Head motion tracking in 3D space for drivers

Authors: Ntawiniga, Frédéric
Advisor: Laurendeau, Denis
Abstract: This work presents a computer vision module capable of tracking the head motion in 3D space for drivers. This computer vision module was designed to be part of an integrated system to analyze the behaviour of the drivers by replacing costly equipments and accessories that track the head of a driver but are often cumbersome for the user. The vision module operates in five stages: image acquisition, head detection, facial features extraction, facial features detection, and 3D reconstruction of the facial features that are being tracked. Firstly, in the image acquisition stage, two synchronized monochromatic cameras are used to set up a stereoscopic system that will later make the 3D reconstruction of the head simpler. Secondly the driver’s head is detected to reduce the size of the search space for finding facial features. Thirdly, after obtaining a pair of images from the two cameras, the facial features extraction stage follows by combining image processing algorithms and epipolar geometry to track the chosen features that, in our case, consist of the two eyes and the tip of the nose. Fourthly, in a detection stage, the 2D tracking results are consolidated by combining a neural network algorithm and the geometry of the human face to discriminate erroneous results. Finally, in the last stage, the 3D model of the head is reconstructed from the 2D tracking results (e.g. tracking performed in each image independently) and calibration of the stereo pair. In addition 3D measurements according to the six axes of motion known as degrees of freedom of the head (longitudinal, vertical and lateral, roll, pitch and yaw) are obtained. The validation of the results is carried out by running our algorithms on pre-recorded video sequences of drivers using a driving simulator in order to obtain 3D measurements to be compared later with the 3D measurements provided by a motion tracking device installed on the driver’s head.
Document Type: Mémoire de maîtrise
Issue Date: 2008
Open Access Date: 13 April 2018
Permalink: http://hdl.handle.net/20.500.11794/19765
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
Description SizeFormat 
25229.pdfTexte3.44 MBAdobe PDFThumbnail
All documents in CorpusUL are protected by Copyright Act of Canada.