Download PDFOpen PDF in browser

Automatic Calibration ‘In the Wild’ of Monocular Surveillance Camera Solely from Pedestrian Height

EasyChair Preprint 3990

4 pagesDate: August 2, 2020

Abstract

We propose a method that relies solely on pedestrian detections in the wild (no known fiducial points) and uses their average height as measuring stick to calibrate monocular surveillance systems. The core idea is that pedestrians feet and head are aligned vertically and the calibration parameters must therefore predict back-projected positions of the feet and head that have the same horizontal coordinates. The advantages are that it does not require any fiducial points; nor to assume that the motion of pedestrians is aligned with vanishing points in the scene; nor relies on a stable framerate as it works on static images. The results are promising: the predicted world positions of pedestrians minimise the vertical misalignment of head and feet and they with the truth positions as well as expected. The approach is very simple and the results encourage us to develop it further with a bayesian approach.

Keyphrases: automatic, calibration, imaging, pedestrian detection, projection matrix, surveillance camera

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
@booklet{EasyChair:3990,
  author    = {Martín E. Oviedo and Sebastián I. Arroyo},
  title     = {Automatic Calibration ‘In the Wild’ of Monocular Surveillance Camera Solely from Pedestrian Height},
  howpublished = {EasyChair Preprint 3990},
  year      = {EasyChair, 2020}}
Download PDFOpen PDF in browser