Head‐Mounted Display‐Based Virtual Reality Locomotion Recognition Using Dense Nested Attention Depthwise Separable Convolutional Neural Network Optimized With Fossa Optimization Algorithm

Head‐mounted display (HMD)‐based locomotion recognition is the detection and classification of user movements in virtual reality environments using HMD sensors to enhance immersion and interaction. To overcome these challenges, this work proposes an innovative HMD‐based locomotion recognition framework employing a dense nested attention depthwise separable convolutional neural network optimized with the fossa optimization algorithm (DNADSCNN–FOA) to achieve precise and efficient locomotion classification. Initially, the 3D Gait Analysis Dataset is pre‐processed using the pretreating and normalizing (PN) process to clean, normalize, and segment gait signals while preserving essential motion information. Key features are then classified by the DNADSCNN to accurately identify locomotion states—static, dynamic, and stand—while the FOA fine‐tunes network parameters to enhance convergence, stability, and classification precision. The experimental results on the 3D Gait Analysis Dataset have shown very high accuracy with the accuracy of 99.6%, recall of 99.5%, precision of 99.3%, and F1 score of 99.7%, thereby proving that DNADSCNN‐FOA approach has high robustness, adaptability, and reliability. Conclusively, the suggested framework has presented a viable, efficient and highly accurate model for recognition of locomotion with HMDs with consistency in performance irrespective of the diversity in locomotion and activity states within immersive VR environments.