Road Boundary Detection in Night Video Sequence: A Novel Technique for Autonomous Vehicles

This paper presents the enhancement of road boundaries in the test video recorded at night time. This is fact that road boundaries become of more importance in night time driving when loss of light makes every thing harder to see. Road boundary detection is a basic and vital research work area because it is a prerequisite for automatic pilot in the intelligent vehicles system project. The boundary detection is a process of analyzing images obtained by vehicle-mounted camera to obtain meaningful information. The research work initially focuses the problem of image enhancement and then for road boundary extraction. Though many different techniques were developed for the extraction of road boundary, the results were not up to the mark as with non uniform illumination a variety of shadows and reflections are created making drastic changes in the actual road scene. The proposed road boundary detection technique brings the preprocessed image stream into frequency domain for fast processing and the video signals are analyzed. In two dimension array of signals the low frequency signals embed in high frequencies are treated differently by bringing them to a very low amplitude signals or zero. The video signal array is then converted into the spatial domain again for further processing. This novel idea can be applied on both painted and unpainted roads, curved and straight roads and with different illumination conditions.