DeepPose: A 2D Image Based Automated Framework for Human Pose Detection and a Trainer App Using Deep Learning

Tracking and estimating human postures through videos taken by various cameras has always been a very important and challenging task. Human posture estimate is not only a significant computer vision problem, but it also plays a crucial role in a variety of real-world applications such as Video Surveillance, Human–Computer Interaction (HCI), Medical Imaging, etc. The aim of the proposed work is to train people in different activity domains such as Sports, Gym, Exercise, Yoga, etc., in settings when physical interaction between instructor and trainee is not possible. Here, we detect human poses and check whether the pose made by the person is correct or not, based on correct input data points. A 2D stick figure model of the person has been used, which consists of joint points which will be displayed to the user along with the predicted accuracy of their posture. The performance of the proposed framework is compared with the existing models and gives remarkable results in terms of an accuracy i.e., 94%. The framework helps the person to improve their posture so that health risks are minimized. By adjusting one's posture and training style, theproposed framework can help to avoid these mishaps. It will also assist athletes in improving their techniques, avoiding injury, and increasing their endurance.