Indoor Surface Classification for Mobile Robots

In this project, we generated a dataset that contains three different types of indoor floor surfaces: carpet, tile and wood. Then, we used this dataset to train eight CNN-based models, including our proposed model, MobileNetV2-modified.

The dataset comprises a total of 2081 samples, consisting of images captured with cameras in various indoor environments and lighting conditions.

These images were taken from different angles in accordance with the overall dimensions of the indoor robots.

This dataset includes samples collected from more than 20 different indoor environments.

The dataset consists of 870 carpet samples, 638 tile samples and 573 wood surface samples.

本研究构建了一款涵盖三类不同室内地面表面的数据集：地毯（carpet）、瓷砖（tile）与木质地面（wood）。随后，我们使用该数据集对8种基于卷积神经网络（Convolutional Neural Network, CNN）的模型开展训练，其中包含我们提出的改进版MobileNetV2（MobileNetV2-modified）模型。

该数据集总计包含2081个样本，均为在多种室内环境与光照条件下通过相机采集得到的图像。

所有图像均适配室内机器人的整体作业尺寸，从不同角度进行拍摄。

该数据集的样本采集自20余个不同的室内环境。

该数据集具体包含870个地毯样本、638个瓷砖样本以及573个木质地面样本。