Automatic Detection of Range Variance of Facades from Vehicle- Based Image Sequence

Vehicle-based image sequence is taken on a moving vehicle, therefore, all scenes passed by will be included, e.g. buildings, crossroad etc.. Images including cross-road are obviously useless for building facade texture. Therefore, those images should be excluded from processing. In this paper, subsection of image sequence is implemented according to range variance automatically detected by the histogram of projective difference of corresponding points. Firstly, projective difference of corresponding points on common projective plane is calculated according to the DC parameters determined. As far as corresponding points on facade be concerned, the projective difference should be very small or even ideal zero. Corresponding points in cross-road or on other facades having obviously range variance with common projective plane, however, have large projective difference. Accordingly 1-D histogram along x-axis of image coordinate system is drawn to show the distribution of projective difference. Experiments show that cross-road and facades having large range variance are represented in the histogram by peak areas. Moreover, the 1-D locations of peak areas in the histogram are obviously corresponding to those having large range variance. By histogram of corresponding points, therefore, not only the existence of range variance is detected, but also the location of range variance in image is indicated because it is drawn following the moving direction of vehicle. According to the location indicated, images including cross-road are excluded from image sequence and subsections of image sequence are divided for facades having large variance with each other as well.