time and alleviate the associated costs. It is proposed that the radial distortion introduced by fish-eye lenses can be calibrated using video normally captured by the camera on a vehicle. Here, it is proposed to heuristically extract real-world straight lines from image frames captured in an automotive environment and use these to calibrate the fish-eye camera for radial distortion.