A Genetic Algorithm for Optimizing Background Subtraction Parameters in Computer Vision

Tracking moving objects in a video sequence is a critical task in several computer vision applications. A common approach is to perform background subtraction which identifies moving objects in a video frame. The mixture of Gaussians model is one of the most popular techniques for performing background subtraction. The performance of the mixture of Gaussian model strongly depends on parameters such as learning rate, background ratio, and number of Gaussians. Fine tuning these parameters is a huge challenge for efficient performance of the background subtraction algorithm. In this work, we propose a genetic algorithm to determine the optimal values of the learning rate and background ratio. Experiments based on the Wallflower test images demonstrate the superior performance of the genetic algorithm when compared to a recently proposed particle swarm optimization approach.