| 摘要 |
<P>PROBLEM TO BE SOLVED: To enhance both detection accuracy in a case where the whole of a target object is photographed and detection accuracy in a case where the target object is at an end portion and only a part thereof is photographed without increasing a processing load. <P>SOLUTION: A target object detection device comprises: a feature amount calculation unit that calculates a feature amount from an image in a search window set for a partial area in an input image according to a predetermined rule; plural discriminators that perform discrimination based on learning using a feature amount of a teacher image; and a detection processing unit that detects a predetermined target object from the image in the search window using the feature amount and one of the discriminators. The plural discriminators include a normal-use discriminator learned using an image in which the whole of the target object is photographed as a positive teacher image, and an end-portion-use discriminator learned using an image in which a part of the target object is photographed as a positive teacher image. The detection processing unit detects the predetermined target object using the normal-use discriminator, and when the search window includes an end portion of the input image, detects the predetermined target object further using the end-portion-use discriminator. <P>COPYRIGHT: (C)2012,JPO&INPIT |
