SI: Human Pose Estimation and Its Applications

Human pose estimation is an important task in computer vision because it not only benefits other vision tasks like action recognition, person re-identification and virtual try-on but also facilitates applications in real-world domains such as robotics, healthcare, sports, and retail. An effective and efficient human pose estimation system can help robots learn skills from demonstrations, help physical therapists diagnose and rehabilitate patients, help sports analysts or coaches track and train athletes, and help retailers build employee-free stores. Thanks to the development of deep learning and large-scale datasets, the performance of state-of-the-art human pose estimation approaches has drastically improved in recent years, and they can estimate postures in daily activities and some sports accurately. However, several challenges still exist. For example, (1) it is challenging to estimate postures which rarely or never occur in the training data; (2) it is difficult to handle complex scenarios such as crowded people, motion blur, low-light conditions, and occlusions; (3) it is desirable to establish efficient models which can estimate human poses in real time or on low-power devices; (4) it is exciting to create new applications of human pose estimation that can benefit society or transform industry.