Enhancing Athletic Performance:2D Human Pose Estimation Using Deep Neural Networks for Movement Analysis

Computer vision is a field that is used to build many neural network-based prototypes efficiently. It analyses images and videos effectively and processes them digitally. Human Pose estimation (HPE) is one of the computer vision applications which additionally uses neural networks for prediction. Human pose estimation is used widely in many fields which includes action recognition, fitness, gaming, and motion capture. HPE identifies and locates the key points on the body joints of a person’s body. These body points are connected to form a Skeletal Structure. This skeletal representation helps to easily visualize the image in 2-D. Focusing on Sports domain, HPE plays major role in estimating athlete movements. It helps in analyzing the feedback on the posture which allows to make adjustments during training. It also helps viewers to analyse the insights of athlete performance. To achieve the model that contains all these features, requires involvement of few frameworks. OpenPose is utilised for fast processing of videos and can detect the multiple people in one frame. OpenPose has the capability of handling the occulsion. On the Other hand, the model also import the capabilities of HRNet, which is used for analyzing athlete movements and tracks the gestures. After detecting the keypoints on the body joints, these points must be integrated to form 2-D images. To achieve the spatial relationship between the bodypoints Joint Graph Convolutional Neural Networks (JGCN) framework is used. The proposed model is estimated to generate approximately (4- 5)percent more efficiency and accuracy in its output 2-D image compared with the previous models.