Principal Investigator: Kristy Arbogast, PhD, Children’s Hospital of Philadelphia

Below is an executive summary of this line of research. Please note that each summary describes results and interpretation that may not be final. Final interpretation of results will be in the peer-reviewed literature.

Vehicle and restraint safety devices have largely been optimized through laboratory testing programs using anthropomorphic test devices (ATDs), or crash test dummies, intended to mimic the human occupant. Most testing protocols evaluate restraint performance with ATDs placed in ideal positions (for example, perfectly upright against the seat back) and the majority of restraints perform very well under those conditions. Unfortunately, recent real world evidence has suggested that ideal test conditions do not always reflect actual conditions and despite being seated in the correct restraint system for their age and size, an unacceptable number of children are injured or killed in motor vehicle crashes. A potential solution lies in the study of the naturalistic behaviors of passengers. For child occupants, however, there have been only a few isolated efforts to describe position and postural changes using naturalistic, observational methods.

Previous research has highlighted the tendency for children to move substantially out of position when restrained in a motor vehicle, not only in response to vehicle dynamics, but also of their own volition. Of particular interest are the circumstances that lead to extreme head positions for rear-seated child occupants to identify whether these circumstances are modifiable. In 2012-2013 and 2013-2014 funding years, CChIPS has supported CHOP’s effort in a large multidisciplinary collaboration of engineers and behavioral scientists led by Monash University in Melbourne, Australia to quantify differences between optimal posture and position of child occupants in the rear seat and actual position and posture through a large scale naturalistic study. CChIPS funded the development of the initial methodology to implement a Microsoft Xbox Kinect Sensor™ in the study vehicles to measure head position. This support continued through the 2014-2015 funding year to develop and test several analytical algorithms that leveraged the quantitative details of the Kinect data.

Ongoing support recognizes that this line of research is a critical next step to understand how to protect rear-seat occupants in actual real-world scenarios – those dominated by less-than-ideal occupant positioning. Future work will quantify head position and lead to the development of countermeasures to recognize and correct sub-optimal positions to improve rear-seat occupant protection.

2014-2015

Left: A single image from one time point displays raw depth data from the Kinect system.
The view is from the front seat looking back at three children positioned across the second row.
The blue dot represents the location of the center-seated child’s head as identified by Kinect’s skeletal tracking system. 
Right: This graph shows the fore-aft distance of the center-seated child’s head throughout the trip as logged 
by Kinect’s skeletal tracking system.
 

In Year 3, the team evaluated the built-in algorithm’s ability to collect accurate data on the position of the head on a random sample of 5 percent of trips. Additionally, in part due to the limitations of the skeletal tracking algorithm, a second approach was developed to find images of extreme head motion. This approach compared the depth profile of a given image to a reference image. These images – defined as extreme motion candidates – were reviewed manually to confirm they indeed represented extreme head motion. Once confirmed, the x, y, z coordinates of the head were extracted from the Kinect data.

Naturalistic data collection was completed in November 2014 with 42 families enrolled and approximately 1,200 trips taken using the study vehicle. Skeleton data were recorded on 714 trips (60 percent success rate), and 1.5 million Kinect images in color and 3D format were made available for analysis. The validation study determined that for 41 percent of these images, the skeletal prediction of the head location was accurate. Though less accurate than expected, the size of the dataset suggests a significant amount of skeletal tracking data remains valuable to analyze. Precise 3D accelerometer data from the Kinect sensor was recorded as well on 492 trips, enabling the research team to study the impact of the dynamics of the car on the child’s position and posture.

2012-2013 & 2013-2014

The aims of this project were to develop and test an innovative data collection and analysis method to determine the naturalistic positions of child occupants while restrained in cars. In collaboration with Monash University in Melbourne, Australia, a naturalistic observation study of children in cars will be completed by the end of 2014, with data collected from 48 Melbourne families. Two study vehicles were fitted with a set of discrete video cameras, a recording system, and a vehicle data acquisition unit, providing a comprehensive view of the child occupants. In addition, one of the vehicles was also fitted with an RGB camera and depth sensor to provide 3D motion capture of the rear seat outboard occupants. The researchers leveraged the capabilities of the Microsoft Xbox Kinect Sensor™ for this purpose as a cost-effective alternative to more expensive technology. Participant families were asked to use the study vehicle for all driving trips over a period of two weeks.

At the beginning of the observation period, participants were asked to complete a mandatory Participant Briefing and Training Session    to familiarize themselves with the vehicle and its video-recording equipment. Participants also completed an online survey to gather    basic demographic information, types of child restraint use and general driver information, such as number of years driving, amount of driving experience and crash history. Upon completion of the observation period, participants were asked to complete a second online survey that addressed general child restraint safety and legislation awareness, driver and child travel behaviors, and beliefs regarding child restraints and common vehicle travel.

The primary objective of the data analysis is to quantify how often and to what degree the child occupant is in a non-ideal position. The upcoming third year of this study will see a continuation of naturalistic data collection and analysis of data, as well as sled tests to explore injury consequences of postures observed in the naturalistic study. 

Project Team Members

Helen Loeb, PhD, Children’s Hospital of Philadelphia (Y1,Y2,Y3); Judith Charlton, PhD, Monash University (Y1,Y2,Y3); Katarina Bohman, PhD, SAFER, Chalmers University (Y1,Y2,Y3); Mats Svensson, PhD, SAFER, Chalmers University (Y1,Y2,Y3); Katarina Bohman, PhD, SAFER, Autoliv Research (Y3); Jinyong Kim, PhD, Children’s Hospital of Philadelphia (Y3); Sjaan Koppel, PhD, Monash University (Y3); Byoung-Keon Park, PhD, University of Michigan (Y3); Matthew Reed, PhD, University of Michigan (Y3)

Students

Suzanne Cross, MS, Monash University (Y1,Y2,Y3); Johnny Kuo, MS, Monash University (Y1,Y2,Y3); Isabelle Stockman, MS, Chalmers University (Y1,Y2,Y3); Gretchen Baker, University of Kansas (Y3); Christian Parker, Drexel University (Y3)

IAB Mentors

John Combest, Nissan Technical Center North America Inc. (Y1,Y2,Y3); Schuyler St. Lawrence, Toyota Motor North America Inc. (Y1,Y2,Y3); Kevin Kramer, Minnesota HealthSolutions (Y1,Y2,Y3); Uwe Meissner, Technical Advisor (Y1,Y2,Y3).