We have developed a statistical head shape model accounting for variations in face and scalp shapes for both male and female. ~300 bald head scans were analyzed using PCA and standard anthropometric parameters were associated with the retained PC scores to generate subject-specific head model. 58 anatomical landmark locations were carefully defined and digitized from the scans.
This is the finest quality of the head model compared to available head models to date. Possible applications would be scalp shape estimation under hair, facial surgery planning, 3D scan anonymization, and face recognition/classification. Please contact me if you have any questions or are interested in collaboration on this. Check this new model out at here: http://humanshape.org/head/
I presented the Inscribed Fitting method at PASS 2018 conference. The inscribed fitting is a method to efficiently estimate body characteristics for individuals wearing a wide range of ensembles, including protective equipment. The inscribed fitting method utilizes a statistical body shape model that is based on hundreds of laser scans from minimally clad individuals with a wide range of body size and shape. This method uses an iterative process to estimate the body shape underlying the body armor, based on the observation that the correct body shape can be approximated by largest feasible body shape that does not protrude through the clothed scan. The results demonstrate that body characteristics and armor fit can be estimated with good accuracy without the additional steps of scanning in a minimally clad condition and superimposing minimal and equipped scans.
The first augmented reality (AR) work using one of our statistical body shape model was presented at 2018 20th Congress International Ergonomics Association. Apple ARKit 1.5 and Unity 3D were used for implementing the concept of this AR vehicle accommodation assessment. Accommodation and body-seat interaction were assessed through a digital body shape model placed in a physical mockup. Detailed information can be found in this paper: Accommodation Assessments for Vehicle Occupants Using Augmented Reality
I presented our recent study on a method to capture occupant dynamics using a Kinect sensor at SAE Government/Industry 2018. This presentation, entitled "Toward Integrated Safety: Occupant Dynamics in Crash Avoidance Maneuvers", addresses the possibility of using the method to bridge the gaps between current active safety and passive safety systems by allowing to measure pre-crash postures in crash avoidance maneuvers. Result showed that the method we developed using a single depth camera system works really well for capturing the head movement and dynamics. [Youtube Video]
Figure. Depth+color data and a tracked 3D head model (blue) in vehicle during an abrupt turn and brake
I gave a presentation about "Monitoring Postures in Work Environments Using Multiple Kinect Sensors" at AHFE 2017. In this study, a posture monitoring system was developed to quantify and classify workers’ postures in an office setting while taking advantage of the Kinect’s posture-tracking capability. Although a previous monitoring system manually coded postures from recorded videos, this system automatically gathers and quantifies worker posture data in real time using Kinect sensors. Multiple sensors were used in the system to gather the Kinect data in different views through a network. A custom calibration method was developed for registering posture data from multiple sensors. This system monitors up to 4 people per sensor and recognizes basic postures such as sitting, standing and walking. Transitions between the target postures are detected using the velocity of tracked posture data. During sitting, torso recline angles are estimated by projecting the joint locations of head and torso to a reference plane that was defined in the calibration step. Predefined information about the environment can be effectively used for quantifying workers’ positions during monitoring.
A new paper on our work modeling child body shape has appeared in Traffic Injury Prevention representing the development of the first-ever statistical body shape model of seated children. The model, which is available online for interactive use, predicts body shape for children ages 3 to 11 years as a function of stature, body weight (expressed as body mass index), and the ratio of sitting height to stature. The posture of the predicted body shape can also be varied with respect to recline angle and lumbar spine flexion. Whole-body laser scans from 135 children in up to 4 postures were used to create the model. Applications include the development of new physical human surrogates for safety system design (e.g., crash test dummies) and parametric human body models. The downloadable body shapes are also appropriate for use in the development of child restraint systems. The new model joins other body shape models for adults and children available online at humanshape.org.
A broad collaboration at U-M several years in the making has produced an innovative tool for visual assessment of child body shape. The lead researchers on the effort are Dr. Julie Lumeng and Dr. Matt Reed. The first validation article on ShapeCoder appeared this month in Pediatric Obesity. The underlying body shape model was developed at UMTRI by Daniel Park and colleagues based on whole-body laser scan data from 147 children. The parametric model used in ShapeCoderallows a rater to select images on screen that best match the body shape of particular child. This tool has value for both estimating the body mass index of children for which measured stature and weight data are not available and for investigating misperception of child body size. An online version of the tool is available. We welcome further collaboration in this domain. Our growing library of body shape models can be used to accurately represent individuals from age 12 months through elderly adults.
I presented a method for quantifying vehicle occupant postures and body shapes at the SAE Congress 2017 . The methodology was demonstrated using children and a single Microsoft Kinect sensor. The challenge posed by the noisy and incomplete data was addressed by fitting the data using a statistical body shape model (SBSM). The SBSM used in this work was developed using laser scan data gathered from 147 children with stature ranging from 100 to 160 cm and BMI from 12 to 27 kg/m^2 in various sitting postures. A principal component (PC) analysis was conducted based on these scans along with the manually-measured body landmarks, and 100 PC scores were retained to account for 99% of variance in the body shape and sitting postures. A PC-based fast fitting method was applied to estimate the occupant characteristics by fitting the SBSM to an incomplete depth image of a subject. The results demonstrate that a fast, inexpensive system can be used to produce useful estimates of occupant characteristics that could be applied to improve personalization of component adjustments, restraint systems, and infotainment systems.
Update: Launched humanshape.org
I have launched HumanShape.org, a web portal for various online body shape models developed by UMTRI. Each model introduced in this site was developed based on hundreds of whole body laser scans, anatomical landmarks, and anthropometric data. As the previous online models, all models provide an intuitive way to generate a body shape, and the predicted 3D geometry and related data can be downloaded into a local repository.
More models will come. Stay tuned.
UPDATE: Launched Childshape.org
My colleague, Dr. Matt Reed and I published a public online body shape modeling site Childshape.org. This is the first online child body shape modeling tool. This site provides an intuitive interface to model a child body shape with a few parameters such as stature, body mass index, and sitting height to standing ratio. Users also can download the modeled body shape to a local repository as an STL file. Also, a set of estimated body landmarks including joint locations can be downloaded as well as the anthropometric dimensions of the selected body shape.
The model was developed using statistical methods based on laser scans of 137 children ages 3 to 11. More information about the methodology behind can be found in this paper "Parametric body shape model of standing children aged 3–11 years".