| 摘要 |
A system (10) for determining the pose (translation, rotation, and scale), or position and orientation, of a model object (12) that best matches a target object (11) located in image data (10). It does this through an iterative process by making small adjustments to the original position and orientation of the model object (12) until it converges to a state that best matches the target object (11) contained in the image data (10). The system (10) comprises an edge detector (22) for processing the digitized image (10) to produce edge data (23) representative of edges of the target object (11). Model storage (24) is provided for storing model data (24) corresponding to the model object (12) that is to be detected. The model data (24) comprises a plurality of data points disposed along edges of the model object (12). A minimum distance vector calculator (25) is coupled to the model storage (24) and to the edge detector (22) for processing the edge data (23) and the model data (24) for each data point in the model object (12) relative to each point in the target object (11) to produce a set of minimum distance vectors between the model object (12) and the target object (11). A neural network (26) is coupled to the the minimum distance vector calculator (25) for estimating translation, rotation, and scaling adjustments that are to be made to the model object (12) that are adapted to translate, rotate, and scale the model relative to the target object (12). A position and orientation adjustment module (27) is coupled between the neural network (26) and the model storage (24) for adjusting the position of the model object (12) relative to the target object (11) based upon the estimated translation, rotation, and scaling adjustments provided by the neural network (26). Iterative calculation of the minimum distance vectors, estimation of the translation, rotation, and scaling adjustments, and adjustment of the position of the model is adapted to reposition the model object (12) until it substantially overlays the target object (11), thus providing an estimate of the position and orientation of the target object (11) in the digitized image (10). A method (20) for determining the position and orientation of the target object (11) in a digitized image (21) is also disclosed. <IMAGE> |
