Deep Learning-based Human Height Estimation from a Stereo Vision System

This paper presents a deep learning-based human height estimation approach using a stereo vision system, which is part of a smart receptionist framework. The proposal consists of a smart screen with an integrated webcam and an additional webcam. The workflow of this approach initially acquires and processes the images in real-time, using deep neural networks detects a human in the images, aligns the images from the two cameras, then obtains the depth of the camera, and finally computes the height of the person under study. The proposed solution is evaluated by different people to check the effectiveness of the system. Obtained results show MAE below 1.0 cm in the computed heights, it is also compared with other techniques of the state-of-the-art.