Tracking-Based Automatic Object Classification

The Problem: While a tracking system is unaware of the identity of any object it tracks, the identity remains the same for the entire tracking sequence. Our system leverages this information by using accumulated joint co-occurrences of the representations within the sequence to create a hierarchical binary-tree classifier of the representations. This classifier is useful to classify sequences as well as individual instances. This work classifies tracked objects based on their shape using the binary motion silhouettes produced by our tracker[3]. Motivation: Infants have an innate ability to do primitive tracking. As they track objects, they are unaware of the identity of the object. But as long as they observe a stable input signal, they can be relatively certain that the object is the same object. We are investigating using the information gained by simple, primitive tracking behavior to aid in visual tasks- in particular, unsupervised hierarchical classification. Our goal is to produce a system that can be situated in a new environment and, without intervention, produce models of the objects in that environment. Because this work depends on tracking moving objects to classify, it will be most useful in tasks which involve tracking because it can first learn to classify and then use the classification in its task. For instance, it may be capable of determining a vocabulary of object types for activity monitoring, interactive environments, home security, outdoor wildlife monitoring, or many other video sources. Previous Work: Baumberg and Hogg have done some early work on learning parametric models of shape fro