Learning flexible models from image sequences

Abstract

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

EigenTracking: Robust matching and tracking of articulated objects using a view-based representation

“…Image-based tracking schemes that emphasize learning of views or motion have focused on region contours (Baumberg and Hogg, 1994;Blake et al, 1994;Cootes et al, 1992;Kervrann and Heitz, 1994). In particular, Baumberg and Hogg (1994) track articulated objects by first computing a silhouette of the object via image differencing.…”

“…In particular, Baumberg and Hogg (1994) track articulated objects by first computing a silhouette of the object via image differencing. They fit a spline to the object's outline and the knot points of the spline form the representation of the current view.…”

Object Modeling for Multicamera Correspondence Using Fuzzy Region Color Adjacency Graphs

“…One approach to the detection and tracking problem is to fit explicit object models of shape, such as rigid wireframe CAD models [15,16] or flexible active shape models [17]. Some model fitting approaches focused on high-level reasoning [18,19].…”

Quasi-Random Sampling for Condensation

“…First, a statistical shape model of a pedestrian was built using automatically segmented pedestrian contours from sequences obtained by a stationary camera (so that we can do background subtraction). We use well-established computer vision techniques (see [22] and [23]) to build a LPDM (Linear Point Distribution Model). We fit a NURB (Non-Uniform Rational B-spline) to each extracted contour using least squares curve approximation to points on the contour [21].…”