A Fast and Robust Pipeline for Generating 3D Human Models Based on Body Measurements Extraction
摘要
The Computer Vision task of extracting body measurements from images has many applications in online shopping, medical, sport & fitness and other fields which require knowing or monitoring body measurements. Ideally, this should be achieved without complex hardware, fast and with high availability. In this chapter, we describe a solution which extracts body measurements using any smartphone with a camera and internet access. Using as input two photos—one frontal and one lateral—and the user’s height, weight, age and gender, our solution can extract more than 100 body measurements with a measurement error less than 5 mm, and an inference time of around 5 s. The solution has two main components, both of which use various Artificial Intelligence techniques. The first component consists of a web widget which gathers the required data from the user and features a virtual assistant which guides the user in the data acquisition process. The second component consists of a measurement engine, which processes the data and returns the body measurements.