A Robust Online Multi-Face Tracking System

Abstract

This paper presents a system for robust online multi-face tracking in video sequences recorded by a stationary camera. Several components and algorithms are combined in the system architecture: an NPD robust face detector, a DSST tracker augmented with FIFO long- and short-term memories (LTMs/STMs), trajectory memories (TMs), a PSR-based tracking failure detector and a Hungarian algorithm for trajectory assignment. The paper gives the preliminary experimental results, expressed by MOTA, MOTP and IDS metrics, obtained on the benchmark dataset consisting of three videos.