A Novel Sports Event Recognition using Pose Estimation and Multi-Fused Features

Several factors, including the dimension and locus of the limbs and head, and other pictures within an environment, were used to determine the management of event identification and recognition in a sequential figure. Body parts were estimated in an attempt to observe and monitor people as far as they were involved in intricate situations. Several feature descriptors were used on both the skeleton points and the silhouettes: MSER, SURF; and BRIEF; as well as HOG. With this combination it becomes easier to increase the discriminative power of the extracted features; hence making identification more robust and efficient. The extracted features are therefore passed through a Random Forest classifier with an accompaniment of Particle Swarm Optimization as a pre-classifier. Performance analysis proved the proposed approach achieves an event detection accuracy of 85.7% on the UCF-50 video dataset. The results show that our proposed method performs better than existing event recognition methods, which makes our research a significant contribution to advancing the state-of-the-art in event recognition.