Accurate Hand Recognition with Neural Architecture Search Technology

Hand gesture recognition is a technology that enables computers to interpret and understand hand movements and gestures made by users. It has various applications across various domains, including human-computer interaction, gaming, virtual reality, sign language interpretation, and robotics. Hand recognition faces challenges such as lighting conditions, occlusions, and variations in hand shape and size. Creating reliable and precise recognition systems frequently necessitates tackling these issues. Neural Architecture Search (NAS) is a technique employed in deep learning and artificial intelligence to automate the creation and optimization of neural network topologies. The objective of NAS is to identify neural network designs that are optimally aligned with certain objectives, including image classification, natural language processing, or reinforcement learning while reducing the necessity for manual design and adjustment. YOLONAS model's integration of YOLO's speed and efficiency with NAS-driven optimization results in improved accuracy and performance in gesture recognition tasks, making it a compelling choice for real-time applications requiring accurate and efficient gesture analysis. In this research, we implement YOLO with NAS technology and training with the Oxford Hand Dataset. Performance metrics are employed for monitoring and quantifying important data, such as the number of Giga Floating Point Operations Per Second (GFLOPS), the mean average precision (mAP), and the time taken for detection. The results of our study indicate that the utilization of YOLONAS with a training time of 100 epochs produces a more reliable output when compared to other approaches.