Gaussian Mixture Model in Dynamic Background of Video Sequences for Human Detection
Bimo Haryo Setyoko, Edi Noersasongko, Guruh Fajar Shidik, Fikri Budiman, Moch Arief Soeleman, Pulung Nurtantio Andono, Pujiono Pujiono · 2022
Background subtraction is the initial key process of human detection. One of the problems faced by background subtraction is various changes or dynamic backgrounds. It can be caused by the illumination or color so the colors of the background and the foreground look similar. To adjust the background changes, a different Gaussian Mixture Model was proposed in this paper. Performance evaluation was conducted qualitatively and quantitatively by comparing GMM -based background subtraction against standard background subtraction. Measurement is carried out using PSNR (Peak Signal Noise Ratio) and MSE (Mean Square Error). The Average MSE of GMM-based was 3,40 and the average PSNR was 11,80. The Average MSE of standard Background Subtraction was 3,98 and the average PSNR was 8,54. The MSE value of GMM-based background subtraction based is lower than the MSE of standard Background Subtraction and the PSNR value is higher. The experiment result proved that GMM-based background subtraction is promising to segment and extract the foreground in detection.