DGT 2014
More information


This project has been funded by the Dirección General de Tráfico (DGT) of the government of Spain, with administrative reference SPIP2014-1468.

The main goal of the project is to research into novel computer vision solutions for the problem of precisely estimating the number of vehicles for situations o high congestion in traffic surveillance images.

In this type of pictures, the vehicles normally appear under severe occlusions, therefore the traditional sliding window approaches for object detection do not perform adequately. Instead, counting by regression strategies must be explored.

The objectives of our research are:

  1. Build and release a novel dataset where the research community can follow a clear experimental validation, in order to compare the performance of different approaches for the challenging problem of vehicle counting in high congestion traffic surveillance images.
  2. Develop novel counting by regression solutions for the problem of vehicle density estimation.


Because better detections are still possible: Multi-aspect Object Detection with Boosted Hough Forest

C. Redondo-Cabrera, R. J. López-Sastre.

BMVC, 2015. PDF Supplementary Code

Extremely Overlapping Vehicle Counting

R. Guerrero-Gómez-Olmedo, B. Torre-Jimenez, R. J. López-Sastre, S. Maldonado-Bascón and Daniel Oñoro-Rubio.

IbPRIA, 2015. PDF TRANCOS Dataset

Fast Head Pose Estimation for Human-Computer Interaction

M. García-Montero, C. Redondo-Cabrera, R. López-Sastre and Tinne Tuytelaars.

IbPRIA, 2015. PDF

Data and Software


  • TRANCOS - TRaffic ANd COngestionS - Download





Gram Reseach Group

Funded by