Best Presentation Award
Hamdi Dibeklioglu won the best presentation award at the CVPR 2014 Workshop on Long-Term Detection and Tracking. Congratulations Hamdi!
Pedestrian Tracking for Safer Cars
Master student Madalin Dumitru-Guzu defended his thesis cum laude. His thesis describes a new method for tracking pedestrians and estimating their orientation in video streams from cameras in recent Mercedes cars. A paper on his joint work with Fabian Flohr, Julian Kooij, and Dariu Gavrila (Daimler R&D) was published in IEEE Intelligent Vehicles Symposium. The Master thesis is available here.
Making Sliding-Window Trackers Faster
Our paper on speeding up trackers and detectors based on sliding-window search was accepted for publication at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) in Columbus OH. The paper describes an algorithm describes a search algorithm that discards locations that are unlikely to contain an object early on with the help of a probabilistic bound. The algorithm allows us to reduce the number of inspected features by approximately 85% without increasing the error of the object tracker / detector.
Dynamic Kinship Verification
In December 2013, Hamdi Dibeklioglu presented his paper titled "Like Father, Like Son: Facial Expression Dynamics for Kinship Verification" at the International Conference on Computer Vision (ICCV) in Sydney, Australia. The paper describes how to use facial expression dynamics for kinship verification.
In June, Laurens van der Maaten gave a Techtalk at Google in Mountain View about his work on visualizing high-dimensional data. A video of the talk is now available on the Google Techtalk Youtube channel.
Structure-Preserving Object Tracker
Our paper on the structure-preserving object tracker (SPOT) was accepted for an oral presentation at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). An extended version of the paper will soon appear in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI). The paper describes a new tracker that can simultaneously track multiple objects based on just a single annotation of those objects. Our tracker outperforms the current state-of-the-art by incorporating the spatial relations between the different objects under consideration in the tracker. An extended version of this work will soon appear in IEEE Transactions on Pattern Analysis and Machine Intelligence. For more details, see this page. For more information, please contact Lu Zhang or Laurens van der Maaten.