ESTIMATED WEIGHT AND HEIGHT OF BODY HUMANS BASED ON FACE IMAGE

• Main Author: Setiawan Hamjaya, Harry
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/54150
• Institution: Institut Teknologi Bandung

Measurement of body weight and height is generally carried out conventionally, which is usually by direct measurement using either a scale or a stature meter. However, in a digital way, we offer a digital approach to measuring weight and height by means of an estimate based on facial images under normal conditions using machine learning techniques. In this study, an experiment was conducted to explore the best approaches and techniques needed in the process of estimating the value of body weight, height and Body Mass Index. The experiment was carried out with two kinds of approaches, namely Embeddings and Landmarks and four types of regression techniques, namely linear, ridge, lasso, and polynomial. Furthermore, the assessment of each technique and approach is carried out using a comparison of the error value between the ground truth and the estimated value. Based on the experimental results, a real-time application will be implemented to facilitate the process of estimating the value of weight, height and Body Mass Index. Based on the research conducted, it was found that a model with input in the form of the 5 most important features from landmark distances with polynomial regression techniques, produces the best performance with an average value of RMSE of around 0.41 and an average value of MAE of about 0.29 which is obtained around 960 experimental data for each value of weight, height and Body Mass Index in normal conditions. Based on the error evaluation value, the application built with the landmark model is successful in estimating the weight, height and BMI of humans in normal conditions close to the true value.