Non parametric tool for vision detection analysis

Abstract

In this work, we deal with the problem of moving object detection using a non-parametric tool represented by the Gaussian process for classification. The technique used relies on the background subtraction approach for motion detection. In this context, a segmentation step is first implemented for pixel clustering before a binary Gaussian process classifier is applied to determine which pixel cluster those of news images belong to. The unclassified pixels are, therefore, labelled as detected targets. This proposed approach enables motion detection to be completed in a comparatively a short execution time with acceptable results. The results outlined here show the effectiveness of the approach to known background subtraction methods.