We aim to develop a smart phone application which can estimate the height and weight of children based on pictures rapidly, easily and precisely. The goal of this application is to improve data collection in developing countries and to facilitate the identification of malnourished children.
The MeasureMe application will allow the precise and reproducible estimation of height and weight of a child without any bulky or faulty material. Moreover, it will allow the automatic diagnosis of the state of malnutrition of a child, removing calculation and variable bias. This will provide an important help to the personnel working in the field of health in regions where the risk of malnutrition is high and where resources are limited.
In addition, this application will allow the automated creation and update of health databases of multiple international health organizations without requiring manual data entry to improve the follow-up of malnourished children and the early detection of food crises.
In order to be able to estimate the height and weight of a person from a photo, the MeasureMe application needs to perform three key steps:
Step 1: Identification of the person’s body and body parts in the image
This step can be done using image processing and machine learning. This step presents many challenges, such as how to separate the person from the background, how to identify body parts of a person with clothes, etc.
Step 2: Estimation of the person’s height and weight
The height is relatively easy to obtain by measuring the distance between the sole of the feet and the tip of the head. The estimation of the weight, on the other hand, is more challenging, as it cannot be directly measured on the image. In order to do this, we use body measures that are related to the weight of a person and with these measures extrapolate the weight. Currently, we are using an equation that uses seven key body measures to estimate the weight of a person (Velardo and Dugelay 2006). With this equation, we manage to estimate the weight of persons quite precisely (mean error of 4%). We are currently working on improving this equation and adapting it for children.
Step 3: Transformation of units
In order to bring back the measures on the image to real units, we currently use a reference object (a small red ball) with a known height. However, in the future, we would like to use the depth captors available in new smart phones and get rid of the reference object.