A Quantitative Analysis on Two RFS-Based Filtering Methods for Multicell Tracking

Yayun Ren,Benlian Xu

doi:10.1155/2014/495765

Abstract

Multiobject filters developed from the theory of random finite sets (RFS) have recently become well-known methods for solving multiobject tracking problem. In this paper, we present two RFS-based filtering methods, Gaussian mixture probability hypothesis density (GM-PHD) filter and multi-Bernoulli filter, to quantitatively analyze their performance on tracking multiple cells in a series of low-contrast image sequences. The GM-PHD filter, under linear Gaussian assumptions on the cell dynamics and birth process, applies the PHD recursion to propagate the posterior intensity in an analytic form, while the multi-Bernoulli filter estimates the multitarget posterior density through propagating the parameters of a multi-Bernoulli RFS that approximates the posterior density of multitarget RFS. Numerous performance comparisons between the two RFS-based methods are carried out on two real cell images sequences and demonstrate that both yield satisfactory results that are in good agreement with manual tracking method.

Highlights

In previous related research in the field of medicine and biology, people often use manual visual inspection to observe living cells behaviors, such as the moving velocity and the density of cell population
We present two random finite sets (RFS)-based filtering methods, Gaussian mixture probability hypothesis density (GM-probabilistic hypothesis density (PHD)) filter and multi-Bernoulli filter, to quantitatively analyze their performance on tracking multiple cells in a series of low-contrast image sequences
label switching rate (LSR) is the number of label switching events normalized over total number of ground truth tracks crossing events, while lost tracks ratio (LTR) is the number of tracks lost for more than 50% of their lifetime normalized over total number of ground truth tracks

Summary

Introduction

In previous related research in the field of medicine and biology, people often use manual visual inspection to observe living cells behaviors, such as the moving velocity and the density of cell population. A sequential Monte Carlo (SMC) implementation of multi-Bernoulli filter is presented in [8], called “track-before-detect” technique This method by passes the detection module and makes use of the spatiotemporal information directly extracted from poor image sequences. It demonstrates very good performance in cell tracking application. Stochastic methods generally make better use of the spatiotemporal information than deterministic methods and yield more robust tracking outcomes especially for poor cell image data These cell tracking methods introduced above have their own properties and show different performance during implementation. This paper first quantitatively analyzes and compares the performance of GM-PHD and multi-Bernoulli filters [8, 9] in tracking multiple cells in a series of lowcontrast image sequences.

Background on RFS Model

Methods

Implementation Issues

Experiments and Discussions

Findings

Conclusions