Our work has focused on developing new cost sensitive feature acquisition and classification algorithms, mapping these algorithms onto camera networks, and creating a test bed of video data and implemented vision algorithms that we can use to implement these. First, we will describe a new algorithm that we have developed for feature acquisition in Hidden Markov Models (HMMs). This is particularly useful for inference tasks involving video from a single camera, in which the relationship between frames of video can be modeled as a Markov chain. We describe this algorithm in the context of using background subtraction results to identify portions of video that contain a moving object. Next, we will describe new algorithms that apply to general graphical models. These can be tested using existing test sets that are drawn from a range of domains in addition to sensor networks.