Isomap Tracking with Particle Filter

The problem of tracking an object in an image sequence involves challenges like translation, in-plane and out-of-plane rotations, scaling, variations in ambient light and occlusions. A model of an object to be tracked is built off-line by making a training set with images of the object with different poses. A dimensionality reduction technique is used to capture the variations in the training images of the object. This gives a low-dimensional representation of high-dimensional data. Isometric feature mapping, also known as Isomap, is the unsupervised nonlinear dimensionality reduction technique used to capture the true degrees of freedom in high-dimensional data. Once the training data is reduced to low-dimensions it forms a part of the state-vector of the object to be tracked. Tracking is done in a stochastic recursive Bayesian framework. Particle filters, which are based on the recursive Bayesian framework, track the state of the object in presence of nonlinearity and non-Gaussianity. The focus of this thesis is the problem of tracking a person’s head and also estimating its pose in each frame using Isomap for dimensionality reduction and particle filter for tracking. ‘Isomap tracking with particle filter’ algorithm is capable of handling rapid translation and out-of-plane rotation of a person’s head with a relatively small amount of training data. The performance of the tracker is demonstrated on an image sequence with a person’s head undergoing translation and out-of-plane rotation.