Outdoor Localization for a Mobile Robot under Different Weather Conditions Using a Deep Learning Algorithm

Abstract

A fundamental issue in robotics is the precise localization of mobile robots in uncertain environments. Due to changing environmental patterns and lighting, localization under difficult perceptual conditions remains problematic. This paper presents an approach for locating an outdoor mobile robot using deep learning algorithms merge with 3D Light Detection and Ranging LiDAR data and RGB-D image. This approach is divided into three levels. The first is the training level, which involves scanning the localization area with a 3D LiDAR sensor and then converting the data into a 2.5D image based on the Principal Component Analysis. The testing is the second level in the Intensity Hue Saturation process. Then, the RGB and Depth images are combined to create a 2.5D fusion image. These datasets are trained and tested using Convolution Neural Networks. The K-Nearest Neighbor algorithm is used in the third level is the classification. The results show that the proposed approach is better in terms of accuracy of 97.46% and the Mean error distance is 0.6 meters.