Landmarks recognition and body reconstruction through an OpenCV AI Kit device

The process of data acquisition is the basis of biomedicine. It is often timedemanding and requires the subject to perform tasks in a lab and to wear sensors. Applying these sensors to the subject requires time, trained personnel, and a laboratory or specialised location. One example is human movement’s reconstruction through stereo-photogrammetry. This method is considered the gold standard of movement analysis but requires more than one specialised camera and some time to prepare the subject for the acquisition (markers placement). Although it is an exceptionally reliable method, it presents some limitations: it is time-consuming, it is not available to everybody, it requires a lab, and it must be performed in the presence of qualified personnel. To improve data collection, we must look for reliable instrumentation that can be placed everywhere, does not require much time to set up, and can be quickly started by the user himself. Given the spread of device cameras capable of registering 3D information, the asserting of the AI, and the general desire for real-time stream of information processing, oak-d devices seem to answer all this demand. We are talking about a low price, compact camera, easy to program and high-performance devices that are ideal for carrying around and experimenting with. This project aims to prove that a marker-less 3D landmark acquisition through the exploitation of an end-to-end pipeline, the improvement of already-coded programs from GitHub, and the development of a calibration protocol is not only possible but also valuable.