| Abstrakt: |
ArUco markers have gained significant attention for their utility in indoor navigation assistance, enhancing the accuracy and efficiency of navigation systems. However, traditional detection methods often need help with computational intensity and effectiveness under varying lighting conditions, necessitating the development of advanced detection techniques. This study investigates deep learning-powered ArUco marker detection using the MobileViTv2 model. MobileViTv2, a lightweight hybrid network combining Convolutional Neural Networks (CNN) and Vision Transformer (ViT) elements, is designed to balance accuracy and computational efficiency. The model was trained and fine-tuned using a transfer learning approach, with comparisons to prominent CNN architectures such as ResNet50, MobileNetV2, DenseNet121, VGG16, InceptionV3, and NASNetMobile. The MobileViTv2 model demonstrated superior performance in detecting and classifying ArUco markers for indoor navigation. The model attained a peak accuracy of 100% using the optimizer developed by Adam with a learning rate of 0.001, surpassing the performance of all models examined in the study. The results showed excellent precision, recall, and F1 scores, confirming the model's robustness and reliability. The total number of trainable parameters was slim, making the model more efficient and faster in processing, which is ideal for real-time applications. The modified MobileViTv2 model significantly enhances the accuracy and efficiency of ArUco marker detection in indoor navigation systems. [ABSTRACT FROM AUTHOR] |
