Motion Segmentation Based on 3D Histogram and Temporal Mode Selection

Abstract

Motion segmentation has been a well explored research topic due to its vast application area. This work proposes a real-time motion segmentation method based on 3D histogram and temporal mode selection. The temporal distribution of a video sequence consists of the motion in the foreground and the relatively immobile background. A 3D histogram provides a short-term memory of the aforementioned distribution. The temporal mode selection process involves identifying the most frequent values in the distribution and construct the background thereafter. This work provides a detailed analysis of the proposed method along with an easy-to-implement algorithm. A number of experimental results and comparisons with some of the leading algorithms are provided to show that the proposed method can provide real-time, robust and highly accurate results.