Computer Vision and Deep Learning for Human Motion Analysis

State-of-the-art technologies usually adopted to characterize human motion rely on wearable sensors, motion capture systems and markers. These system provide useful and accurate quantitative kinematic measures. Unfortunately, marker-based systems require expensive laboratory settings with several infrared cameras, limiting their use in uncontrolled environments. This could modify the naturalness of subjects movements and induce discomfort. Also, markers are intrusive and their number and location must be determined a priori. Recent advances on pose estimation and semantic features detectors based on computer vision and deep neural networks are opening the possibility of adopting efficient video-based methods for extracting movement information from RGB video data. In this contest, in the last few years, we introduced and tested the effectiveness of a video-based markerless pipeline for the quantitative analysis of human motion in the rehabilitation domain. In this paper, we summarize the implemented pipeline, we highlight possible application fields where its use can be beneficial also providing examples of applications.