Research

 

Intelligent Vehicle Experimental Test-beds

 

Human-Centered Driver Assistance Systems

 

Vision Systems for Occupant Posture Analysis

 

Vision Systems for Driver Analysis and Interface

 

Related Research

 

Real-Time Vision for Occupant Posture Analysis for "Smart" Airbags

According to the National Highway Traffic Safety Administration (NHTSA), in the past 10 years, airbags were deployed more than 3.3 million times in the United States. Airbags are credited with saving more than 6,000 lives and preventing a much greater number of serious injuries. These numbers clearly highlight the life saving attributes of airbag technology. Alas, there are other rather disheartening numbers that are presented in the same report. It states that since 1990, over 200 fatalities have been recorded as a direct result of an airbag deployment. The majority of these deaths have been children. The number of severe injuries is much higher. Obviously, these deaths and injuries must be prevented by exploring and adopting the most appropriate technologies, policies and practices.

A new law, which will go into effect in the United States in 2004 requires that airbag systems be able to distinguish persons too small or persons in unsafe positions from persons of a safe size and in safe position for airbag deployment. One of the main difficulties encountered by the decision logic systems used in airbag deployment deals with the critical assumption abut the occupant size and position in the car at the time of a crash. Most airbag systems consider a single standard for the occupant's size and the nature of the crash. Vision based technology enables the use of precise information about the occupant's size, position, and posture to aid the single standard airbag system in deciding whether the occupant is of the right type for deployment.

Our primary research objective is to describe the design, implementation and evaluation of vision-based occupant posture analysis systems to control the deployment of an airbag in a safe manner. These efforts resulted in the development of a framework for sensing the most relevant visual information, a set of robust and efficient algorithms for extracting features which characterize the size, position and posture of the occupant, and a pattern recognition module to classify the visual cues into categories, which can trigger the safe deployment logic of the airbag system.

To this end, we are investigating the viability of three different occupant posture analysis systems, each utilizing different cameras setups and extractable features. These systems include using thermal infrared imaging, stereo and reflectance imaging, and shape-from-silhouette based voxel reconstruction methods.

Thermal Long-wavelength Infrared Imaging

Stereo and Reflectance Imaging

Shape-from-Silhouette Voxel Reconstruction

Comparative Performance Analysis

Related Publications

Video Gallery