| 摘要 |
For a thermal imaging camera (1), features are extracted from a series of at least two infrared images (10, 11) or visible images (19, 20) associated therewith by a feature analysis, and an optimal correspondence between features extracted from the images (9, 10, 19, 20) is determined, and a translation vector (18) that relates the pixels of the first infrared image (10) to pixels of the second infrared image (11) is determined for the image positions (16, 17) of the corresponding features. |
| 主权项 |
1. An image processing method for a sequence of IR images (10, 11), recorded by a thermal imaging camera (1), the sequence comprising at least a first IR image (10) and a second IR image (11), the first IR image (10) being assigned to a first instant, and the second IR image (11) being assigned to a second, later instant, the method comprising recording a first VIS image (19) in a visible spectral region with an additional VIS camera (4), of a scene at the first instant, recording a second VIS image (20) in the visible spectral region with the VIS camera (4), of the scene at the second instant, the first VIS image (19) being recorded simultaneously or virtually simultaneously with the first IR image (10), and the second VIS image (20) being recorded simultaneously or virtually simultaneously with the second IR image (11), selecting a first image region (14) in the first VIS image (19), extracting at least one feature from the first VIS image (19) for the first image region (14), determining and selecting a second image region (15) in the second VIS image (20) using at least one of a feature or pattern recognition algorithm such that for the selected second image region (15) a correspondence of at least one extracted feature to the at least one feature extracted in relation to the first image region (14) is optimal, and determining by a comparison of an image position (16) of the first image region (14) in the first VIS image (19) with an image position (17) of the second VIS image region (15) in the second VIS image a VIS translation vector (18) that relates at least one pixel of the first IR image (10) to at least one pixel in the second IR image (11), and using a stored correspondence between the pixels of the first VIS image (19) and of the first IR image (10), or of the second VIS image (20) and the second IR image (11) in order to convert the VIS translation vector into a corresponding IR translation vector that relates at least one pixel of the first IR image (10) to at least one pixel in the second IR image (11), and shifting the first IR image (10) relative to the second IR image (11) by the IR translation vector such that the pictures (12, 13) of the IR images (10, 11) are superposed,
wherein the sequence comprises a third IR image, the third IR image being assigned to a third instant that is later in time than the second instant, a third VIS image of a scene is produced at the third instant, and there is selected in the second VIS image (20) a new first image region that is derived from the previously determined second image region (15) of the second VIS image (20), for the new first image region at least one feature is extracted from the second VIS image (20), a new second image region is determined and selected in the third VIS image using the at least one of the feature or pattern recognition algorithm such that for the selected new second image region a correspondence of at least one feature to the at least one feature extracted in relation to the new first image region is optimal, and there is determined by a comparison of the image position of the new first image region in the second VIS image (20) with the image position of the new second image region in the third VIS image a new IR translation vector that relates at least one pixel of the second IR image (20) to at least one pixel in the third IR image. |
