| Summary: | 碩士 === 國立雲林科技大學 === 電機工程系碩士班 === 99 === Most of License Plate Recognition (LPR) systems have still been applied to fixed scenes. In order to make the LPR system suitable for diverse applications, border stain issue caused by inaccurate license plate localization under complex scenes is studied out in this thesis.
Well-known conventional stain removal method , histogram-based stain removal, suffers severely from trapezoid stains residing at the right side and left side, and is difficult to reliably define the threshold for removing the border stains at the upside and downside. So the license plate character recognition rate after conventional histogram-based stain removal is only 79.48%.
In this paper, mathematical morphology and fill-in methods are proposed to remove trapezoidal stains at the left-side and right-side. And, a novel method of vertical-edge projection histogram segmentation is also proposed to discriminate r the threshold for removing the border stains at the upside and downside effectively and easily. Experimental results show the proposed methods can achieve the license plate character recognition rate of 98.56% under complex scenes.
Finally, this thesis integrates the propose border stain removal methods into the original LPR system, and implements as an Android-based patrol robot featuring automatic LPR.
|
|---|
