Keynote Speaker at SBMI2015

Pierre-Marc Jodoin, Motion Detection: Unsolved issues and [Potential] Solutions

Abstract. Motion detection has long been used in video analytics to detect moving objects. Although a plethora of papers have been published in that area, almost every method is based on the same underlying assumptions. These assumptions are that a video always comes with a decently high framerate, that it was shot by a fix camera, and that it shows moving objects made of colors (or texture) different from those in the background. Although these assumptions have their reasons to be, since almost every method relies on it, they often stumble on the same issues. It is thus unlikely that a method based on traditional approaches (KNN, GMM, KDE, PCA, etc.) will lead to a major breakthrough in regards to those issues. In this presentation, I will review unsolved motion detection problems and potential solutions to it. I will propose a slightly new motion detection definition, I will discuss applications that go beyond the traditional scope of motion detection (UAVs, ultra-low framerate videos, etc.), datasets that shall be proposed in order to validate future methods, and various ideas related to recent machine learning breakthroughs.

is associate professor at the computer science department of the University of Sherbrooke, Canada. He specialized in medical imaging, video analytics, image processing and computer vision. In 2011, he co-founded a company called Imeka, that develops software for the medical imaging community. He is also the co-director of the image processing and visualisation service at the Centre Hospitalier de l'Université de Sherbrooke (plateforme d'analyse et de visualisation d'images) and director of the research center for intelligent environments. He is associate editor of IEEE transactions on image processing, and guest editor of Pattern Recognition and Signal Processing. Pierre-Marc Jodoin organized the 2012 and 2014 editions of the IEEE CVPR Workshop on Change Detection and founded the changedetection.net initiative, the largest change detection benchmarking dataset in the world.