Custom Deep CNN-Based Staircase Classification for Smart Environments

Effective indoor navigation is pivotal for robots operating in smart homes and commercial settings. This paper delves into the intricate challenge of staircase classification and segmentation, introducing a customized deep convolutional neural network (CNN) architecture tailored for robotic applications, particularly relevant to tasks like staircase cleaning. Based on a meticulously curated dataset comprising 9000 to 9500 images with two classes (stairs and non-stairs), the proposed deep CNN model demonstrates exceptional accuracy in classifying images and excels in staircase segmentation. Benchmarking against established algorithms, including the 12-layer CNN, AlexNet, ResNet, Inception (GoogLeNet), and MobileNetV2, underscores the superior performance of our approach. These results highlight the efficacy of the proposed deep CNN in providing accurate spatial information for robots navigating in complex indoor spaces. The paper contributes to the advancement of computer vision in robotic applications, demonstrating a practical solution for enhanced navigational capabilities in real-world scenarios. This system provides crucial spatial information necessary for robots navigating and executing tasks in complex indoor environments.