Statistical approach for distance estimation using Inverse Perspective Mapping on embedded platform

The paper presents a real time image processing algorithm for distance estimation on Beagleboard. The system uses a single camera fixed at a stationary position to capture the real time image of target object and determine its distance in contrast to existing and most common vision algorithms of stereo vision. The image captured from a single forward facing camera suffers from high degree of uncertainty in object distance estimation due to the nonlinear relation between object height and its actual distance from camera. This uncertainty is eliminated by using Inverse Perspective Mapping to translate the front view captured by the photosensitive sensor to a bird's eye view. The focus of the paper is to present a statistical approach to scale the position of object in pixel to the real world physical distance. The proposed algorithm offers high efficiency in determining the distance of target object in a short time and thus aids the autonomous vehicle in intelligent navigation and feedback. The implementation is done using Intel OpenCV image processing libraries to reduce system overhead and is intended to work at least at 30 fps with VGA resolution.