Adaptive node dimension (AND) calculation algorithm for mesh based motion estimation

Abstract

For any video-based applications, estimating the motion of dynamic object(s) inside a video frame is required to be as accurate as possible. Different standard methods (block and mesh based) of motion estimations have their own merits and demerits. Block based approach gives reasonably good performance in terms of accuracy with limited amount of bit rate but suffers from discontinuity of motion at block boundaries. On the other hand, mesh based approach can capture the motion of an moving object inside a video frame more accurately than block based approach with high computational complexity. If motion in the camera is present, then this technique calculates large number of nodes to represent the camera movement. Since, all nodes are required to transmit from encoder to decoder, the camera movement may cause a large bandwidth requirement for the mesh based motion estimation technique. In this paper, a new approach of motion estimation has been proposed, which can be applied for generating a mesh for moving object(s) inside a video frame and node dimension of the mesh is adaptively changes according to the motion vectors of the neighborhood. This will be very useful to estimate of motion of both object and camera movement with a very little error metric.