As part of my course "Optimization for engineers," I completed an individual project where I applied optimization methods covered in the course to a real-world problem. The aim of optimization algorithms is to find the best possible solution by searching through a vast number of possibilities in a given space. The best solution is defined as the one that satisfies a specific criterion the most.
For my project, I chose to focus on genetic algorithms, a bio-inspired algorithm that mimics the process of natural selection. Specifically, I conducted a comparative study of three different optimization algorithms - Random Search, micro-genetical algorithm, and evolutionary strategy - to enhance the RANdom SAmple Consensus (RANSAC) algorithm for a computer vision problem.
The goal was to find the best eight points among thousands for the external calibration of a stereo vision setup. In my experience, the evolutionary strategy algorithm showed faster convergence, required less computation effort, and produced more precise results than the original RANSAC algorithm and micro-genetical algorithm. This study could help improve the calibration of stereo vision setups, thereby improving their accuracy and reliability.