Computing Optical Flow to Obtain Motion Field

In this article, we describe a method for optical flow computation that produces motion fields. To achieve this objective, we employ the Lucas-Kanade algorithm - a classical computer vision technique that efficiently tracks motion patterns within image sequences. The algorithm operates by applying a local gradient-based approach, assuming constant brightness between frames and solving the optical flow equation using least squares optimization over a neighborhood of pixels. Specifically, we implement this algorithm on sequential images to estimate per-pixel displacement vectors between consecutive frames. This process yields an optical flow field where each pixel is assigned movement components in both x and y directions. These displacement vectors provide crucial information about object motion within the scene, which is fundamental for numerous computer vision applications such as motion segmentation, object tracking, and video stabilization. The implementation typically involves key functions like gradient computation, matrix inversion for solving linear equations, and iterative refinement for handling larger displacements.