Tutorial 5 : Spatio‐temporal filtering for multi‐object tracking in image sequences

Presenters:

Prof. Andrea Cavallaro, Queen Mary University of London
Dr. Emilio Maggio, Vicon, UK

Abstract

This proposed ICIP 2011 tutorial will cover the fundamental aspects of spatio-temporal filtering for multiobject tracking in image sequences. The tutorial sets forth the state-of-the-art in object detection and representation, data association and random finite sets for video tracking. The tutorial will discuss and demonstrate the latest multi-target tracking algorithms with a unified and comprehensive coverage.

Using practical examples and illustration as support, we will introduce the participants in a discussion of the advantages and the limitations of traditional and modern approaches and we will guide them toward more efficient and accurate multi-target tracking algorithms, in particular those based on the Probabilistic Hypothesis Density Filter, a recent advance in image-based tracking that allows to reduce the computation complexity for multiple target tracking as well as to incorporate contextual knowledge to improve the efficiency and robustness of the filter.

Multiple-target tracking applications will be discussed in real-world tracking scenarios. We will conclude the tutorial by discussing tracking evaluation methodologies and introducing a collection of software resources and publicly available datasets to help the attendees develop and test multi-object trackers.

Outline

The tutorial material will be organized according to the proposed syllabus below. The tutorial slides will be provided as PDF for inclusion in the course distribution material. A website will be developed for the course, which will contain links to supporting material and video segments that enrich the learning experience of the participants.

Part 1 - Introduction Tracking in image sequences: definitions and problem formulation Applications Multi-target management: challenges

Part B – Background Introduction to Bayesian tracking

• Dynamic and observation models
• Kalman filter
• Monte Carlo approximation

Object detection and representation

Part C – Multi-target algorithms Measurement validation
Data association

• Nearest Neighbour
• Graph matching
• Multiple Hypothesis Tracking

Random Finite Sets for tracking
Probabilistic Hypothesis Density filter
The Particle PHD filter

• Dynamic and observation models
• Birth and clutter models
• Importance sampling and resampling

Tracking with context modelling

• Contextual information
• ''Influence of the context on the filter

Birth and clutter intensity estimation
Tracking with contextual feedback

Part D - Performance evaluation Analytical vs. empirical methods
Evaluation scores, protocols and datasets
Comparing multi-target trackers

• Target life-span
• Statistical significance
• Repeatability

Part E – Open problems and research outlook