下向きカメラを有する自律走行車のための模様の劣化にロバストな自己位置修正法

抄録

This paper addresses a position correction using an original absolute difference of difference (ADD) correlation that is robust to texture degradation such as occlusion. The correlation is calculated by summing absolute differences of the difference between two successive intensities in a template image. The quality of the position correction is guaranteed by cross-checking the search results of four image groups. Using the proposed method, our experimental vehicle equipped with a downward facing camera is most successful in correcting position error compared with other correlation methods including zero-mean normalized cross-correlation (ZNCC) on an indoor floor with the texture 67% occluded. For outdoor environment on a paved road, the proposed correlation is more robust against texture degradation by sand than conventional sum of absolute differences (SAD) and increment sign correlation (ISC) calculations, and moreover, the processing time for ADD is practically short to search every template image recorded for the position correction.