Last update: 05/04/2013
This work proposes a holistic method for highway traffic video classification based on vehicle crowd properties. The method classifies the traffic congestion into three classes: light, medium and heavy. This is done by usage of average crowd density and crowd speed. Firstly, the crowd density is estimated by background subtraction and the crowd speed is performed by pyramidal Kanade-Lucas-Tomasi (KLT) tracker algorithm. The features classification with neural networks show 94.50% of accuracy on experimental results from 254 highway traffic videos of UCSD data set.
Best Paper Award of The 10th IASTED International Conference on Signal Processing, Pattern Recognition and Applications (SPPRA'2013)
For more information, a copy of the full paper can be found here:
http://www.academia.edu/2154120/Highway_Traffic_Congestion_Classification_using_Holistic_Properties
The full dissertation / master thesis can be downloaded here (in brazilian-portuguese):
Classificação automática do estado do trânsito baseada em contexto global
http://repositorio.ufba.br/ri/handle/ri/13300 (PDF ~12.3Mb)
Presentation slides (in brazilian-portuguese):
Bibtex:
@article{asobral2013,
author={Andrews Sobral and Luciano Oliveira and Leizer Schnitman and Felippe De Souza},
journal={10th IASTED International Conference on Signal Processing, Pattern Recognition and Applications (SPPRA'2013)},
title={Highway Traffic Congestion Classification Using Holistic Properties},
year={2013},
month={fev.},
}
@article{asobral2013,
author={Andrews Sobral and Luciano Oliveira and Leizer Schnitman and Felippe De Souza},
journal={10th IASTED International Conference on Signal Processing, Pattern Recognition and Applications (SPPRA'2013)},
title={Highway Traffic Congestion Classification Using Holistic Properties},
year={2013},
month={fev.},
}
SPPRA'2013 Paper Presentation
Example of crowd properties extracted from six videos with three distinct traffic patterns (light, medium and heavy). In this video two holistic properties are extracted: avg. of crowd density and avg. of crowd speed.
The features are classified with a multi-layer perceptron (MLP) using RPROP algorithm for training, and a sigmoid function as an activation function and four hidden neurons.
Normalized features extracted from UCSD traffic videos
MLP evaluation in four test trials varying the training algorithm (TA), activation function (AF) and number of hidden neurons (HN)
Cumulative confusion matrix of the proposed system using MLP network with best configuration.
