Using images to avoid collisions and bypass obstacles in indoor environments

Abstract

Convolutional Neural Network (CNN) has contributed a lot to the advancement of autonomous navigation techniques, and such systems can be adapted to facilitate the movement of robots and visually impaired people. This work presents an approach that uses images to avoid collisions and bypass obstacles in indoor environments. The constructed dataset uses information from forward and lateral speeds during walks to determine collisions and obstacle avoidance. VGG16, ResNet50, and Dronet architectures were used to evaluate the dataset. Finally, reflections on the dataset characteristics are added, and the CNNs performance is presented.