This project aims to resolve the rotational motion (3Dof) estimation using the Contrast Maximisation framework globally. Iterative optimisation was employed by previous work, in which proper initialisation is necessary. Based on Branch-and-Bound, our method is able to solve the problem globally with a novel bounding function for Contrast Maximisation. The theoretical validity of our bounding function is rigorously established.

This project propose Spatiotemporal Registration (STR) as a cogent alternative approach to Contrast Maximisation framework. The complexity of STR is independent to the resolution of the optical sensor. Moreover, The accuracy is less afftected by the event batch size, which can produce a higher temporal resolution estimation. In particular, the STR also generate feature tracks as by-product, which can be futher incorporated into visual odometry pipepline for motion averaging.