摘要 |
<P>PROBLEM TO BE SOLVED: To create a landmark database so as to estimate a self-position of a mobile body at high speed by use of a rotationally invariant and scale-invariant characteristic amount. Ž<P>SOLUTION: A landmark learning device has: a characteristic point extraction part 102 extracting a characteristic point in each image based on time-series image data, and calculating a characteristic amount; a characteristic point tracking part 103 outputting present image coordinates of a point to be tracked based on a present image, and an ID and image coordinates of the point to be tracked obtained in a past image; and a characteristic point decision part 104 selecting a characteristic point wherein characteristic amounts are similar between the images by pattern recognition in subsequent image processing, setting the characteristic point as a landmark candidate point when the selected characteristic point can be sequentially tracked, and deciding left points as landmarks when the number of the landmark candidate points becomes a threshold value or below. Ž<P>COPYRIGHT: (C)2010,JPO&INPIT Ž
|