| 摘要 |
<p>A target fish-eye image (100) for object recognition is split (120) into regions in accordance with the direction of distortion of the target fish-eye image (100). An object recognition portion (130) performs object recognition by using databases (140) prepared in accordance with the regions respectively. The same process is applied to a plurality of target fish-eye images (110) created by rotation of the target fish-eye image (100). A detected coordinate transformation portion (150) restores the obtained position of an object into the original position by inverse rotation, and outputs a thus obtained result as a detection result. Accordingly, detection accuracy in object recognition can be improved, and the quality of data in databases can be reduced. Rotated object images required for database creation are used for creation of databases for object recognition for the fish-eye image. Thus, it is possible to directly use a distorted fish-eye image for object recognition without projective conversion or image conversion based on corrective operation to the fish-eye image.</p> |
