Learning flexible models from image sequences

摘要: The “Point Distribution Model”, derived by analysing the modes of variation a set training examples, can be useful tool in machine vision. One drawbacks this approach to date is that data acquired with human intervention where fixed points must selected eye from example images. A method described for generating similar flexible shape model automatically real image data. cubic B-spline used as vector model. Large sets are generate robust profile use labelling and tracking pedestrians real-world scenes. Furthermore, an extended which incorporates direction motion, allowing extrapolation shape.