Comparative Analysis of Optical Flow Techniques: Classical Computer Vision vs Deep Learning Approach
This analysis entails a comprehensive examination of Optical Flow using both Classical Computer Vision techniques and the Deep Learning Model, specifically FlowNet 2.0. The comparison hinges on crucial performance metrics: L1 error, Average Endpoint Error, and Average Angular Error. The obtained flow is then visually represented through a grid of arrows, facilitating face tracking via a Haar-Cascade Classifier. The optical flow field is subsequently scrutinized to assess both the velocity of movement and its directional aspects. Furthermore, a user-friendly user interface is designed for the Classical Computer Vision approach, allowing users to visualize optical flow with different step sizes and reset the region of interest.