Computer Vision And Pattern Analysis Laboratory Home Page  Home
People  People
Publications  Publications
Publications  Databases
Contact Information  Contact
Supported Research Projects  Supported Research Projects
Research Activites  Research Activites
Research Groups
SPIS - Signal Processing and Information Systems Lab.SPIS - Signal Processing and Information Systems Lab.
Medical Vision and Analysis Group  Medical Research Activities
Biometrics Research Group  Biometrics Research Group
SPIS - Signal Processing and Information Systems Lab.MISAM - Machine Intelligence for Speech Audio and Multimedia.
Knowledge Base
  Paper Library
  VPA Lab Inventory
  Databases in VPALAB
Video-based driver identification using local appearance face recognition
Authors: Stallkamp, J., Ekenel, H. K., Erdogan, H., Stiefelhagen, R. and Ercil, A.
Published in: Biennial on DSP for in-Vehicle and Mobile Systems
Publication year: 2007
Abstract: In this paper, we present a person identification system for vehicular environments. The proposed system uses face images of the driver and utilizes local appearance-based face recognition over the video sequence. To perform local appearance-based face recognition, the input face image is decomposed into non-overlapping blocks and on each local block, discrete cosine transform is applied to extract the local features. The extracted local features are then combined to construct the overall feature vector. This process is repeated for each video frame. The distributions of the feature vectors over the video sequence are modeled using a Gaussian distribution function at the training stage. During testing, the feature vector extracted from each frame is compared to each person’s distribution, and individual likelihood scores are generated. Finally, the person is identified as the one who has maximum joint-likelihood score over the whole video sequence. To assess the performance of the developed system, extensive experiments are conducted on different identification scenarios, such as closed-set identification, open-set identification and verification. For the experiments a subset of the CIAIR-HCC database, an in-vehicle data corpus that is collected at the Nagoya University, Japan is used. We show that, despite varying environment and illumination conditions, that commonly exist in vehicular environments, it is possible to identify individuals robustly from their face images.
  download full paper

Home Back