Object boundary detection in image sequences using multistage hypothesis testing

Summary form only given. It has been shown that multistage hypothesis testing can be applied to the detection of moving object boundaries. An object boundary is modeled as a collection of connected linear edge segments, each translating in a direction perpendicular to its orientation. The motion of each segment is estimated by posing a multiple hypothesis testing problem: a large number (1000´s) of edge segments are hypothesized at each pixel in each image. Performing all these tests in a brute-force manner is clearly not feasible. However, the computation time can be greatly reduced by structuring the search space hierarchically: sequential hypothesis testing is used to prune a tree-structured list of spatially overlapping, multiple hypotheses. Each tree is pruned using a multistage, two-threshold test. Moreover, the performance and computational requirements of the algorithm can be predicted for the case of white Gaussian noise backgrounds