Sergey Ioffe

Probabilistic Methods for Finding People

Finding people in pictures presents a particularly difficult object recognition problem. We show how to find people by finding candidate body segments, and then constructing assemblies of segments that are consistent with the constraints on the appearance of a person that result from kinematic properties. Since a reasonable model of a person requires at least nine segments, it is not possible to inspect every group, due to the huge combinatorial complexity.We propose two approaches to this problem. In one, the search can be pruned by using projected versions of a classifier that accepts groups corresponding to people. We describe an efficient projection algorithm for one popular classifier, and demonstrate that our approach can be used to determine whether images of real scenes contain people.The second approach employs a probabilistic framework, so that we can draw samples of assemblies, with probabilities proportional to their likelihood, which allows to draw human-like assemblies more often than the non-person ones. The main performance problem is in segmentation of images, but the overall results of both approaches on real images of people are encouraging.

Human tracking with mixtures of trees

Tree-structured probabilistic models admit simple, fast inference. However they are not well suited to phenonena such as occlusion, where multiple components of an object may disappear simultaneously. We address this problem with mixtures of trees, and demonstrate an efficient and compact representation of this mixture, which admits simple learning and inference algorithms. We use this method to build an automated tracker for Muybridge sequences of a variety of human activities. Tracking is difficult, because the temporal dependencies rule out simple inference methods. We show how to use our model for efficient inference, using a method that employs alternate spatial and temporal inference. The result is a cracker that (a) uses a very loose motion model, and so can track many different activities at a variable frame rate and (b) is entirely, automatic.

Bayesian structure from motion

Formulates structure from motion as a Bayesian inference problem and uses a Markov-chain Monte Carlo sampler to sample the posterior on this problem. This results in a method that can identify both small and large tracker errors and yields reconstructions that are stable in the presence of these errors. Furthermore, the method gives detailed information on the range of ambiguities in structure given a particular data set and requires no special geometric formulation to cope with degenerate situations. Motion segmentation is obtained by a layer of discrete variables associating a point with an object. We demonstrate a sampler that successfully samples an approximation to the marginal on this domain, producing a relatively unambiguous segmentation.

The Joy of Sampling

A standard method for handling Bayesian models is to use Markov chain Monte Carlo methods to draw samples from the posterior. We demonstrate this method on two core problems in computer vision—structure from motion and colour constancy. These examples illustrate a samplers producing useful representations for very large problems. We demonstrate that the sampled representations are trustworthy, using consistency checks in the experimental design. The sampling solution to structure from motion is strictly better than the factorisation approach, because: it reports uncertainty on structure and position measurements in a direct way; it can identify tracking errors; and its estimates of covariance in marginal point position are reliable. Our colour constancy solution is strictly better than competing approaches, because: it reports uncertainty on surface colour and illuminant measurements in a direct way; it incorporates all available constraints on surface reflectance and on illumination in a direct way; and it integrates a spatial model of reflectance and illumination distribution with a rendering model in a natural way. One advantage of a sampled representation is that it can be resampled to take into account other information. We demonstrate the effect of knowing that, in our colour constancy example, a surface viewed in two different images is in fact the same object. We conclude with a general discussion of the strengths and weaknesses of the sampling paradigm as a tool for computer vision.

Mixtures of trees for object recognition

Efficient detection of objects in images is complicated by variations of object appearance due to intra-class object differences, articulation, lighting, occlusions, and aspect variations. To reduce the search required for detection, we employ the bottom-up approach where we find candidate image features and associate some of them with parts of the object model. We represent objects as collections of local features, and would like to allow any of them to be absent, with only a small subset sufficient for detection;furthermore, our model should allow efficient correspondence search. We propose a model, Mixture of Trees, that achieves these goals. With a mixture of trees, we can model the individual appearances of the features, relationships among them, and the aspect, and handle occlusions. Independences captured in the model make efficient inference possible. In our earlier work, we have shown that mixtures of trees can be used to model objects with a natural tree structure, in the context of human tracking. Now we show that a natural tree structure is not required, and use a mixture of trees for both frontal and view-invariant face detection. We also show that by modeling faces as collections of features we can establish an intrinsic coordinate frame for a face, and estimate the out-of-plane rotation of a face.

Finding objects by grouping primitives

We describe the use of a representation, called a body plan, to segment and to recognize people and animals in complex environments. The representation is an organized collection of grouping hints obtained from a combination of constraints on color and texture and constraints on geometric properties such as the structure of individual parts and the relationships, between parts. The approach is illustrated with two examples of programs that successfully use body plans for recognition: one example involves determining whether a picture contains a scantily clad human, using a body plan built by hand; the other involves determining whether a picture contains a horse, using a body plan learned from image data. In both cases, the system demonstrates excellent performance on large, uncontrolled test sets and very large and diverse control sets. The mechanism of recognition by assembly is very general; we describe previous work on finding clothing by marking folds and then assembling groups of folds.