Implementation of motion detectors: A case study

Conclusions Motion Detector Models • Two types of motion detectors were implemented in order to evaluate their applicability in a real-time computer vision scenario. In particular, we are interested in implementing the algorithms on an FPGA, which allows for flexible hardware implementation of algorithms at the price of introducing constraints on real-time capability. • The Reichardt correlation dectector requires the least amount of computational effort, due to its simple architecture. However, its output is noisy and brittle. Two variants were thus implemented to improve performance, one based on averaging inputs and another based on cross-correlation. Pooling results over space and time was also implemented. • The second type of detector implemented is a basic version of the oriented spatio-temporal energy filtering model (Adelson and Bergen). • Detectors were evaluated both on controlled artificial stimuli and more importantly on real-world video sequences of moving faces.