Corridors Classification Using Convolutional Neural Network on Mobile Robot

Corridor detection is a critical challenge in mobile robotics, especially for navigation in indoor environments such as office buildings or hospitals. Despite the continuous advancement of robotics, accurate corridor detection and navigation remain complex due to variations in lighting conditions, differences in architectural designs, and other constraints. This study has developed a wheeled robot with Convolutional Neural Network (CNN) to enhance the corridor detection capability. The system is proposed to generate a campus building corridor dataset and evaluate six CNN-based classification methods, DenseNet121, EfficientNetV2B0, InceptionResNetV2, MobileNetV2, ResNet50V2, and VGG16. Each one is known for its robustness in image recognition tasks. The main contribution of this study is to develop an effective and efficient corridor detection system using CNN, along with a comprehensive analysis of six different CNN architectures for application in mobile robots. The result of the study shows that EfficientNetV2B0 achieves the highest accuracy of about 0.97. This study is integrated into a mobile robot tasked with detecting and classifying corridor types.