Project information
- Problem: Object Detection
- ML Areas: Image Processing and Computer Vision
- Learning technique: Unsupervised Learning
- Tools: Python, OpenCV, Git
- Project date: January 2019 - June 2019
- Project URL: Background Subtraction project
Description
Background Subtraction is a Computer Vision problem of understanding and concretely detecting what is background in a scene, clean from any kind of foreground of that scene. It is applied to a video as input and it suggests frame by frame the detected background. Gaussian Mixture Models (GMM) are able to provide a good solution for the background subtractrion problem: for each frame of a video, the GMM can reproduce the background behind the foreground.
Goal: Improvement of the GMM algorithm used for background subtraction, in order to overcome its limits: GMM is not enough adaptive to the change of natural light.
Technical Solutions: AGM (Adaptive Gaussian Model): it is a model, based on the GMM, but makes it adaptive to the change of natural lights (sunrise, sunset).
Results:
