CUDA accelerated visual relative motion estimation
Safa Ouerghi, Fethi Tlili · 2016
Egomotion estimation is a fundamental issue in structure from motion and particularly for ADAS systems. Several camera motion estimation methods from a set of variable number of image correspondances were proposed. Seven-point method represent the minimal number of required correspondences to estimate the fundamental matrix, raised special interest for their application in a hypothesize-and-test framework. This algorithm allows relative pose recovery at the expense of a much higher computational time. In this work, we present a GPU-based solution for the essential matrix estimation performed using the 7-point algorithm, and complemented with robust estimation. The description of the hardware-specific implementation considerations is given in detail. Performance analysis against existing CPU implementations is also given. Our implementation showed good performance with a speedup reaching 36×.