Outdoor Localization of 4-Wheels for Mobile Robot Using CNN with 3D Data

One of the possible problems for a mobile robot is the localization. This is due to GPS systems' difficulty in detecting the location of a moving robot and the effects of weathering on sensors, such as the light sensitivity of RGBs sensors. In addition, mapping techniques in severe environments requires time and effort. This research seeks to enhance the localization of mobile robots by merging 3D LiDAR data with RGB-D images and using deep learning techniques. The suggested method entails using a simulator to design a four-wheel mobile robot controlled by a LiDAR sensor and testing them in an outdoor environment. The proposed localization system works in three steps. The first step is the training step, in which the 3D point cloud LiDAR sensor scans the entire city and then uses the PCA method to compress the dimensions of the 3D LiDAR data to a 2.5D image. The testing data stage is the second step. First, the RGB and depth images have merged using the IHS technique to create a 2.5D fusion image. Next, Convolution Neural Networks are used to train and test these datasets to extract features from the images. Finally, the K-Nearest Neighbor method was used in the third step. The classification step allows high accuracy while also reducing training time. The experimental findings show that the suggested technique is better in yielding results up to an accuracy of 98.15 % and a Mean Square Error of 0.25, and the Mean Error Distance is 1.36 meters