Vision stéréoscopique par ordinateur pour la détection et le suivi de cibles pour une application automobile

Julien Morat 1
1 PERCEPTION - Interpretation and Modelling of Images and Videos
Inria Grenoble - Rhône-Alpes, LJK - Laboratoire Jean Kuntzmann, INPG - Institut National Polytechnique de Grenoble
Abstract : The increasing complexity of the driving environment and the efforts
for improving safety in the roads, explain the interest car
manufacturers put on the development of automatic driver assistance
technology. Many vehicles available on the market are already equipped
with such systems. While we have become proficient in judging the
state of the vehicle (speed, position, etc.), judging the surrounding
environment remains a difficult task. Among all the sensors capable of
perceiving the complexity of an urban environment, stereo-vision
has interesting capabilities, a large spectrum of applications
(pedestrian detection, vehicle tracking, white-line detection, etc.)
and a competitive price. For these reasons, Renault aims to identify
and solve the problems related to the implementation of such a system
in a vehicle series. The first related issue concerns the
development of the stereo-system calibration. Indeed the capability of
the system to provide measurements depends on parameters being
properly estimated, even when the system undergoes extreme conditions
(high temperatures, shock, vibration ...). Therefore, we present an
evaluation methodology that measures the degradation of the system's
performance with respect to the calibration. The second problem deals
with the detection of obstacles. The method we propose uses the
properties of rectification in an original fashion, and segments the
scene in road and obstacles. The last issue relates to the calculation
of the obstacles' speed. A large majority of approaches in the
literature approximate the speed of an obstacle from its successive
positions. In this calculation, the accumulation of uncertainties
makes this estimate extremely noisy. Our approach effectively combines
the strengths of the stereo-vision and optical flows to directly
obtain a robust and precise measure of 3-D speed.
Document type :
Theses
Complete list of metadatas

Cited literature [68 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00343675
Contributor : Frédéric Devernay <>
Submitted on : Tuesday, December 2, 2008 - 3:18:10 PM
Last modification on : Wednesday, April 11, 2018 - 1:58:41 AM
Long-term archiving on : Monday, June 7, 2010 - 9:32:19 PM

Identifiers

  • HAL Id : tel-00343675, version 1

Collections

Citation

Julien Morat. Vision stéréoscopique par ordinateur pour la détection et le suivi de cibles pour une application automobile. Interface homme-machine [cs.HC]. Institut National Polytechnique de Grenoble - INPG, 2008. Français. ⟨tel-00343675⟩

Share

Metrics

Record views

1011

Files downloads

6817