On the sequential accumulation of evidence

In this paper, we introduce a method for sequentially accumulating evidence as it pertains to an active observer seeking to identify an object in a kn own environment. We develop a probabilistic framework, based on a generaliz ed inverse theory, where assertions are represented by conditional probabil ity density functions. This leads to a sequential recognition strategy in w hich evidence is accumulated over successive viewpoints using Bayesian chai ning until a definitive assertion can be made. To illustrate the theory we show how the characteristics of belief distributions can be exploited in a model-based recognition problem, where the task is to identify an unknown m odel from a database of known objects on the basis of parameter estimates. We illustrate the robustness of the algorithm through recognition experimen ts in two very different contexts: (1) a highly structured recognition cont ext where 3-D parametric models can be estimated directly from range data, (2) a complex environment, where the relationship between the data and the model is learned through an appearance-based strategy. Specifically, the fl ow fields computed through the object's motion are used as structural signa tures for recognition.