STGCN-SHOA: A Spatial-Temporal Graph Convolutional Network for athletes’ gait analysis using Spotted Hyena Optimization Algorithm

Gait recognition from videos is an emerging research topic in computer vision-based applications. In sports sceneries, vision-based system investigates the different gait pattern that assists the athlete to improve the performance. The gait data acquired from videotapes have complexities namely, variation in illumination, recording conditions, angle variations and speed of stride. To tackle these issues, the proposed work constructs Spatial-Temporal traits based Graph Convolutional Network (STGCN) model for athletes’ gait analysis via multimodal fusion technique. The proposed STGCN framework employs a Spotted Hyena optimization Algorithm (SHOA) along with the Minimum Cross Entropy (MCE) to diminish the cross entropy loss for improving the effectiveness of the prediction. The hybrid STGCN- SHOA network model was assessed on three benchmark datasets namely, UCF 50 sports action dataset, Olympics sports dataset and Sports Videos in the Wild (SVW). The proposed method were compared to different algorithms, which show improved performance with 98.6%, 97.3%, and 98.9% recognition accuracy, respectively.