High-performance template tracking

Raúl Cabido, Antonio S. Montemayor, Juan José Pantrigo, Mario M. Zarzuela, Bryson R. Payne

Index

• Abstract
• Hypothesis and improvements
• Kernel-based particle filter for visual tracking
• GPU Platform
• Demo videos
• Binaries/Source code
• Links/Other resources
• About
• Authors
• CAPO
• Acknowledgements

Abstract

Tracking systems are important in computer vision, with applications in video surveillance, human computer interfaces (HCI), etc. Consumer graphics processing units (GPUs) have experienced an extraordinary evolution in both computing performance and programmability, leading to a greater use of the GPU for non-rendering applications, such as image processing and computer vision tasks. In this work we show an effective particle filtering implementation for real-time template tracking based on the use of a graphics card as a streaming architecture in a translation-rotation-scale model.

---

Hypothesis and improvements

Our hypothesis about the kernel-based particle filter algorithm and the GPU hardware proposals are based on some foundations:

• Consumer hardware is evolving from high frequency single-core machines to multi-core processors: need for parallel approaches
• Graphics processing unit (GPU) is an example of a future architecture: many-core architecture, high throughput, high memory bandwidth, low cost
• PF is algorithmically very suitable for GPU hardware (independency, pixel-based approach, computationally intensive,...)

Kernel-based particle filter for visual tracking

• It may be used with fixed appearance models (we could also use dynamic models)
• We may enhance features with geometric templates while weighting particles
• Workload reduction enables real time performance and stable tracking

GPU Platform

• Very good performance (2-10 times faster than CPU)
• Consumer parallel platform: advanced and suitable for data-parallel processing
• Not expensive but high performance hardware (more than 1000 M transistors), free and high level programming frameworks
• Very rapid evolution hardware, almost doubling performance each 9 months

---

Demo Videos

Not shown in the paper. Caviar video benchmark. Comparison sequence showing our proposal against ground truth and a standard particle filter (.avi)	Not shown in the paper. Caviar video benchmark. Another comparison sequence showing our proposal against ground truth and a standard particle filter (.avi)
Fig. 10. Synthetic sequence comparing our proposal against ground truth and a standard particle filter (.avi)	Fig. 12. Face tracking using a Haar-like template embeded in a particle filter (.wmv)
Fig. 13. Comparison between the Haar-like template tracking and particle filter using a fixed appearance model (.wmv)	Fig. 14. Multiview application for pose recognition using templates (.wmv)

---

Binaries/Source code

• Binaries are now available. Before running the application, please read the readme.txt file. Note that video sequences are also included.


• Source code (Visual Studio 2008 solution) to make the performance results reproducible.

---

Links/Other resources

• First proposal of particle filtering partially using the GPU: Antonio S. Montemayor, Juan José Pantrigo, Ángel Sánchez, Felipe Fernández. Particle Filter on GPUs for Real-Time Tracking, Proc. of the ACM SIGGRAPH 2004 (research poster)
• First and complete particle filtering using the GPU: Antonio S. Montemayor, Juan José Pantrigo, Raúl Cabido, Bryson Payne, Ángel Sánchez, Felipe Fernández. Improving GPU Particle Filter with Shader Model 3.0 for Visual Tracking, Proc. of the ACM SIGGRAPH 2006 (research poster).

---

About

Authors

Antonio S. Montemayor: Associate Professor at Universidad Rey Juan Carlos. The title of his Ph.D. was: Optimización de Algoritmos de Procesamiento de Vídeo para su Implementación sobre Tarjetas Gráficas de Consumo (Video Processing Algorithmic Optimization for Implementation on Consumer Graphics Cards). Personal Web: http://www.etsii.urjc.es/~asanz/.

Bryson R. Payne: Associate Professor of Computer Science at North Georgia College & State University. Ph.D. in Computer Science from Georgia State University entitled Accelerating Scientific Computation in Bioinformatics by Using Graphics Processing Units as Parallel Vector Processors. Personal Web: http://www.professorpayne.com/.

Juan José Pantrigo: Ph.D. in Computer Science and Associate Professor at Universidad Rey Juan Carlos. His main interests include Computer Vision and Optimization Techniques. Personal Web: http://www.etsii.urjc.es/~jjpantrigo/

Mario M. Zarzuela: Ph.D. in Computer Science and Assistant Professor at Universidad de Valladolid. His main interests are GPU Programming and Computer Vision, including face tracking techniques and neural netwoks in image processing.

Raúl Cabido: Ph.D. Student and grant holder at Universidad Rey Juan Carlos. His main interests are GPU Programming, Video Processing and Computer Vision.

CAPO

CAPO is the Spanish acronym for High-Performance Computing and Optimization (Computación de Altas Prestaciones y Optimización). CAPO is one of the research lines of the GAVAB group at Universidad Rey Juan Carlos (Madrid, Spain).

---

Acknowledgements

 

We would like to thank the Spanish Ministry of Education and Science that has been supported this research by CICYT TIN2008-06890-C02-02, URJC and CAM by URJC-CM-2008-CET-3625, and the Nvidia Professor Partnership Program