City Research Online

Abstract

The aim of this thesis is to explore strategies for real-time image segmentation of non-rigid objects in a spatio-temporal domain with a stationary camera within an optical high dynamic range environment. Camera, illumination and segmentation techniques are discussed for image processing in environments which are characterized by large intensity fluctuations and hence a high optical dynamic range (HDR), in particular for vehicle interior surveillance.

Since the introduction of the airbag in 1981 numberless lives were saved and bad injuries were avoided. But in recent years the airbag has frequently been in the headlines due to the increasing number of injuries caused by it. To avoid these injuries a new generation of ’smart airbags’ has been designed which shows the ability to inflate in multiple steps and with different volumes. In order to determine the optimal inflation mode for a crash it is necessary to consider information about the interior situation and the occupants of the vehicle. This thesis presents a real-time visual occupant detection and classification system for advanced airbag deployment, utilizing a custom CMOS camera and motion based image segmentation algorithms for embedded systems under adverse illumination conditions.

A novel illumination method is presented which combines a set of images flashed with different radiant intensities, which significantly simplifies image segmentation in HDR environments. With a constant exposure time for the imager a single image can be produced with a compressed dynamic range and a simultaneously reduced offset. This makes it possible to capture a vehicle interior under adverse light conditions without using high dynamic range cameras and without losing image detail. The expansion of this active illumination experiment leads to a novel shadow detection and removal technique that produces a shadow-free scene by simulating an artificial infinite illuminant plane over the held of view. Finally a shadowless image without loss of texture details is obtained without any region extraction phase.

Furthermore, a texture based segmentation approach for stationary cam-eras is presented which is neither effected by sudden illumination changes nor by shadow effects.

Publication Type:	Thesis (Doctoral)
Subjects:	Q Science > Q Science (General) T Technology > T Technology (General)
Departments:	School of Science & Technology School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads