Moving Object Detection based on RGBD Information

Abstract

This thesis is targeting the Moving Object Detection topic, more specifically, the Background Subtraction. In this study, we proposed two approaches using color and depth information to solve the background subtraction. The following two paragraphs will give a brief abstract for each approach. In this research study, we propose a framework for improving traditional Background Subtraction techniques. This framework is based on two data types: color and depth; it stands for obtaining preliminary results of the background segmentation using Depth and RGB channels independently, then using an algorithm to fuse them to create the final results. The experiments on the SBM-RGBD dataset using four methods: ViBe, LOBSTER, SuBSENSE, and PAWCS, proved that the proposed framework achieves an impressive performance compared to the original RGB-based techniques from the state-of-the-art. This dissertation also proposes a novel deep learning model called Deep Multi-Scale Network (DMSN) for Background Subtraction. This convolutional neural network is built to use RGB color channels and Depth maps as inputs with which it can fuse semantic and spatial information. Compared with previous Deep Learning Background Subtraction techniques that lack information due to their use of only RGB channels, our RGBD version can overcome most of the drawbacks, especially in some particular challenges. Further, this study introduces a new protocol for the SBM-RGBD dataset regarding scene-independent evaluation, dedicated to Deep Learning methods to set up a competitive platform that includes more challenging situations. The proposed method proved its efficiency in solving the background subtraction in complex problems at different levels. The experimental results verify that the proposed work outperforms the state-of-the-art on SBM-RGBD and GSM datasets.