TR2004-130

A Hidden Markov Model Framework for Traffic Event Detection Using Video Features


    •  Li, X.; Porikli, F.M., "A Hidden Markov Model Framework for Traffic Event Detection Using Video Features", IEEE International Conference on Image Processing (ICIP), ISSN: 1522-4880, October 2004, vol. 5, pp. 2901-2904.
      BibTeX Download PDF
      • @inproceedings{Li2004oct,
      • author = {Li, X. and Porikli, F.M.},
      • title = {A Hidden Markov Model Framework for Traffic Event Detection Using Video Features},
      • booktitle = {IEEE International Conference on Image Processing (ICIP)},
      • year = 2004,
      • volume = 5,
      • pages = {2901--2904},
      • month = oct,
      • issn = {1522-4880},
      • url = {http://www.merl.com/publications/TR2004-130}
      • }
  • Research Areas:

    Computer Vision, Machine Learning


We present a novel approach for highway traffic event detection. Our algorithm extracts features directly from the compressed video and automatically detects traffic events using a Gaussian Mixture Hidden Markov Model (GMHMM) framework. First, a feature vector is computed for a group of picture from the Discrete Cosine transform (DCT) coefficients and macro-block motion vectors after MPEG video bitstream is parsed. We show that the feature vector is robust towards different camera setups and illumination conditions such as sunny, overcast, dark, night, etc. Then, we use Viterbi algorithm to determine the most likely traffic condition. We define six traffic patterns, and each pattern is modeled by a separate GMHMM that is trained using the EM algorithm. The proposed system is efficient both in terms of computational complexity and memory requirement. The experimental results prove that the system has a high detection rate. The presented model-based system can be easily extended for detection of similar traffic events.