Researchers in biology regularly produce large data sets of noisy images and videos that contain hundreds of fluorescent objects interacting in a cluttered background. This dissertation presents a statistical framework for analyzing images and videos of this kind. Specifically we analyze images that contain hundreds of overlapping polonies and videos of hundreds of vesicles and tubular organelles produced by total internal reflection fluorescent microscopy.

We approach the problem of detecting and tracking multiple fluorescent objects by defining statistical data models for individual objects and background, with clear rules to compose them in an image. A statistical model for the data allows us to formulate well defined hypotheses and properly weigh them on-line. The computational challenge of object detection is addressed by defining a sequence of coarse-to-fine tests, derived from the statistical model, to quickly eliminate most candidate locations for the objects. The computational load of the tests is initially very low and gradually increases as the false positives become more difficult to eliminate. Only at the last step, state variables are estimated from a complete time-dependent model. Processing time thus mainly depends on the number and size of the objects in the image and not on image size.

The main contributions of this dissertation are: (a) a general statistical model for image and video data presenting multiple fluorescent objects such as polonies, vesicles and tubular objects, (b) the use of these models to derive coarse-to-fine stable algorithms for efficient detection and tracking of overlapping objects (c) the derivation of simple tests to identify types of vesicle dynamics, and (d) two end-user applications: one for polony detection and another for vesicle tracking and dynamics identification.