<![CDATA[The rapid expansion of smart spaces has increased the need for robotic systems capable of interpreting visual cues, recognizing human behavior, and responding safely in real time. However, existing vision-based models often struggle with occlusion, lighting variation, latency constraints, and limited contextual understanding in dynamic human-centered environments. This study develops a hybrid vision-based pattern recognition framework that integrates Convolutional Neural Networks (CNNs), Transformer-based attention mechanisms, multi-scale feature fusion, supervised learning, and reinforcement learning. The model is trained and validated using publicly available human–robot interaction datasets and simulated smart space scenarios involving gesture recognition, object detection, activity recognition, and intention prediction. The objective is to enhance intelligent human–robot interaction by improving visual perception accuracy, contextual interpretation, adaptive decision-making, and real-time responsiveness in smart environments. The proposed framework achieves stronger performance than baseline CNN-only and Vision Transformer models, with improved accuracy in gesture recognition, object detection, activity recognition, and intention prediction while maintaining low-latency inference suitable for real-time robotic interaction. The model also demonstrates better adaptability under dynamic lighting, occlusion, and multi-person interaction scenarios. This study concludes that combining CNN-based local feature extraction, Transformer-based global attention, and reinforcement learning-based policy optimization provides a reliable, adaptive, and context-aware framework for intelligent robotic systems. The findings support safer and more efficient human–robot collaboration in healthcare, smart homes, collaborative workplaces, and smart city environments.]]>
Copyrights © 2026
