Mapping and Assessing the State of Traffic InFrastructure

Demonstrations

During the work on the project we have created several demonstration videos illustrating the principal achievements. Most of these videos are already available from the results page. and we summarize them here in order to improve their visibility.
 
  1. Detection and recognition of triangular warning signs. Excellent performance has been achieved of 100% recall and near 100% precision. The developed prototype system comprises:
  • boosted Haar cascade detection [ITSC10],
  • filtering detection responses with a suitably trained strong classifier [CVWW11],
  • enforcing temporal consistence of the detection responses [submitted],
  • recognition by employing LDA+1NN [ITSC10].
The recognition results would be considerably better if we used a properly conditioned training set [CVWW11]. The number of false positive detections would be even less if we employed an additional filter based on our discriminative trajectory model [submitted]. 		Triangles demo 2010-11
 
  2. Creating orthogonal road appearance mosaics from georeferenced perspective images taken from the driver's perspective. The developed prototype system comprises:
  • inverse perspective transformation for images acquired by a camera mounted on a moving vehicle [MIPRO09],
  • cubic interpolation of positional readings provided by a standard GPS receiver [MIPRO11],
  • stitching individual inverse perspective images into the comprehensive mosaic [MIPRO11].
Mosaic demo 2011-03
 
  3. Producing spatio-temporal appearance descriptors (STA). The developed prototype system comprises:
  • calculating HOG features in subwindows of the tracked object [CVWW11],
  • determining STA descriptors (1st and 2nd order) [SCIA11].
STA demo 2011-04
 
  4. Detection, tracking and recognition of the centerline road surface marking. The developed prototype system comprises:
  • inverse perspective transformation for images acquired by a camera mounted on a moving vehicle [MIPRO09],
  • detection of road surface lines based on steerable filter response [MIPRO09],
  • robust parameter estimation for the parabolic centerline model [MIPRO11].
Recent developments resulted in improved performance with respect to what we reported in [MIPRO11]. 		Centerline demo 2010-12