ECCV 2016 Daily - Friday

Yoni Kasten prepared this project together with Shmuel Peleg and Michael Werman who are professors at the Hebrew University along with Gil Ben-Artzi , the additional author. The work that they presented finds the fundamental matrix between two views with wide angle when they recorded videos of multiple objects moving inside. The main novelty of this work, with respect to previous work, is that this work can deal with multiple objects. Previous methods could deal with one object moving in such a video scenario in cases where cameras have a wide angle between them. When using cameras without wide angles, you can use feature points in order to find the fundamental matrix. If the angle is wide, you cannot do that because the background is very different. However, you can use the motion in the video sequences in order to find the fundamental matrix using the convex hull of the foreground object in each view, which can not be used when there are multiple objects in the scene. This model has very important and specific practical applications. For instance, there are surveillance cameras that recorded the same scene and the cameras can be calibrated using this method. The model has been tested on both synthetic and real datasets including PETS 2009, which is a very challenging dataset. Fundamental matrices were found with high accuracy. The method is based on line motion barcodes, descriptors for lines in a video sequence which are insensitive to the wide angles of cameras. Follow the link to this video to better understand what motion barcode is. Fundamental Matrices fromMoving Objects Using Line Motion Barcodes 7 ECCV Daily: Fr iday Yoni Kasten is a student at Hebrew University of Jerusalem having just completed his Master’s degree.