Statistical Models for Tracking and Detection

We explore the behavior of two different statistical models , one based on the beta-binomial distribution and another based on sim- ple unigrams. We show that tracking systems based on these models have complementary strengths and weaknesses: the Beta-Binomial system yields high precision at high decision threshold, bu t perfor- mance quickly degrades as the threshold drops; the Unigram sys- tem is not as strong at high decision threshold, but is very go od at suppressing false-alarms at lower threshold. We will descr ibe the features of these systems that give rise to this behavior, an d discuss ways that each system might be improved by borrowing from the other. We will also discuss our detection system, and how improve- ments in tracking can be carried over to improvements in detection.