Detection of moving objects using adaptive multi-feature histograms

Abstract

This paper presents a moving object detection scheme that incorporates three innovations. First, considering the inter-frame consistency of pixels, we extend the compact binary face descriptor (CBFD) to the temporal domain and propose a novel local binary descriptor named temporally-consistent local compact binary descriptor (TC-LCBD), which exploits the useful correlation of the intensities of inter-frame pixels to guarantee good performance in complex scenes. We do this mainly because the background scene between frames has a significant coherence. Second, both color and TC-LCBD features are modeled as a group of adaptive histograms for characterizing each pixel, which can enhance the robustness to dynamic backgrounds and illumination changes. Third, by comparing changes in histogram proximity between two adjacent frames, we can dynamically adjust the model sensitivity and adaptation rate without user intervention. Experimental results on well-known, challenging data sets demonstrate that the proposed method significantly outperforms many state-of-the-art methods.