learning scene specific shape priors


Detecting, isolating, and tracking moving objects in an outdoor scene is a fundamental problem of visual surveillance. A key component of most approaches to this problem is the construction of a background model of intensity values. We propose extending background modeling to include learning a model of the expected shape of foreground objects. This paper describes our approach to shape description, shape space density estimation, and unsupervised model training. A key contribution is a description of properties of the joint distribution of object shape and image location. We show object segmentation and anomalous shape detection results on video captured from road intersections. Our results demonstrate the usefulness of building scene-specific and spatially-localized shape background models.


First a large database of examples shapes are collected from the scene, the process is described in the figure below. Then a PCA basis is built from this database. This basis is then used to improve object segmentation and anomaly detection.

An overview of the process we use to collect example shapes.


This figure (bottom) shows the coordinates of example shapes with respect to a PCA subspace (top). The first principal direction, left, corresponds to size of the object. The second principal direction, right, corresponds to the orientation of the object.

Example shapes and their position in the first two dimensions of a local PCA subspace. Points are colored by when they arrived in the scene. Notice that two different buses moved through the scene.